Comment out erroniously removed del statement, comment out opt tests

## What type of PR is this? (check all applicable)

- [ ] Refactor
- [ ] Feature
- [ ] Bug Fix
- [ ] Optimization
- [X ] Documentation Update


## Have you discussed this change with the InvokeAI team?
- [ X] Yes
- [ ] No, because:

      
## Description
This PR points mkdocs to the `main` branch again, so that the 3.0.0
documentation appears in gh-pages.

It also makes a minor tweak to the tooltip for model imports, so that
users know that URLs are accepted.

Also rebuilds frontend for use in beta testing.

.dockerignore

@@ -1,25 +1,9 @@
-# use this file as a whitelist
 *
 !invokeai
-!ldm
 !pyproject.toml
+!docker/docker-entrypoint.sh
+!LICENSE
 
-# ignore frontend/web but whitelist dist
-invokeai/frontend/web/
-!invokeai/frontend/web/dist/
-
-# ignore invokeai/assets but whitelist invokeai/assets/web
-invokeai/assets/
-!invokeai/assets/web/
-
-# Guard against pulling in any models that might exist in the directory tree
-**/*.pt*
-**/*.ckpt
-
-# Byte-compiled / optimized / DLL files
-**/__pycache__/
-**/*.py[cod]
-
-# Distribution / packaging
-**/*.egg-info/
-**/*.egg
+**/node_modules
+**/__pycache__
+**/*.egg-info

.github/CODEOWNERS

@@ -6,7 +6,7 @@
 /mkdocs.yml @lstein  @blessedcoolant
 
 # nodes
-/invokeai/app/ @Kyle0654 @blessedcoolant
+/invokeai/app/ @Kyle0654 @blessedcoolant @psychedelicious @brandonrising 
 
 # installation and configuration
 /pyproject.toml  @lstein @blessedcoolant
@@ -22,7 +22,7 @@
 /invokeai/backend @blessedcoolant @psychedelicious @lstein @maryhipp
 
 # generation, model management, postprocessing
-/invokeai/backend  @damian0815 @lstein @blessedcoolant @jpphoto @gregghelt2 @StAlKeR7779
+/invokeai/backend  @damian0815 @lstein @blessedcoolant @gregghelt2 @StAlKeR7779 @brandonrising
 
 # front ends
 /invokeai/frontend/CLI @lstein

.github/workflows/build-container.yml

@@ -3,17 +3,15 @@ on:
   push:
     branches:
       - 'main'
-      - 'update/ci/docker/*'
-      - 'update/docker/*'
-      - 'dev/ci/docker/*'
-      - 'dev/docker/*'
     paths:
       - 'pyproject.toml'
       - '.dockerignore'
       - 'invokeai/**'
       - 'docker/Dockerfile'
+      - 'docker/docker-entrypoint.sh'
+      - 'workflows/build-container.yml'
     tags:
-      - 'v*.*.*'
+      - 'v*'
   workflow_dispatch:
 
 permissions:
@@ -26,23 +24,27 @@ jobs:
     strategy:
       fail-fast: false
       matrix:
-        flavor:
-          - rocm
-          - cuda
-          - cpu
-        include:
-          - flavor: rocm
-            pip-extra-index-url: 'https://download.pytorch.org/whl/rocm5.2'
-          - flavor: cuda
-            pip-extra-index-url: ''
-          - flavor: cpu
-            pip-extra-index-url: 'https://download.pytorch.org/whl/cpu'
+        gpu-driver:
+        - cuda
+        - cpu
+        - rocm
     runs-on: ubuntu-latest
-    name: ${{ matrix.flavor }}
+    name: ${{ matrix.gpu-driver }}
     env:
-      PLATFORMS: 'linux/amd64,linux/arm64'
-      DOCKERFILE: 'docker/Dockerfile'
+      # torch/arm64 does not support GPU currently, so arm64 builds
+      # would not be GPU-accelerated.
+      # re-enable arm64 if there is sufficient demand.
+      # PLATFORMS: 'linux/amd64,linux/arm64'
+      PLATFORMS: 'linux/amd64'
     steps:
+      - name: Free up more disk space on the runner
+        # https://github.com/actions/runner-images/issues/2840#issuecomment-1284059930
+        run: |
+          sudo rm -rf /usr/share/dotnet
+          sudo rm -rf "$AGENT_TOOLSDIRECTORY"
+          sudo swapoff /mnt/swapfile
+          sudo rm -rf /mnt/swapfile
+
       - name: Checkout
         uses: actions/checkout@v3
 
@@ -53,7 +55,7 @@ jobs:
           github-token: ${{ secrets.GITHUB_TOKEN }}
           images: |
             ghcr.io/${{ github.repository }}
-            ${{ vars.DOCKERHUB_REPOSITORY }}
+            ${{ env.DOCKERHUB_REPOSITORY }}
           tags: |
             type=ref,event=branch
             type=ref,event=tag
@@ -62,8 +64,8 @@ jobs:
             type=pep440,pattern={{major}}
             type=sha,enable=true,prefix=sha-,format=short
           flavor: |
-            latest=${{ matrix.flavor == 'cuda' && github.ref == 'refs/heads/main' }}
-            suffix=-${{ matrix.flavor }},onlatest=false
+            latest=${{ matrix.gpu-driver == 'cuda' && github.ref == 'refs/heads/main' }}
+            suffix=-${{ matrix.gpu-driver }},onlatest=false
 
       - name: Set up QEMU
         uses: docker/setup-qemu-action@v2
@@ -81,34 +83,33 @@ jobs:
           username: ${{ github.repository_owner }}
           password: ${{ secrets.GITHUB_TOKEN }}
 
-      - name: Login to Docker Hub
-        if: github.event_name != 'pull_request' && vars.DOCKERHUB_REPOSITORY != ''
-        uses: docker/login-action@v2
-        with:
-          username: ${{ secrets.DOCKERHUB_USERNAME }}
-          password: ${{ secrets.DOCKERHUB_TOKEN }}
+      # - name: Login to Docker Hub
+      #   if: github.event_name != 'pull_request' && vars.DOCKERHUB_REPOSITORY != ''
+      #   uses: docker/login-action@v2
+      #   with:
+      #     username: ${{ secrets.DOCKERHUB_USERNAME }}
+      #     password: ${{ secrets.DOCKERHUB_TOKEN }}
 
       - name: Build container
         id: docker_build
         uses: docker/build-push-action@v4
         with:
           context: .
-          file: ${{ env.DOCKERFILE }}
+          file: docker/Dockerfile
           platforms: ${{ env.PLATFORMS }}
           push: ${{ github.ref == 'refs/heads/main' || github.ref_type == 'tag' }}
           tags: ${{ steps.meta.outputs.tags }}
           labels: ${{ steps.meta.outputs.labels }}
-          build-args: PIP_EXTRA_INDEX_URL=${{ matrix.pip-extra-index-url }}
           cache-from: |
-            type=gha,scope=${{ github.ref_name }}-${{ matrix.flavor }}
-            type=gha,scope=main-${{ matrix.flavor }}
-          cache-to: type=gha,mode=max,scope=${{ github.ref_name }}-${{ matrix.flavor }}
+            type=gha,scope=${{ github.ref_name }}-${{ matrix.gpu-driver }}
+            type=gha,scope=main-${{ matrix.gpu-driver }}
+          cache-to: type=gha,mode=max,scope=${{ github.ref_name }}-${{ matrix.gpu-driver }}
 
-      - name: Docker Hub Description
-        if: github.ref == 'refs/heads/main' || github.ref == 'refs/tags/*' && vars.DOCKERHUB_REPOSITORY != ''
-        uses: peter-evans/dockerhub-description@v3
-        with:
-          username: ${{ secrets.DOCKERHUB_USERNAME }}
-          password: ${{ secrets.DOCKERHUB_TOKEN }}
-          repository: ${{ vars.DOCKERHUB_REPOSITORY }}
-          short-description: ${{ github.event.repository.description }}
+      # - name: Docker Hub Description
+      #   if: github.ref == 'refs/heads/main' || github.ref == 'refs/tags/*' && vars.DOCKERHUB_REPOSITORY != ''
+      #   uses: peter-evans/dockerhub-description@v3
+      #   with:
+      #     username: ${{ secrets.DOCKERHUB_USERNAME }}
+      #     password: ${{ secrets.DOCKERHUB_TOKEN }}
+      #     repository: ${{ vars.DOCKERHUB_REPOSITORY }}
+      #     short-description: ${{ github.event.repository.description }}

.github/workflows/mkdocs-material.yml

@@ -43,7 +43,7 @@ jobs:
             --verbose
 
       - name: deploy to gh-pages
-        if: ${{ github.ref == 'refs/heads/v2.3' }}
+        if: ${{ github.ref == 'refs/heads/main' }}
         run: |
           python -m \
             mkdocs gh-deploy \

LICENSE

@@ -1,21 +1,176 @@
-MIT License
+                                 Apache License
+                           Version 2.0, January 2004
+                        http://www.apache.org/licenses/
 
-Copyright (c) 2022 InvokeAI Team
+   TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
 
-Permission is hereby granted, free of charge, to any person obtaining a copy
-of this software and associated documentation files (the "Software"), to deal
-in the Software without restriction, including without limitation the rights
-to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
-copies of the Software, and to permit persons to whom the Software is
-furnished to do so, subject to the following conditions:
+   1. Definitions.
 
-The above copyright notice and this permission notice shall be included in all
-copies or substantial portions of the Software.
+      "License" shall mean the terms and conditions for use, reproduction,
+      and distribution as defined by Sections 1 through 9 of this document.
 
-THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
-IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
-FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
-AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
-LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
-OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
-SOFTWARE.
+      "Licensor" shall mean the copyright owner or entity authorized by
+      the copyright owner that is granting the License.
+
+      "Legal Entity" shall mean the union of the acting entity and all
+      other entities that control, are controlled by, or are under common
+      control with that entity. For the purposes of this definition,
+      "control" means (i) the power, direct or indirect, to cause the
+      direction or management of such entity, whether by contract or
+      otherwise, or (ii) ownership of fifty percent (50%) or more of the
+      outstanding shares, or (iii) beneficial ownership of such entity.
+
+      "You" (or "Your") shall mean an individual or Legal Entity
+      exercising permissions granted by this License.
+
+      "Source" form shall mean the preferred form for making modifications,
+      including but not limited to software source code, documentation
+      source, and configuration files.
+
+      "Object" form shall mean any form resulting from mechanical
+      transformation or translation of a Source form, including but
+      not limited to compiled object code, generated documentation,
+      and conversions to other media types.
+
+      "Work" shall mean the work of authorship, whether in Source or
+      Object form, made available under the License, as indicated by a
+      copyright notice that is included in or attached to the work
+      (an example is provided in the Appendix below).
+
+      "Derivative Works" shall mean any work, whether in Source or Object
+      form, that is based on (or derived from) the Work and for which the
+      editorial revisions, annotations, elaborations, or other modifications
+      represent, as a whole, an original work of authorship. For the purposes
+      of this License, Derivative Works shall not include works that remain
+      separable from, or merely link (or bind by name) to the interfaces of,
+      the Work and Derivative Works thereof.
+
+      "Contribution" shall mean any work of authorship, including
+      the original version of the Work and any modifications or additions
+      to that Work or Derivative Works thereof, that is intentionally
+      submitted to Licensor for inclusion in the Work by the copyright owner
+      or by an individual or Legal Entity authorized to submit on behalf of
+      the copyright owner. For the purposes of this definition, "submitted"
+      means any form of electronic, verbal, or written communication sent
+      to the Licensor or its representatives, including but not limited to
+      communication on electronic mailing lists, source code control systems,
+      and issue tracking systems that are managed by, or on behalf of, the
+      Licensor for the purpose of discussing and improving the Work, but
+      excluding communication that is conspicuously marked or otherwise
+      designated in writing by the copyright owner as "Not a Contribution."
+
+      "Contributor" shall mean Licensor and any individual or Legal Entity
+      on behalf of whom a Contribution has been received by Licensor and
+      subsequently incorporated within the Work.
+
+   2. Grant of Copyright License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      copyright license to reproduce, prepare Derivative Works of,
+      publicly display, publicly perform, sublicense, and distribute the
+      Work and such Derivative Works in Source or Object form.
+
+   3. Grant of Patent License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      (except as stated in this section) patent license to make, have made,
+      use, offer to sell, sell, import, and otherwise transfer the Work,
+      where such license applies only to those patent claims licensable
+      by such Contributor that are necessarily infringed by their
+      Contribution(s) alone or by combination of their Contribution(s)
+      with the Work to which such Contribution(s) was submitted. If You
+      institute patent litigation against any entity (including a
+      cross-claim or counterclaim in a lawsuit) alleging that the Work
+      or a Contribution incorporated within the Work constitutes direct
+      or contributory patent infringement, then any patent licenses
+      granted to You under this License for that Work shall terminate
+      as of the date such litigation is filed.
+
+   4. Redistribution. You may reproduce and distribute copies of the
+      Work or Derivative Works thereof in any medium, with or without
+      modifications, and in Source or Object form, provided that You
+      meet the following conditions:
+
+      (a) You must give any other recipients of the Work or
+          Derivative Works a copy of this License; and
+
+      (b) You must cause any modified files to carry prominent notices
+          stating that You changed the files; and
+
+      (c) You must retain, in the Source form of any Derivative Works
+          that You distribute, all copyright, patent, trademark, and
+          attribution notices from the Source form of the Work,
+          excluding those notices that do not pertain to any part of
+          the Derivative Works; and
+
+      (d) If the Work includes a "NOTICE" text file as part of its
+          distribution, then any Derivative Works that You distribute must
+          include a readable copy of the attribution notices contained
+          within such NOTICE file, excluding those notices that do not
+          pertain to any part of the Derivative Works, in at least one
+          of the following places: within a NOTICE text file distributed
+          as part of the Derivative Works; within the Source form or
+          documentation, if provided along with the Derivative Works; or,
+          within a display generated by the Derivative Works, if and
+          wherever such third-party notices normally appear. The contents
+          of the NOTICE file are for informational purposes only and
+          do not modify the License. You may add Your own attribution
+          notices within Derivative Works that You distribute, alongside
+          or as an addendum to the NOTICE text from the Work, provided
+          that such additional attribution notices cannot be construed
+          as modifying the License.
+
+      You may add Your own copyright statement to Your modifications and
+      may provide additional or different license terms and conditions
+      for use, reproduction, or distribution of Your modifications, or
+      for any such Derivative Works as a whole, provided Your use,
+      reproduction, and distribution of the Work otherwise complies with
+      the conditions stated in this License.
+
+   5. Submission of Contributions. Unless You explicitly state otherwise,
+      any Contribution intentionally submitted for inclusion in the Work
+      by You to the Licensor shall be under the terms and conditions of
+      this License, without any additional terms or conditions.
+      Notwithstanding the above, nothing herein shall supersede or modify
+      the terms of any separate license agreement you may have executed
+      with Licensor regarding such Contributions.
+
+   6. Trademarks. This License does not grant permission to use the trade
+      names, trademarks, service marks, or product names of the Licensor,
+      except as required for reasonable and customary use in describing the
+      origin of the Work and reproducing the content of the NOTICE file.
+
+   7. Disclaimer of Warranty. Unless required by applicable law or
+      agreed to in writing, Licensor provides the Work (and each
+      Contributor provides its Contributions) on an "AS IS" BASIS,
+      WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
+      implied, including, without limitation, any warranties or conditions
+      of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
+      PARTICULAR PURPOSE. You are solely responsible for determining the
+      appropriateness of using or redistributing the Work and assume any
+      risks associated with Your exercise of permissions under this License.
+
+   8. Limitation of Liability. In no event and under no legal theory,
+      whether in tort (including negligence), contract, or otherwise,
+      unless required by applicable law (such as deliberate and grossly
+      negligent acts) or agreed to in writing, shall any Contributor be
+      liable to You for damages, including any direct, indirect, special,
+      incidental, or consequential damages of any character arising as a
+      result of this License or out of the use or inability to use the
+      Work (including but not limited to damages for loss of goodwill,
+      work stoppage, computer failure or malfunction, or any and all
+      other commercial damages or losses), even if such Contributor
+      has been advised of the possibility of such damages.
+
+   9. Accepting Warranty or Additional Liability. While redistributing
+      the Work or Derivative Works thereof, You may choose to offer,
+      and charge a fee for, acceptance of support, warranty, indemnity,
+      or other liability obligations and/or rights consistent with this
+      License. However, in accepting such obligations, You may act only
+      on Your own behalf and on Your sole responsibility, not on behalf
+      of any other Contributor, and only if You agree to indemnify,
+      defend, and hold each Contributor harmless for any liability
+      incurred by, or claims asserted against, such Contributor by reason
+      of your accepting any such warranty or additional liability.
+
+   

README.md

@@ -3,8 +3,8 @@
 ![project hero](https://github.com/invoke-ai/InvokeAI/assets/31807370/1a917d94-e099-4fa1-a70f-7dd8d0691018)
 
 # Invoke AI - Generative AI for Professional Creatives
-## Image Generation for Stable Diffusion, Custom-Trained Models, and more. 
-  Learn more about us and get started instantly at [invoke.ai](https://invoke.ai)
+## Professional Creative Tools for Stable Diffusion, Custom-Trained Models, and more. 
+  To learn more about Invoke AI, get started instantly, or implement our Business solutions, visit [invoke.ai](https://invoke.ai)
 
 
 [![discord badge]][discord link]
@@ -132,8 +132,10 @@ and go to http://localhost:9090.
 
 ### Command-Line Installation (for developers and users familiar with Terminals)
 
-You must have Python 3.9 or 3.10 installed on your machine. Earlier or later versions are
-not supported.
+You must have Python 3.9 or 3.10 installed on your machine. Earlier or
+later versions are not supported.
+Node.js also needs to be installed along with yarn (can be installed with
+the command `npm install -g yarn` if needed)
 
 1. Open a command-line window on your machine. The PowerShell is recommended for Windows.
 2. Create a directory to install InvokeAI into. You'll need at least 15 GB of free space:
@@ -197,11 +199,18 @@ not supported.
 7. Launch the web server (do it every time you run InvokeAI):
 
     ```terminal
-    invokeai --web
+    invokeai-web
     ```
 
-8. Point your browser to http://localhost:9090 to bring up the web interface.
-9. Type `banana sushi` in the box on the top left and click `Invoke`.
+8. Build Node.js assets
+
+  ```terminal
+  cd invokeai/frontend/web/
+  yarn vite build
+  ```
+
+9. Point your browser to http://localhost:9090 to bring up the web interface.
+10. Type `banana sushi` in the box on the top left and click `Invoke`.
 
 Be sure to activate the virtual environment each time before re-launching InvokeAI,
 using `source .venv/bin/activate` or `.venv\Scripts\activate`.
@@ -329,24 +338,24 @@ InvokeAI offers a locally hosted Web Server & React Frontend, with an industry l
 
 The Unified Canvas is a fully integrated canvas implementation with support for all core generation capabilities, in/outpainting, brush tools, and more. This creative tool unlocks the capability for artists to create with AI as a creative collaborator, and can be used to augment AI-generated imagery, sketches, photography, renders, and more.
 
-### *Advanced Prompt Syntax*
+### *Node Architecture & Editor (Beta)*
 
-Invoke AI's advanced prompt syntax allows for token weighting, cross-attention control, and prompt blending, allowing for fine-tuned tweaking of your invocations and exploration of the latent space. 
+Invoke AI's backend is built on a graph-based execution architecture. This allows for customizable generation pipelines to be developed by professional users looking to create specific workflows to support their production use-cases, and will be extended in the future with additional capabilities.
 
-### *Command Line Interface*
+### *Board & Gallery Management*
 
-For users utilizing a terminal-based environment, or who want to take advantage of CLI features, InvokeAI offers an extensive and actively supported command-line interface that provides the full suite of generation functionality available in the tool.
+Invoke AI provides an organized gallery system for easily storing, accessing, and remixing your content in the Invoke workspace. Images can be dragged/dropped onto any Image-base UI element in the application, and rich metadata within the Image allows for easy recall of key prompts or settings used in your workflow. 
 
 ### Other features
 
 - *Support for both ckpt and diffusers models*
 - *SD 2.0, 2.1 support*
-- *Upscaling & Face Restoration Tools*
+- *Upscaling Tools*
 - *Embedding Manager & Support*
 - *Model Manager & Support*
 - *Node-Based Architecture*
 - *Node-Based Plug-&-Play UI (Beta)*
-- *Boards & Gallery Management
+- *SDXL Support* (Coming soon)
 
 ### Latest Changes
 
@@ -359,7 +368,7 @@ Notes](https://github.com/invoke-ai/InvokeAI/releases) and the
 Please check out our **[Q&A](https://invoke-ai.github.io/InvokeAI/help/TROUBLESHOOT/#faq)** to get solutions for common installation
 problems and other issues.
 
-## 🤝 Contributing
+## Contributing
 
 Anyone who wishes to contribute to this project, whether documentation, features, bug fixes, code
 cleanup, testing, or code reviews, is very much encouraged to do so.
@@ -378,7 +387,7 @@ to become part of our community.
 
 Welcome to InvokeAI!
 
-### 👥 Contributors
+### Contributors
 
 This fork is a combined effort of various people from across the world.
 [Check out the list of all these amazing people](https://invoke-ai.github.io/InvokeAI/other/CONTRIBUTORS/). We thank them for

docker/.env.sample

@@ -0,0 +1,13 @@
+## Make a copy of this file named `.env` and fill in the values below.
+## Any environment variables supported by InvokeAI can be specified here.
+
+# INVOKEAI_ROOT is the path to a path on the local filesystem where InvokeAI will store data.
+# Outputs will also be stored here by default.
+# This **must** be an absolute path.
+INVOKEAI_ROOT=
+
+HUGGINGFACE_TOKEN=
+
+## optional variables specific to the docker setup
+# GPU_DRIVER=cuda
+# CONTAINER_UID=1000

docker/Dockerfile

@@ -1,107 +1,129 @@
-# syntax=docker/dockerfile:1
+# syntax=docker/dockerfile:1.4
 
-ARG PYTHON_VERSION=3.9
-##################
-##  base image  ##
-##################
-FROM --platform=${TARGETPLATFORM} python:${PYTHON_VERSION}-slim AS python-base
+## Builder stage
 
-LABEL org.opencontainers.image.authors="mauwii@outlook.de"
+FROM library/ubuntu:22.04 AS builder
 
-# Prepare apt for buildkit cache
-RUN rm -f /etc/apt/apt.conf.d/docker-clean \
-  && echo 'Binary::apt::APT::Keep-Downloaded-Packages "true";' >/etc/apt/apt.conf.d/keep-cache
+ARG DEBIAN_FRONTEND=noninteractive
+RUN rm -f /etc/apt/apt.conf.d/docker-clean; echo 'Binary::apt::APT::Keep-Downloaded-Packages "true";' > /etc/apt/apt.conf.d/keep-cache
+RUN --mount=type=cache,target=/var/cache/apt,sharing=locked \
+    --mount=type=cache,target=/var/lib/apt,sharing=locked \
+    apt update && apt-get install -y \
+        git \
+        python3.10-venv \
+        python3-pip \
+        build-essential
 
-# Install dependencies
-RUN \
-  --mount=type=cache,target=/var/cache/apt,sharing=locked \
-  --mount=type=cache,target=/var/lib/apt,sharing=locked \
-  apt-get update \
-  && apt-get install -y \
-    --no-install-recommends \
-    libgl1-mesa-glx=20.3.* \
-    libglib2.0-0=2.66.* \
-    libopencv-dev=4.5.*
+ENV INVOKEAI_SRC=/opt/invokeai
+ENV VIRTUAL_ENV=/opt/venv/invokeai
 
-# Set working directory and env
-ARG APPDIR=/usr/src
-ARG APPNAME=InvokeAI
-WORKDIR ${APPDIR}
-ENV PATH ${APPDIR}/${APPNAME}/bin:$PATH
-# Keeps Python from generating .pyc files in the container
-ENV PYTHONDONTWRITEBYTECODE 1
-# Turns off buffering for easier container logging
-ENV PYTHONUNBUFFERED 1
-# Don't fall back to legacy build system
-ENV PIP_USE_PEP517=1
+ENV PATH="$VIRTUAL_ENV/bin:$PATH"
+ARG TORCH_VERSION=2.0.1
+ARG TORCHVISION_VERSION=0.15.2
+ARG GPU_DRIVER=cuda
+ARG TARGETPLATFORM="linux/amd64"
+# unused but available
+ARG BUILDPLATFORM
 
-#######################
-##  build pyproject  ##
-#######################
-FROM python-base AS pyproject-builder
+WORKDIR ${INVOKEAI_SRC}
 
-# Install build dependencies
-RUN \
-  --mount=type=cache,target=/var/cache/apt,sharing=locked \
-  --mount=type=cache,target=/var/lib/apt,sharing=locked \
-  apt-get update \
-  && apt-get install -y \
-    --no-install-recommends \
-    build-essential=12.9 \
-    gcc=4:10.2.* \
-    python3-dev=3.9.*
+# Install pytorch before all other pip packages
+# NOTE: there are no pytorch builds for arm64 + cuda, only cpu
+# x86_64/CUDA is default
+RUN --mount=type=cache,target=/root/.cache/pip \
+    python3 -m venv ${VIRTUAL_ENV} &&\
+    if [ "$TARGETPLATFORM" = "linux/arm64" ] || [ "$GPU_DRIVER" = "cpu" ]; then \
+        extra_index_url_arg="--extra-index-url https://download.pytorch.org/whl/cpu"; \
+    elif [ "$GPU_DRIVER" = "rocm" ]; then \
+        extra_index_url_arg="--extra-index-url https://download.pytorch.org/whl/rocm5.4.2"; \
+    else \
+        extra_index_url_arg="--extra-index-url https://download.pytorch.org/whl/cu118"; \
+    fi &&\
+    pip install $extra_index_url_arg \
+        torch==$TORCH_VERSION \
+        torchvision==$TORCHVISION_VERSION
 
-# Prepare pip for buildkit cache
-ARG PIP_CACHE_DIR=/var/cache/buildkit/pip
-ENV PIP_CACHE_DIR ${PIP_CACHE_DIR}
-RUN mkdir -p ${PIP_CACHE_DIR}
+# Install the local package.
+# Editable mode helps use the same image for development:
+# the local working copy can be bind-mounted into the image
+# at path defined by ${INVOKEAI_SRC}
+COPY invokeai ./invokeai
+COPY pyproject.toml ./
+RUN --mount=type=cache,target=/root/.cache/pip \
+    # xformers + triton fails to install on arm64
+    if [ "$GPU_DRIVER" = "cuda" ] && [ "$TARGETPLATFORM" = "linux/amd64" ]; then \
+        pip install -e ".[xformers]"; \
+    else \
+        pip install -e "."; \
+    fi
 
-# Create virtual environment
-RUN --mount=type=cache,target=${PIP_CACHE_DIR} \
-  python3 -m venv "${APPNAME}" \
-  --upgrade-deps
+# #### Build the Web UI ------------------------------------
 
-# Install requirements
-COPY --link pyproject.toml .
-COPY --link invokeai/version/invokeai_version.py invokeai/version/__init__.py invokeai/version/
-ARG PIP_EXTRA_INDEX_URL
-ENV PIP_EXTRA_INDEX_URL ${PIP_EXTRA_INDEX_URL}
-RUN --mount=type=cache,target=${PIP_CACHE_DIR} \
-  "${APPNAME}"/bin/pip install .
+FROM node:18 AS web-builder
+WORKDIR /build
+COPY invokeai/frontend/web/ ./
+RUN --mount=type=cache,target=/usr/lib/node_modules \
+    npm install --include dev
+RUN --mount=type=cache,target=/usr/lib/node_modules \
+    yarn vite build
 
-# Install pyproject.toml
-COPY --link . .
-RUN --mount=type=cache,target=${PIP_CACHE_DIR} \
-  "${APPNAME}/bin/pip" install .
 
-# Build patchmatch
+#### Runtime stage ---------------------------------------
+
+FROM library/ubuntu:22.04 AS runtime
+
+ARG DEBIAN_FRONTEND=noninteractive
+ENV PYTHONUNBUFFERED=1
+ENV PYTHONDONTWRITEBYTECODE=1
+
+RUN apt update && apt install -y --no-install-recommends \
+        git \
+        curl \
+        vim \
+        tmux \
+        ncdu \
+        iotop \
+        bzip2 \
+        gosu \
+        libglib2.0-0 \
+        libgl1-mesa-glx \
+        python3-venv \
+        python3-pip \
+        build-essential \
+        libopencv-dev \
+        libstdc++-10-dev &&\
+    apt-get clean && apt-get autoclean
+
+# globally add magic-wormhole
+# for ease of transferring data to and from the container
+# when running in sandboxed cloud environments; e.g. Runpod etc.
+RUN pip install magic-wormhole
+
+ENV INVOKEAI_SRC=/opt/invokeai
+ENV VIRTUAL_ENV=/opt/venv/invokeai
+ENV INVOKEAI_ROOT=/invokeai
+ENV PATH="$VIRTUAL_ENV/bin:$INVOKEAI_SRC:$PATH"
+
+# --link requires buldkit w/ dockerfile syntax 1.4
+COPY --link --from=builder ${INVOKEAI_SRC} ${INVOKEAI_SRC}
+COPY --link --from=builder ${VIRTUAL_ENV} ${VIRTUAL_ENV}
+COPY --link --from=web-builder /build/dist ${INVOKEAI_SRC}/invokeai/frontend/web/dist
+
+# Link amdgpu.ids for ROCm builds
+# contributed by https://github.com/Rubonnek
+RUN mkdir -p "/opt/amdgpu/share/libdrm" &&\
+  ln -s "/usr/share/libdrm/amdgpu.ids" "/opt/amdgpu/share/libdrm/amdgpu.ids"
+
+WORKDIR ${INVOKEAI_SRC}
+
+# build patchmatch
+RUN cd /usr/lib/$(uname -p)-linux-gnu/pkgconfig/ && ln -sf opencv4.pc opencv.pc
 RUN python3 -c "from patchmatch import patch_match"
 
-#####################
-##  runtime image  ##
-#####################
-FROM python-base AS runtime
+# Create unprivileged user and make the local dir
+RUN useradd --create-home --shell /bin/bash -u 1000 --comment "container local user" invoke
+RUN mkdir -p ${INVOKEAI_ROOT} && chown -R invoke:invoke ${INVOKEAI_ROOT}
 
-# Create a new user
-ARG UNAME=appuser
-RUN useradd \
-  --no-log-init \
-  -m \
-  -U \
-  "${UNAME}"
-
-# Create volume directory
-ARG VOLUME_DIR=/data
-RUN mkdir -p "${VOLUME_DIR}" \
-  && chown -hR "${UNAME}:${UNAME}" "${VOLUME_DIR}"
-
-# Setup runtime environment
-USER ${UNAME}:${UNAME}
-COPY --chown=${UNAME}:${UNAME} --from=pyproject-builder ${APPDIR}/${APPNAME} ${APPNAME}
-ENV INVOKEAI_ROOT ${VOLUME_DIR}
-ENV TRANSFORMERS_CACHE ${VOLUME_DIR}/.cache
-ENV INVOKE_MODEL_RECONFIGURE "--yes --default_only"
-EXPOSE 9090
-ENTRYPOINT [ "invokeai" ]
-CMD [ "--web", "--host", "0.0.0.0", "--port", "9090" ]
-VOLUME [ "${VOLUME_DIR}" ]
+COPY docker/docker-entrypoint.sh ./
+ENTRYPOINT ["/opt/invokeai/docker-entrypoint.sh"]
+CMD ["invokeai-web", "--host", "0.0.0.0"]

docker/README.md

@@ -0,0 +1,77 @@
+# InvokeAI Containerized
+
+All commands are to be run from the `docker` directory: `cd docker`
+
+#### Linux
+
+1. Ensure builkit is enabled in the Docker daemon settings (`/etc/docker/daemon.json`)
+2. Install the `docker compose` plugin using your package manager, or follow a [tutorial](https://www.digitalocean.com/community/tutorials/how-to-install-and-use-docker-compose-on-ubuntu-22-04).
+    - The deprecated `docker-compose` (hyphenated) CLI continues to work for now.
+3. Ensure docker daemon is able to access the GPU.
+    - You may need to install [nvidia-container-toolkit](https://docs.nvidia.com/datacenter/cloud-native/container-toolkit/latest/install-guide.html)
+
+#### macOS
+
+1. Ensure Docker has at least 16GB RAM
+2. Enable VirtioFS for file sharing
+3. Enable `docker compose` V2 support
+
+This is done via Docker Desktop preferences
+
+## Quickstart
+
+
+1. Make a copy of `env.sample` and name it `.env` (`cp env.sample .env` (Mac/Linux) or `copy example.env .env` (Windows)). Make changes as necessary. Set `INVOKEAI_ROOT` to an absolute path to:
+    a. the desired location of the InvokeAI runtime directory, or
+    b. an existing, v3.0.0 compatible runtime directory.
+1. `docker compose up`
+
+The image will be built automatically if needed.
+
+The runtime directory (holding models and outputs) will be created in the location specified by `INVOKEAI_ROOT`. The default location is `~/invokeai`. The runtime directory will be populated with the base configs and models necessary to start generating.
+
+### Use a GPU
+
+- Linux is *recommended* for GPU support in Docker.
+- WSL2 is *required* for Windows.
+- only `x86_64` architecture is supported.
+
+The Docker daemon on the system must be already set up to use the GPU. In case of Linux, this involves installing `nvidia-docker-runtime` and configuring the `nvidia` runtime as default. Steps will be different for AMD. Please see Docker documentation for the most up-to-date instructions for using your GPU with Docker.
+
+## Customize
+
+Check the `.env.sample` file. It contains some environment variables for running in Docker. Copy it, name it `.env`, and fill it in with your own values. Next time you run `docker compose up`, your custom values will be used.
+
+You can also set these values in `docker compose.yml` directly, but `.env` will help avoid conflicts when code is updated.
+
+Example (most values are optional):
+
+```
+INVOKEAI_ROOT=/Volumes/WorkDrive/invokeai
+HUGGINGFACE_TOKEN=the_actual_token
+CONTAINER_UID=1000
+GPU_DRIVER=cuda
+```
+
+## Even Moar Customizing!
+
+See the `docker compose.yaml` file. The `command` instruction can be uncommented and used to run arbitrary startup commands. Some examples below.
+
+### Reconfigure the runtime directory
+
+Can be used to download additional models from the supported model list
+
+In conjunction with `INVOKEAI_ROOT` can be also used to initialize a runtime directory
+
+```
+command:
+  - invokeai-configure
+  - --yes
+```
+
+Or install models:
+
+```
+command:
+  - invokeai-model-install
+```

docker/build.sh

@@ -1,51 +1,11 @@
 #!/usr/bin/env bash
 set -e
 
-# If you want to build a specific flavor, set the CONTAINER_FLAVOR environment variable
-#   e.g. CONTAINER_FLAVOR=cpu ./build.sh
-#   Possible Values are:
-#     - cpu
-#     - cuda
-#     - rocm
-#   Don't forget to also set it when executing run.sh
-#   if it is not set, the script will try to detect the flavor by itself.
-#
-# Doc can be found here:
-#   https://invoke-ai.github.io/InvokeAI/installation/040_INSTALL_DOCKER/
+build_args=""
 
-SCRIPTDIR=$(dirname "${BASH_SOURCE[0]}")
-cd "$SCRIPTDIR" || exit 1
+[[ -f ".env" ]] && build_args=$(awk '$1 ~ /\=[^$]/ {print "--build-arg " $0 " "}' .env)
 
-source ./env.sh
+echo "docker-compose build args:"
+echo $build_args
 
-DOCKERFILE=${INVOKE_DOCKERFILE:-./Dockerfile}
-
-# print the settings
-echo -e "You are using these values:\n"
-echo -e "Dockerfile:\t\t${DOCKERFILE}"
-echo -e "index-url:\t\t${PIP_EXTRA_INDEX_URL:-none}"
-echo -e "Volumename:\t\t${VOLUMENAME}"
-echo -e "Platform:\t\t${PLATFORM}"
-echo -e "Container Registry:\t${CONTAINER_REGISTRY}"
-echo -e "Container Repository:\t${CONTAINER_REPOSITORY}"
-echo -e "Container Tag:\t\t${CONTAINER_TAG}"
-echo -e "Container Flavor:\t${CONTAINER_FLAVOR}"
-echo -e "Container Image:\t${CONTAINER_IMAGE}\n"
-
-# Create docker volume
-if [[ -n "$(docker volume ls -f name="${VOLUMENAME}" -q)" ]]; then
-    echo -e "Volume already exists\n"
-else
-    echo -n "creating docker volume "
-    docker volume create "${VOLUMENAME}"
-fi
-
-# Build Container
-docker build \
-    --platform="${PLATFORM:-linux/amd64}" \
-    --tag="${CONTAINER_IMAGE:-invokeai}" \
-    ${CONTAINER_FLAVOR:+--build-arg="CONTAINER_FLAVOR=${CONTAINER_FLAVOR}"} \
-    ${PIP_EXTRA_INDEX_URL:+--build-arg="PIP_EXTRA_INDEX_URL=${PIP_EXTRA_INDEX_URL}"} \
-    ${PIP_PACKAGE:+--build-arg="PIP_PACKAGE=${PIP_PACKAGE}"} \
-    --file="${DOCKERFILE}" \
-    ..
+docker-compose build $build_args

docker/docker-compose.yml

@@ -0,0 +1,48 @@
+# Copyright (c) 2023 Eugene Brodsky https://github.com/ebr
+
+version: '3.8'
+
+services:
+  invokeai:
+    image: "local/invokeai:latest"
+    # edit below to run on a container runtime other than nvidia-container-runtime.
+    # not yet tested with rocm/AMD GPUs
+    # Comment out the "deploy" section to run on CPU only
+    deploy:
+      resources:
+        reservations:
+          devices:
+            - driver: nvidia
+              count: 1
+              capabilities: [gpu]
+    build:
+      context: ..
+      dockerfile: docker/Dockerfile
+
+    # variables without a default will automatically inherit from the host environment
+    environment:
+      - INVOKEAI_ROOT
+      - HF_HOME
+
+    # Create a .env file in the same directory as this docker-compose.yml file
+    # and populate it with environment variables. See .env.sample
+    env_file:
+      - .env
+
+    ports:
+      - "${INVOKEAI_PORT:-9090}:9090"
+    volumes:
+      - ${INVOKEAI_ROOT:-~/invokeai}:${INVOKEAI_ROOT:-/invokeai}
+      - ${HF_HOME:-~/.cache/huggingface}:${HF_HOME:-/invokeai/.cache/huggingface}
+      # - ${INVOKEAI_MODELS_DIR:-${INVOKEAI_ROOT:-/invokeai/models}}
+      # - ${INVOKEAI_MODELS_CONFIG_PATH:-${INVOKEAI_ROOT:-/invokeai/configs/models.yaml}}
+    tty: true
+    stdin_open: true
+
+    # # Example of running alternative commands/scripts in the container
+    # command:
+    #   - bash
+    #   - -c
+    #   - |
+    #     invokeai-model-install --yes --default-only --config_file ${INVOKEAI_ROOT}/config_custom.yaml
+    #     invokeai-nodes-web --host 0.0.0.0

docker/docker-entrypoint.sh

@@ -0,0 +1,65 @@
+#!/bin/bash
+set -e -o pipefail
+
+### Container entrypoint
+# Runs the CMD as defined by the Dockerfile or passed to `docker run`
+# Can be used to configure the runtime dir
+# Bypass by using ENTRYPOINT or `--entrypoint`
+
+### Set INVOKEAI_ROOT pointing to a valid runtime directory
+# Otherwise configure the runtime dir first.
+
+### Configure the InvokeAI runtime directory (done by default)):
+# docker run --rm -it <this image> --configure
+# or skip with --no-configure
+
+### Set the CONTAINER_UID envvar to match your user.
+# Ensures files created in the container are owned by you:
+#   docker run --rm -it -v /some/path:/invokeai -e CONTAINER_UID=$(id -u) <this image>
+# Default UID: 1000 chosen due to popularity on Linux systems. Possibly 501 on MacOS.
+
+USER_ID=${CONTAINER_UID:-1000}
+USER=invoke
+usermod -u ${USER_ID} ${USER} 1>/dev/null
+
+configure() {
+    # Configure the runtime directory
+    if [[ -f ${INVOKEAI_ROOT}/invokeai.yaml ]]; then
+        echo "${INVOKEAI_ROOT}/invokeai.yaml exists. InvokeAI is already configured."
+        echo "To reconfigure InvokeAI, delete the above file."
+        echo "======================================================================"
+    else
+        mkdir -p ${INVOKEAI_ROOT}
+        chown --recursive ${USER} ${INVOKEAI_ROOT}
+        gosu ${USER} invokeai-configure --yes --default_only
+    fi
+}
+
+## Skip attempting to configure.
+## Must be passed first, before any other args.
+if [[ $1 != "--no-configure" ]]; then
+    configure
+else
+    shift
+fi
+
+### Set the $PUBLIC_KEY env var to enable SSH access.
+# We do not install openssh-server in the image by default to avoid bloat.
+# but it is useful to have the full SSH server e.g. on Runpod.
+# (use SCP to copy files to/from the image, etc)
+if [[ -v "PUBLIC_KEY" ]] && [[ ! -d "${HOME}/.ssh" ]]; then
+    apt-get update
+    apt-get install -y openssh-server
+    pushd $HOME
+    mkdir -p .ssh
+    echo ${PUBLIC_KEY} > .ssh/authorized_keys
+    chmod -R 700 .ssh
+    popd
+    service ssh start
+fi
+
+
+cd ${INVOKEAI_ROOT}
+
+# Run the CMD as the Container User (not root).
+exec gosu ${USER} "$@"

docker/env.sh

@@ -1,54 +0,0 @@
-#!/usr/bin/env bash
-
-# This file is used to set environment variables for the build.sh and run.sh scripts.
-
-# Try to detect the container flavor if no PIP_EXTRA_INDEX_URL got specified
-if [[ -z "$PIP_EXTRA_INDEX_URL" ]]; then
-
-  # Activate virtual environment if not already activated and exists
-  if [[ -z $VIRTUAL_ENV ]]; then
-    [[ -e "$(dirname "${BASH_SOURCE[0]}")/../.venv/bin/activate" ]] \
-      && source "$(dirname "${BASH_SOURCE[0]}")/../.venv/bin/activate" \
-      && echo "Activated virtual environment: $VIRTUAL_ENV"
-  fi
-
-  # Decide which container flavor to build if not specified
-  if [[ -z "$CONTAINER_FLAVOR" ]] && python -c "import torch" &>/dev/null; then
-    # Check for CUDA and ROCm
-    CUDA_AVAILABLE=$(python -c "import torch;print(torch.cuda.is_available())")
-    ROCM_AVAILABLE=$(python -c "import torch;print(torch.version.hip is not None)")
-    if [[ "${CUDA_AVAILABLE}" == "True" ]]; then
-      CONTAINER_FLAVOR="cuda"
-    elif [[ "${ROCM_AVAILABLE}" == "True" ]]; then
-      CONTAINER_FLAVOR="rocm"
-    else
-      CONTAINER_FLAVOR="cpu"
-    fi
-  fi
-
-  # Set PIP_EXTRA_INDEX_URL based on container flavor
-  if [[ "$CONTAINER_FLAVOR" == "rocm" ]]; then
-    PIP_EXTRA_INDEX_URL="https://download.pytorch.org/whl/rocm"
-  elif [[ "$CONTAINER_FLAVOR" == "cpu" ]]; then
-    PIP_EXTRA_INDEX_URL="https://download.pytorch.org/whl/cpu"
-  # elif [[ -z "$CONTAINER_FLAVOR" || "$CONTAINER_FLAVOR" == "cuda" ]]; then
-  #   PIP_PACKAGE=${PIP_PACKAGE-".[xformers]"}
-  fi
-fi
-
-# Variables shared by build.sh and run.sh
-REPOSITORY_NAME="${REPOSITORY_NAME-$(basename "$(git rev-parse --show-toplevel)")}"
-REPOSITORY_NAME="${REPOSITORY_NAME,,}"
-VOLUMENAME="${VOLUMENAME-"${REPOSITORY_NAME}_data"}"
-ARCH="${ARCH-$(uname -m)}"
-PLATFORM="${PLATFORM-linux/${ARCH}}"
-INVOKEAI_BRANCH="${INVOKEAI_BRANCH-$(git branch --show)}"
-CONTAINER_REGISTRY="${CONTAINER_REGISTRY-"ghcr.io"}"
-CONTAINER_REPOSITORY="${CONTAINER_REPOSITORY-"$(whoami)/${REPOSITORY_NAME}"}"
-CONTAINER_FLAVOR="${CONTAINER_FLAVOR-cuda}"
-CONTAINER_TAG="${CONTAINER_TAG-"${INVOKEAI_BRANCH##*/}-${CONTAINER_FLAVOR}"}"
-CONTAINER_IMAGE="${CONTAINER_REGISTRY}/${CONTAINER_REPOSITORY}:${CONTAINER_TAG}"
-CONTAINER_IMAGE="${CONTAINER_IMAGE,,}"
-
-# enable docker buildkit
-export DOCKER_BUILDKIT=1

docker/run.sh

@@ -1,41 +1,8 @@
 #!/usr/bin/env bash
 set -e
 
-# How to use: https://invoke-ai.github.io/InvokeAI/installation/040_INSTALL_DOCKER/
-
 SCRIPTDIR=$(dirname "${BASH_SOURCE[0]}")
 cd "$SCRIPTDIR" || exit 1
 
-source ./env.sh
-
-# Create outputs directory if it does not exist
-[[ -d ./outputs ]] || mkdir ./outputs
-
-echo -e "You are using these values:\n"
-echo -e "Volumename:\t${VOLUMENAME}"
-echo -e "Invokeai_tag:\t${CONTAINER_IMAGE}"
-echo -e "local Models:\t${MODELSPATH:-unset}\n"
-
-docker run \
-  --interactive \
-  --tty \
-  --rm \
-  --platform="${PLATFORM}" \
-  --name="${REPOSITORY_NAME}" \
-  --hostname="${REPOSITORY_NAME}" \
-  --mount type=volume,volume-driver=local,source="${VOLUMENAME}",target=/data \
-  --mount type=bind,source="$(pwd)"/outputs/,target=/data/outputs/ \
-  ${MODELSPATH:+--mount="type=bind,source=${MODELSPATH},target=/data/models"} \
-  ${HUGGING_FACE_HUB_TOKEN:+--env="HUGGING_FACE_HUB_TOKEN=${HUGGING_FACE_HUB_TOKEN}"} \
-  --publish=9090:9090 \
-  --cap-add=sys_nice \
-  ${GPU_FLAGS:+--gpus="${GPU_FLAGS}"} \
-  "${CONTAINER_IMAGE}" ${@:+$@}
-
-echo -e "\nCleaning trash folder ..."
-for f in outputs/.Trash*; do
-  if [ -e "$f" ]; then
-    rm -Rf "$f"
-    break
-  fi
-done
+docker-compose up --build -d
+docker-compose logs -f

docker/runpod-readme.md

@@ -0,0 +1,60 @@
+# InvokeAI - A Stable Diffusion Toolkit
+
+Stable Diffusion distribution by InvokeAI: https://github.com/invoke-ai
+
+The Docker image tracks the `main` branch of the InvokeAI project, which means it includes the latest features, but may contain some bugs.
+
+Your working directory is mounted under the `/workspace` path inside the pod. The models are in `/workspace/invokeai/models`, and outputs are in `/workspace/invokeai/outputs`.
+
+> **Only the /workspace directory will persist between pod restarts!**
+
+> **If you _terminate_ (not just _stop_) the pod, the /workspace will be lost.**
+
+## Quickstart
+
+1. Launch a pod from this template. **It will take about 5-10 minutes to run through the initial setup**. Be patient.
+1. Wait for the application to load.
+    - TIP: you know it's ready when the CPU usage goes idle
+    - You can also check the logs for a line that says "_Point your browser at..._"
+1. Open the Invoke AI web UI: click the `Connect` => `connect over  HTTP` button.
+1. Generate some art!
+
+## Other things you can do
+
+At any point you may edit the pod configuration and set an arbitrary Docker command. For example, you could run a command to downloads some models using `curl`, or fetch some images and place them into your outputs to continue a working session.
+
+If you need to run *multiple commands*, define them in the Docker Command field like this:
+
+`bash -c "cd ${INVOKEAI_ROOT}/outputs; wormhole receive 2-foo-bar; invoke.py --web --host 0.0.0.0"`
+
+### Copying your data in and out of the pod
+
+This image includes a couple of handy tools to help you get the data into the pod (such as your custom models or embeddings), and out of the pod (such as downloading your outputs). Here are your options for getting your data in and out of the pod:
+
+- **SSH server**:
+    1. Make sure to create and set your Public Key in the RunPod settings (follow the official instructions)
+    1. Add an exposed port 22 (TCP) in the pod settings!
+    1. When your pod restarts, you will see a new entry in the `Connect` dialog. Use this SSH server to `scp` or `sftp` your files as necessary, or SSH into the pod using the fully fledged SSH server.
+
+- [**Magic Wormhole**](https://magic-wormhole.readthedocs.io/en/latest/welcome.html):
+    1. On your computer, `pip install magic-wormhole` (see above instructions for details)
+    1. Connect to the command line **using the "light" SSH client** or the browser-based console. _Currently there's a bug where `wormhole` isn't available when connected to "full" SSH server, as described above_.
+    1. `wormhole send /workspace/invokeai/outputs` will send the entire `outputs` directory. You can also send individual files.
+    1. Once packaged, you will see a `wormhole receive <123-some-words>` command. Copy it
+    1. Paste this command into the terminal on your local machine to securely download the payload.
+    1. It works the same in reverse: you can `wormhole send` some models from your computer to the pod. Again, save your files somewhere in `/workspace` or they will be lost when the pod is stopped.
+
+- **RunPod's Cloud Sync feature** may be used to sync the persistent volume to cloud storage. You could, for example, copy the entire `/workspace` to S3, add some custom models to it, and copy it back from S3 when launching new pod configurations. Follow the Cloud Sync instructions.
+
+
+### Disable the NSFW checker
+
+The NSFW checker is enabled by default. To disable it, edit the pod configuration and set the following command:
+
+```
+invoke --web --host 0.0.0.0 --no-nsfw_checker
+```
+
+---
+
+Template ©2023 Eugene Brodsky [ebr](https://github.com/ebr)

docs/CHANGELOG.md

@@ -617,8 +617,6 @@ sections describe what's new for InvokeAI.
 - `dream.py` script renamed `invoke.py`. A `dream.py` script wrapper remains for
   backward compatibility.
 - Completely new WebGUI - launch with `python3 scripts/invoke.py --web`
-- Support for [inpainting](deprecated/INPAINTING.md) and
-  [outpainting](features/OUTPAINTING.md)
 - img2img runs on all k\* samplers
 - Support for
   [negative prompts](features/PROMPTS.md#negative-and-unconditioned-prompts)

docs/contributing/INVOCATIONS.md

@@ -1,8 +1,521 @@
 # Invocations
 
-Invocations represent a single operation, its inputs, and its outputs. These
-operations and their outputs can be chained together to generate and modify
-images.
+Features in InvokeAI are added in the form of modular node-like systems called
+**Invocations**.
+
+An Invocation is simply a single operation that takes in some inputs and gives
+out some outputs. We can then chain multiple Invocations together to create more
+complex functionality.
+
+## Invocations Directory
+
+InvokeAI Invocations can be found in the `invokeai/app/invocations` directory.
+
+You can add your new functionality to one of the existing Invocations in this
+directory or create a new file in this directory as per your needs.
+
+**Note:** _All Invocations must be inside this directory for InvokeAI to
+recognize them as valid Invocations._
+
+## Creating A New Invocation
+
+In order to understand the process of creating a new Invocation, let us actually
+create one.
+
+In our example, let us create an Invocation that will take in an image, resize
+it and output the resized image.
+
+The first set of things we need to do when creating a new Invocation are -
+
+- Create a new class that derives from a predefined parent class called
+  `BaseInvocation`.
+- The name of every Invocation must end with the word `Invocation` in order for
+  it to be recognized as an Invocation.
+- Every Invocation must have a `docstring` that describes what this Invocation
+  does.
+- Every Invocation must have a unique `type` field defined which becomes its
+  indentifier.
+- Invocations are strictly typed. We make use of the native
+  [typing](https://docs.python.org/3/library/typing.html) library and the
+  installed [pydantic](https://pydantic-docs.helpmanual.io/) library for
+  validation.
+
+So let us do that.
+
+```python
+from typing import Literal
+from .baseinvocation import BaseInvocation
+
+class ResizeInvocation(BaseInvocation):
+    '''Resizes an image'''
+    type: Literal['resize'] = 'resize'
+```
+
+That's great.
+
+Now we have setup the base of our new Invocation. Let us think about what inputs
+our Invocation takes.
+
+- We need an `image` that we are going to resize.
+- We will need new `width` and `height` values to which we need to resize the
+  image to.
+
+### **Inputs**
+
+Every Invocation input is a pydantic `Field` and like everything else should be
+strictly typed and defined.
+
+So let us create these inputs for our Invocation. First up, the `image` input we
+need. Generally, we can use standard variable types in Python but InvokeAI
+already has a custom `ImageField` type that handles all the stuff that is needed
+for image inputs.
+
+But what is this `ImageField` ..? It is a special class type specifically
+written to handle how images are dealt with in InvokeAI. We will cover how to
+create your own custom field types later in this guide. For now, let's go ahead
+and use it.
+
+```python
+from typing import Literal, Union
+from pydantic import Field
+
+from .baseinvocation import BaseInvocation
+from ..models.image import ImageField
+
+class ResizeInvocation(BaseInvocation):
+    '''Resizes an image'''
+    type: Literal['resize'] = 'resize'
+
+    # Inputs
+    image: Union[ImageField, None] = Field(description="The input image", default=None)
+```
+
+Let us break down our input code.
+
+```python
+image: Union[ImageField, None] = Field(description="The input image", default=None)
+```
+
+| Part      | Value                                                | Description                                                                                        |
+| --------- | ---------------------------------------------------- | -------------------------------------------------------------------------------------------------- |
+| Name      | `image`                                              | The variable that will hold our image                                                              |
+| Type Hint | `Union[ImageField, None]`                            | The types for our field. Indicates that the image can either be an `ImageField` type or `None`     |
+| Field     | `Field(description="The input image", default=None)` | The image variable is a field which needs a description and a default value that we set to `None`. |
+
+Great. Now let us create our other inputs for `width` and `height`
+
+```python
+from typing import Literal, Union
+from pydantic import Field
+
+from .baseinvocation import BaseInvocation
+from ..models.image import ImageField
+
+class ResizeInvocation(BaseInvocation):
+    '''Resizes an image'''
+    type: Literal['resize'] = 'resize'
+
+    # Inputs
+    image: Union[ImageField, None] = Field(description="The input image", default=None)
+    width: int = Field(default=512, ge=64, le=2048, description="Width of the new image")
+    height: int = Field(default=512, ge=64, le=2048, description="Height of the new image")
+```
+
+As you might have noticed, we added two new parameters to the field type for
+`width` and `height` called `gt` and `le`. These basically stand for _greater
+than or equal to_ and _less than or equal to_. There are various other param
+types for field that you can find on the **pydantic** documentation.
+
+**Note:** _Any time it is possible to define constraints for our field, we
+should do it so the frontend has more information on how to parse this field._
+
+Perfect. We now have our inputs. Let us do something with these.
+
+### **Invoke Function**
+
+The `invoke` function is where all the magic happens. This function provides you
+the `context` parameter that is of the type `InvocationContext` which will give
+you access to the current context of the generation and all the other services
+that are provided by it by InvokeAI.
+
+Let us create this function first.
+
+```python
+from typing import Literal, Union
+from pydantic import Field
+
+from .baseinvocation import BaseInvocation, InvocationContext
+from ..models.image import ImageField
+
+class ResizeInvocation(BaseInvocation):
+    '''Resizes an image'''
+    type: Literal['resize'] = 'resize'
+
+    # Inputs
+    image: Union[ImageField, None] = Field(description="The input image", default=None)
+    width: int = Field(default=512, ge=64, le=2048, description="Width of the new image")
+    height: int = Field(default=512, ge=64, le=2048, description="Height of the new image")
+
+    def invoke(self, context: InvocationContext):
+        pass
+```
+
+### **Outputs**
+
+The output of our Invocation will be whatever is returned by this `invoke`
+function. Like with our inputs, we need to strongly type and define our outputs
+too.
+
+What is our output going to be? Another image. Normally you'd have to create a
+type for this but InvokeAI already offers you an `ImageOutput` type that handles
+all the necessary info related to image outputs. So let us use that.
+
+We will cover how to create your own output types later in this guide.
+
+```python
+from typing import Literal, Union
+from pydantic import Field
+
+from .baseinvocation import BaseInvocation, InvocationContext
+from ..models.image import ImageField
+from .image import ImageOutput
+
+class ResizeInvocation(BaseInvocation):
+    '''Resizes an image'''
+    type: Literal['resize'] = 'resize'
+
+    # Inputs
+    image: Union[ImageField, None] = Field(description="The input image", default=None)
+    width: int = Field(default=512, ge=64, le=2048, description="Width of the new image")
+    height: int = Field(default=512, ge=64, le=2048, description="Height of the new image")
+
+    def invoke(self, context: InvocationContext) -> ImageOutput:
+        pass
+```
+
+Perfect. Now that we have our Invocation setup, let us do what we want to do.
+
+- We will first load the image. Generally we do this using the `PIL` library but
+  we can use one of the services provided by InvokeAI to load the image.
+- We will resize the image using `PIL` to our input data.
+- We will output this image in the format we set above.
+
+So let's do that.
+
+```python
+from typing import Literal, Union
+from pydantic import Field
+
+from .baseinvocation import BaseInvocation, InvocationContext
+from ..models.image import ImageField, ResourceOrigin, ImageCategory
+from .image import ImageOutput
+
+class ResizeInvocation(BaseInvocation):
+    '''Resizes an image'''
+    type: Literal['resize'] = 'resize'
+
+    # Inputs
+    image: Union[ImageField, None] = Field(description="The input image", default=None)
+    width: int = Field(default=512, ge=64, le=2048, description="Width of the new image")
+    height: int = Field(default=512, ge=64, le=2048, description="Height of the new image")
+
+    def invoke(self, context: InvocationContext) -> ImageOutput:
+        # Load the image using InvokeAI's predefined Image Service.
+        image = context.services.images.get_pil_image(self.image.image_origin, self.image.image_name)
+
+        # Resizing the image
+        # Because we used the above service, we already have a PIL image. So we can simply resize.
+        resized_image = image.resize((self.width, self.height))
+
+        # Preparing the image for output using InvokeAI's predefined Image Service.
+        output_image = context.services.images.create(
+            image=resized_image,
+            image_origin=ResourceOrigin.INTERNAL,
+            image_category=ImageCategory.GENERAL,
+            node_id=self.id,
+            session_id=context.graph_execution_state_id,
+            is_intermediate=self.is_intermediate,
+        )
+
+        # Returning the Image
+        return ImageOutput(
+            image=ImageField(
+                image_name=output_image.image_name,
+                image_origin=output_image.image_origin,
+            ),
+            width=output_image.width,
+            height=output_image.height,
+        )
+```
+
+**Note:** Do not be overwhelmed by the `ImageOutput` process. InvokeAI has a
+certain way that the images need to be dispatched in order to be stored and read
+correctly. In 99% of the cases when dealing with an image output, you can simply
+copy-paste the template above.
+
+That's it. You made your own **Resize Invocation**.
+
+## Result
+
+Once you make your Invocation correctly, the rest of the process is fully
+automated for you.
+
+When you launch InvokeAI, you can go to `http://localhost:9090/docs` and see
+your new Invocation show up there with all the relevant info.
+
+![resize invocation](../assets/contributing/resize_invocation.png)
+
+When you launch the frontend UI, you can go to the Node Editor tab and find your
+new Invocation ready to be used.
+
+![resize node editor](../assets/contributing/resize_node_editor.png)
+
+# Advanced
+
+## Custom Input Fields
+
+Now that you know how to create your own Invocations, let us dive into slightly
+more advanced topics.
+
+While creating your own Invocations, you might run into a scenario where the
+existing input types in InvokeAI do not meet your requirements. In such cases,
+you can create your own input types.
+
+Let us create one as an example. Let us say we want to create a color input
+field that represents a color code. But before we start on that here are some
+general good practices to keep in mind.
+
+**Good Practices**
+
+- There is no naming convention for input fields but we highly recommend that
+  you name it something appropriate like `ColorField`.
+- It is not mandatory but it is heavily recommended to add a relevant
+  `docstring` to describe your input field.
+- Keep your field in the same file as the Invocation that it is made for or in
+  another file where it is relevant.
+
+All input types a class that derive from the `BaseModel` type from `pydantic`.
+So let's create one.
+
+```python
+from pydantic import BaseModel
+
+class ColorField(BaseModel):
+    '''A field that holds the rgba values of a color'''
+    pass
+```
+
+Perfect. Now let us create our custom inputs for our field. This is exactly
+similar how you created input fields for your Invocation. All the same rules
+apply. Let us create four fields representing the _red(r)_, _blue(b)_,
+_green(g)_ and _alpha(a)_ channel of the color.
+
+```python
+class ColorField(BaseModel):
+    '''A field that holds the rgba values of a color'''
+    r: int = Field(ge=0, le=255, description="The red channel")
+    g: int = Field(ge=0, le=255, description="The green channel")
+    b: int = Field(ge=0, le=255, description="The blue channel")
+    a: int = Field(ge=0, le=255, description="The alpha channel")
+```
+
+That's it. We now have a new input field type that we can use in our Invocations
+like this.
+
+```python
+color: ColorField = Field(default=ColorField(r=0, g=0, b=0, a=0), description='Background color of an image')
+```
+
+**Extra Config**
+
+All input fields also take an additional `Config` class that you can use to do
+various advanced things like setting required parameters and etc.
+
+Let us do that for our _ColorField_ and enforce all the values because we did
+not define any defaults for our fields.
+
+```python
+class ColorField(BaseModel):
+    '''A field that holds the rgba values of a color'''
+    r: int = Field(ge=0, le=255, description="The red channel")
+    g: int = Field(ge=0, le=255, description="The green channel")
+    b: int = Field(ge=0, le=255, description="The blue channel")
+    a: int = Field(ge=0, le=255, description="The alpha channel")
+
+    class Config:
+        schema_extra = {"required": ["r", "g", "b", "a"]}
+```
+
+Now it becomes mandatory for the user to supply all the values required by our
+input field.
+
+We will discuss the `Config` class in extra detail later in this guide and how
+you can use it to make your Invocations more robust.
+
+## Custom Output Types
+
+Like with custom inputs, sometimes you might find yourself needing custom
+outputs that InvokeAI does not provide. We can easily set one up.
+
+Now that you are familiar with Invocations and Inputs, let us use that knowledge
+to put together a custom output type for an Invocation that returns _width_,
+_height_ and _background_color_ that we need to create a blank image.
+
+- A custom output type is a class that derives from the parent class of
+  `BaseInvocationOutput`.
+- It is not mandatory but we recommend using names ending with `Output` for
+  output types. So we'll call our class `BlankImageOutput`
+- It is not mandatory but we highly recommend adding a `docstring` to describe
+  what your output type is for.
+- Like Invocations, each output type should have a `type` variable that is
+  **unique**
+
+Now that we know the basic rules for creating a new output type, let us go ahead
+and make it.
+
+```python
+from typing import Literal
+from pydantic import Field
+
+from .baseinvocation import BaseInvocationOutput
+
+class BlankImageOutput(BaseInvocationOutput):
+    '''Base output type for creating a blank image'''
+    type: Literal['blank_image_output'] = 'blank_image_output'
+
+    # Inputs
+    width: int = Field(description='Width of blank image')
+    height: int = Field(description='Height of blank image')
+    bg_color: ColorField = Field(description='Background color of blank image')
+
+    class Config:
+        schema_extra = {"required": ["type", "width", "height", "bg_color"]}
+```
+
+All set. We now have an output type that requires what we need to create a
+blank_image. And if you noticed it, we even used the `Config` class to ensure
+the fields are required.
+
+## Custom Configuration
+
+As you might have noticed when making inputs and outputs, we used a class called
+`Config` from _pydantic_ to further customize them. Because our inputs and
+outputs essentially inherit from _pydantic_'s `BaseModel` class, all
+[configuration options](https://docs.pydantic.dev/latest/usage/schema/#schema-customization)
+that are valid for _pydantic_ classes are also valid for our inputs and outputs.
+You can do the same for your Invocations too but InvokeAI makes our life a
+little bit easier on that end.
+
+InvokeAI provides a custom configuration class called `InvocationConfig`
+particularly for configuring Invocations. This is exactly the same as the raw
+`Config` class from _pydantic_ with some extra stuff on top to help faciliate
+parsing of the scheme in the frontend UI.
+
+At the current moment, tihs `InvocationConfig` class is further improved with
+the following features related the `ui`.
+
+| Config Option | Field Type                                                                                                    | Example                                                                                                               |
+| ------------- | ------------------------------------------------------------------------------------------------------------- | --------------------------------------------------------------------------------------------------------------------- |
+| type_hints    | `Dict[str, Literal["integer", "float", "boolean", "string", "enum", "image", "latents", "model", "control"]]` | `type_hint: "model"` provides type hints related to the model like displaying a list of available models              |
+| tags          | `List[str]`                                                                                                   | `tags: ['resize', 'image']` will classify your invocation under the tags of resize and image.                         |
+| title         | `str`                                                                                                         | `title: 'Resize Image` will rename your to this custom title rather than infer from the name of the Invocation class. |
+
+So let us update your `ResizeInvocation` with some extra configuration and see
+how that works.
+
+```python
+from typing import Literal, Union
+from pydantic import Field
+
+from .baseinvocation import BaseInvocation, InvocationContext, InvocationConfig
+from ..models.image import ImageField, ResourceOrigin, ImageCategory
+from .image import ImageOutput
+
+class ResizeInvocation(BaseInvocation):
+    '''Resizes an image'''
+    type: Literal['resize'] = 'resize'
+
+    # Inputs
+    image: Union[ImageField, None] = Field(description="The input image", default=None)
+    width: int = Field(default=512, ge=64, le=2048, description="Width of the new image")
+    height: int = Field(default=512, ge=64, le=2048, description="Height of the new image")
+
+    class Config(InvocationConfig):
+        schema_extra: {
+            ui: {
+                tags: ['resize', 'image'],
+                title: ['My Custom Resize']
+            }
+        }
+
+    def invoke(self, context: InvocationContext) -> ImageOutput:
+        # Load the image using InvokeAI's predefined Image Service.
+        image = context.services.images.get_pil_image(self.image.image_origin, self.image.image_name)
+
+        # Resizing the image
+        # Because we used the above service, we already have a PIL image. So we can simply resize.
+        resized_image = image.resize((self.width, self.height))
+
+        # Preparing the image for output using InvokeAI's predefined Image Service.
+        output_image = context.services.images.create(
+            image=resized_image,
+            image_origin=ResourceOrigin.INTERNAL,
+            image_category=ImageCategory.GENERAL,
+            node_id=self.id,
+            session_id=context.graph_execution_state_id,
+            is_intermediate=self.is_intermediate,
+        )
+
+        # Returning the Image
+        return ImageOutput(
+            image=ImageField(
+                image_name=output_image.image_name,
+                image_origin=output_image.image_origin,
+            ),
+            width=output_image.width,
+            height=output_image.height,
+        )
+```
+
+We now customized our code to let the frontend know that our Invocation falls
+under `resize` and `image` categories. So when the user searches for these
+particular words, our Invocation will show up too.
+
+We also set a custom title for our Invocation. So instead of being called
+`Resize`, it will be called `My Custom Resize`.
+
+As simple as that.
+
+As time goes by, InvokeAI will further improve and add more customizability for
+Invocation configuration. We will have more documentation regarding this at a
+later time.
+
+# **[TODO]**
+
+## Custom Components For Frontend
+
+Every backend input type should have a corresponding frontend component so the
+UI knows what to render when you use a particular field type.
+
+If you are using existing field types, we already have components for those. So
+you don't have to worry about creating anything new. But this might not always
+be the case. Sometimes you might want to create new field types and have the
+frontend UI deal with it in a different way.
+
+This is where we venture into the world of React and Javascript and create our
+own new components for our Invocations. Do not fear the world of JS. It's
+actually pretty straightforward.
+
+Let us create a new component for our custom color field we created above. When
+we use a color field, let us say we want the UI to display a color picker for
+the user to pick from rather than entering values. That is what we will build
+now.
+
+---
+
+# OLD -- TO BE DELETED OR MOVED LATER
+
+---
 
 ## Creating a new invocation
 

docs/contributing/LOCAL_DEVELOPMENT.md

@@ -81,3 +81,193 @@ pytest --cov; open ./coverage/html/index.html
 <!--#TODO: get input from blessedcoolant here, for the moment inserted the frontend README via snippets extension.-->
 
 --8<-- "invokeai/frontend/web/README.md"
+
+## Developing InvokeAI in VSCode
+
+VSCode offers some nice tools:
+
+- python debugger
+- automatic `venv` activation
+- remote dev (e.g. run InvokeAI on a beefy linux desktop while you type in
+  comfort on your macbook)
+
+### Setup
+
+You'll need the
+[Python](https://marketplace.visualstudio.com/items?itemName=ms-python.python)
+and
+[Pylance](https://marketplace.visualstudio.com/items?itemName=ms-python.vscode-pylance)
+extensions installed first.
+
+It's also really handy to install the `Jupyter` extensions:
+
+- [Jupyter](https://marketplace.visualstudio.com/items?itemName=ms-toolsai.jupyter)
+- [Jupyter Cell Tags](https://marketplace.visualstudio.com/items?itemName=ms-toolsai.vscode-jupyter-cell-tags)
+- [Jupyter Notebook Renderers](https://marketplace.visualstudio.com/items?itemName=ms-toolsai.jupyter-renderers)
+- [Jupyter Slide Show](https://marketplace.visualstudio.com/items?itemName=ms-toolsai.vscode-jupyter-slideshow)
+
+#### InvokeAI workspace
+
+Creating a VSCode workspace for working on InvokeAI is highly recommended. It
+can hold InvokeAI-specific settings and configs.
+
+To make a workspace:
+
+- Open the InvokeAI repo dir in VSCode
+- `File` > `Save Workspace As` > save it _outside_ the repo
+
+#### Default python interpreter (i.e. automatic virtual environment activation)
+
+- Use command palette to run command
+  `Preferences: Open Workspace Settings (JSON)`
+- Add `python.defaultInterpreterPath` to `settings`, pointing to your `venv`'s
+  python
+
+Should look something like this:
+
+```jsonc
+{
+  // I like to have all InvokeAI-related folders in my workspace
+  "folders": [
+    {
+      // repo root
+      "path": "InvokeAI"
+    },
+    {
+      // InvokeAI root dir, where `invokeai.yaml` lives
+      "path": "/path/to/invokeai_root"
+    }
+  ],
+  "settings": {
+    // Where your InvokeAI `venv`'s python executable lives
+    "python.defaultInterpreterPath": "/path/to/invokeai_root/.venv/bin/python"
+  }
+}
+```
+
+Now when you open the VSCode integrated terminal, or do anything that needs to
+run python, it will automatically be in your InvokeAI virtual environment.
+
+Bonus: When you create a Jupyter notebook, when you run it, you'll be prompted
+for the python interpreter to run in. This will default to your `venv` python,
+and so you'll have access to the same python environment as the InvokeAI app.
+
+This is _super_ handy.
+
+#### Debugging configs with `launch.json`
+
+Debugging configs are managed in a `launch.json` file. Like most VSCode configs,
+these can be scoped to a workspace or folder.
+
+Follow the [official guide](https://code.visualstudio.com/docs/python/debugging)
+to set up your `launch.json` and try it out.
+
+Now we can create the InvokeAI debugging configs:
+
+```jsonc
+{
+  // Use IntelliSense to learn about possible attributes.
+  // Hover to view descriptions of existing attributes.
+  // For more information, visit: https://go.microsoft.com/fwlink/?linkid=830387
+  "version": "0.2.0",
+  "configurations": [
+    {
+      // Run the InvokeAI backend & serve the pre-built UI
+      "name": "InvokeAI Web",
+      "type": "python",
+      "request": "launch",
+      "program": "scripts/invokeai-web.py",
+      "args": [
+        // Your InvokeAI root dir (where `invokeai.yaml` lives)
+        "--root",
+        "/path/to/invokeai_root",
+        // Access the app from anywhere on your local network
+        "--host",
+        "0.0.0.0"
+      ],
+      "justMyCode": true
+    },
+    {
+      // Run the nodes-based CLI
+      "name": "InvokeAI CLI",
+      "type": "python",
+      "request": "launch",
+      "program": "scripts/invokeai-cli.py",
+      "justMyCode": true
+    },
+    {
+      // Run tests
+      "name": "InvokeAI Test",
+      "type": "python",
+      "request": "launch",
+      "module": "pytest",
+      "args": ["--capture=no"],
+      "justMyCode": true
+    },
+    {
+      // Run a single test
+      "name": "InvokeAI Single Test",
+      "type": "python",
+      "request": "launch",
+      "module": "pytest",
+      "args": [
+        // Change this to point to the specific test you are working on
+        "tests/nodes/test_invoker.py"
+      ],
+      "justMyCode": true
+    },
+    {
+      // This is the default, useful to just run a single file
+      "name": "Python: File",
+      "type": "python",
+      "request": "launch",
+      "program": "${file}",
+      "justMyCode": true
+    }
+  ]
+}
+```
+
+You'll see these configs in the debugging configs drop down. Running them will
+start InvokeAI with attached debugger, in the correct environment, and work just
+like the normal app.
+
+Enjoy debugging InvokeAI with ease (not that we have any bugs of course).
+
+#### Remote dev
+
+This is very easy to set up and provides the same very smooth experience as
+local development. Environments and debugging, as set up above, just work,
+though you'd need to recreate the workspace and debugging configs on the remote.
+
+Consult the
+[official guide](https://code.visualstudio.com/docs/remote/remote-overview) to
+get it set up.
+
+Suggest using VSCode's included settings sync so that your remote dev host has
+all the same app settings and extensions automagically.
+
+##### One remote dev gotcha
+
+I've found the automatic port forwarding to be very flakey. You can disable it
+in `Preferences: Open Remote Settings (ssh: hostname)`. Search for
+`remote.autoForwardPorts` and untick the box.
+
+To forward ports very reliably, use SSH on the remote dev client (e.g. your
+macbook). Here's how to forward both backend API port (`9090`) and the frontend
+live dev server port (`5173`):
+
+```bash
+ssh \
+    -L 9090:localhost:9090 \
+    -L 5173:localhost:5173 \
+    user@remote-dev-host
+```
+
+The forwarding stops when you close the terminal window, so suggest to do this
+_outside_ the VSCode integrated terminal in case you need to restart VSCode for
+an extension update or something
+
+Now, on your remote dev client, you can open `localhost:9090` and access the UI,
+now served from the remote dev host, just the same as if it was running on the
+client.

docs/features/CONCEPTS.md

@@ -1,9 +1,12 @@
 ---
-title: Concepts Library
+title: Concepts 
 ---
 
 # :material-library-shelves: The Hugging Face Concepts Library and Importing Textual Inversion files
 
+With the advances in research, many new capabilities are available to customize the knowledge and understanding of novel concepts not originally contained in the base model. 
+
+
 ## Using Textual Inversion Files
 
 Textual inversion (TI) files are small models that customize the output of
@@ -12,18 +15,16 @@ and artistic styles. They are also known as "embeds" in the machine learning
 world.
 
 Each TI file introduces one or more vocabulary terms to the SD model. These are
-known in InvokeAI as "triggers." Triggers are often, but not always, denoted
-using angle brackets as in "<trigger-phrase>". The two most common type of
+known in InvokeAI as "triggers." Triggers are denoted using angle brackets 
+as in "<trigger-phrase>". The two most common type of
 TI files that you'll encounter are `.pt` and `.bin` files, which are produced by
 different TI training packages. InvokeAI supports both formats, but its
-[built-in TI training system](TEXTUAL_INVERSION.md) produces `.pt`.
+[built-in TI training system](TRAINING.md) produces `.pt`.
 
 The [Hugging Face company](https://huggingface.co/sd-concepts-library) has
 amassed a large ligrary of >800 community-contributed TI files covering a
-broad range of subjects and styles. InvokeAI has built-in support for this
-library which downloads and merges TI files automatically upon request. You can
-also install your own or others' TI files by placing them in a designated
-directory.
+broad range of subjects and styles. You can also install your own or others' TI files 
+by placing them in the designated directory for the compatible model type
 
 ### An Example
 
@@ -41,66 +42,43 @@ You can also combine styles and concepts:
   | :--------------------------------------------------------: |
   | ![](../assets/concepts/image5.png)                         |
 </figure>
-## Using a Hugging Face Concept
 
-!!! warning "Authenticating to HuggingFace"
-
-    Some concepts require valid authentication to HuggingFace. Without it, they will not be downloaded
-    and will be silently ignored.
-
-    If you used an installer to install InvokeAI, you may have already set a HuggingFace token.
-    If you skipped this step, you can:
-
-    - run the InvokeAI configuration script again (if you used a manual installer): `invokeai-configure`
-    - set one of the `HUGGINGFACE_TOKEN` or `HUGGING_FACE_HUB_TOKEN` environment variables to contain your token
-
-    Finally, if you already used any HuggingFace library on your computer, you might already have a token
-    in your local cache. Check for a hidden `.huggingface` directory in your home folder. If it
-    contains a `token` file, then you are all set.
-
-
-Hugging Face TI concepts are downloaded and installed automatically as you
-require them. This requires your machine to be connected to the Internet. To
-find out what each concept is for, you can browse the
-[Hugging Face concepts library](https://huggingface.co/sd-concepts-library) and
-look at examples of what each concept produces.
-
-To load concepts, you will need to open the Web UI's configuration
-dialogue and activate "Show Textual Inversions from HF Concepts
-Library". This will then add a list of HF Concepts to the dropdown
-"Add Textual Inversion" menu. Select the concept(s) of your choice and
-they will be incorporated into the positive prompt. A few concepts are
-designed for the negative prompt, in which case you can add them to
-the negative prompt box by select the down arrow icon next to the
-textual inversion menu.
-
-There are nearly 1000 HF concepts, more than will fit into a menu. For
-this reason we only show the most popular concepts (those which have
-received 5 or more likes). If you wish to use a concept that is not on
-the list, you may simply type its name surrounded by brackets. For
-example, to load the concept named "xidiversity", add `<xidiversity>`
-to the positive or negative prompt text.
 
 ## Installing your Own TI Files
 
 You may install any number of `.pt` and `.bin` files simply by copying them into
-the `embeddings` directory of the InvokeAI runtime directory (usually `invokeai`
-in your home directory). You may create subdirectories in order to organize the
-files in any way you wish. Be careful not to overwrite one file with another.
+the `embedding` directory of the corresponding InvokeAI models directory (usually `invokeai`
+in your home directory). For example, you can simply move a Stable Diffusion 1.5 embedding file to
+the `sd-1/embedding` folder. Be careful not to overwrite one file with another.
 For example, TI files generated by the Hugging Face toolkit share the named
-`learned_embedding.bin`. You can use subdirectories to keep them distinct.
+`learned_embedding.bin`. You can rename these, or use subdirectories to keep them distinct.
 
-At startup time, InvokeAI will scan the `embeddings` directory and load any TI
-files it finds there. At startup you will see a message similar to this one:
+At startup time, InvokeAI will scan the various `embedding` directories and load any TI
+files it finds there for compatible models. At startup you will see a message similar to this one:
 
 ```bash
 >> Current embedding manager terms: <HOI4-Leader>, <princess-knight>
 ```
+To use these when generating, simply type the `<` key in your prompt to open the Textual Inversion WebUI and 
+select the embedding you'd like to use. This UI has type-ahead support, so you can easily find supported embeddings.
 
-The terms you can use will appear in the "Add Textual Inversion"
-dropdown menu above the HF Concepts.
+## Using LoRAs
 
-## Further Reading
+LoRA files are models that customize the output of Stable Diffusion image generation.
+Larger than embeddings, but much smaller than full models, they augment SD with improved
+understanding of subjects and artistic styles.
+
+Unlike TI files, LoRAs do not introduce novel vocabulary into the model's known tokens. Instead,
+LoRAs augment the model's weights that are applied to generate imagery. LoRAs may be supplied
+with a "trigger" word that they have been explicitly trained on, or may simply apply their 
+effect without being triggered.
+
+LoRAs are typically stored in .safetensors files, which are the most secure way to store and transmit
+these types of weights. You may install any number of `.safetensors` LoRA files simply by copying them into
+the `lora` directory of the corresponding InvokeAI models directory (usually `invokeai`
+in your home directory). For example, you can simply move a Stable Diffusion 1.5 LoRA file to
+the `sd-1/lora` folder.
+
+To use these when generating, open the LoRA menu item in the options panel, select the LoRAs you want to apply
+and ensure that they have the appropriate weight recommended by the model provider. Typically, most LoRAs perform best at a weight of .75-1. 
 
-Please see [the repository](https://github.com/rinongal/textual_inversion) and
-associated paper for details and limitations.

docs/features/CONFIGURATION.md

@@ -0,0 +1,287 @@
+---
+title: Configuration
+---
+
+# :material-tune-variant: InvokeAI Configuration
+
+## Intro
+
+InvokeAI has numerous runtime settings which can be used to adjust
+many aspects of its operations, including the location of files and
+directories, memory usage, and performance. These settings can be
+viewed and customized in several ways:
+
+1. By editing settings in the `invokeai.yaml` file.
+2. By setting environment variables.
+3. On the command-line, when InvokeAI is launched.
+
+In addition, the most commonly changed settings are accessible
+graphically via the `invokeai-configure` script.
+
+### How the Configuration System Works
+
+When InvokeAI is launched, the very first thing it needs to do is to
+find its "root" directory, which contains its configuration files,
+installed models, its database of images, and the folder(s) of
+generated images themselves. In this document, the root directory will
+be referred to as ROOT.
+
+#### Finding the Root Directory
+
+To find its root directory, InvokeAI uses the following recipe:
+
+1. It first looks for the argument `--root <path>` on the command line
+it was launched from, and uses the indicated path if present.
+
+2. Next it looks for the environment variable INVOKEAI_ROOT, and uses
+the directory path found there if present.
+
+3. If neither of these are present, then InvokeAI looks for the
+folder containing the `.venv` Python virtual environment directory for
+the currently active environment. This directory is checked for files
+expected inside the InvokeAI root before it is used.
+
+4. Finally, InvokeAI looks for a directory in the current user's home
+directory named `invokeai`.
+
+#### Reading the InvokeAI Configuration File
+
+Once the root directory has been located, InvokeAI looks for a file
+named `ROOT/invokeai.yaml`, and if present reads configuration values
+from it. The top of this file looks like this:
+
+```
+InvokeAI:
+  Web Server:
+    host: localhost
+    port: 9090
+    allow_origins: []
+    allow_credentials: true
+    allow_methods:
+    - '*'
+    allow_headers:
+    - '*'
+  Features:
+    esrgan: true
+    internet_available: true
+    log_tokenization: false
+    nsfw_checker: false
+    patchmatch: true
+    restore: true
+...
+```
+
+This lines in this file are used to establish default values for
+Invoke's settings. In the above fragment, the Web Server's listening
+port is set to 9090 by the `port` setting.
+
+You can edit this file with a text editor such as "Notepad" (do not
+use Word or any other word processor). When editing, be careful to
+maintain the indentation, and do not add extraneous text, as syntax
+errors will prevent InvokeAI from launching. A basic guide to the
+format of YAML files can be found
+[here](https://circleci.com/blog/what-is-yaml-a-beginner-s-guide/).
+
+You can fix a broken `invokeai.yaml` by deleting it and running the
+configuration script again -- option [7] in the launcher, "Re-run the
+configure script".
+
+#### Reading Environment Variables
+
+Next InvokeAI looks for defined environment variables in the format
+`INVOKEAI_<setting_name>`, for example `INVOKEAI_port`. Environment
+variable values take precedence over configuration file variables. On
+a Macintosh system, for example, you could change the port that the
+web server listens on by setting the environment variable this way:
+
+```
+export INVOKEAI_port=8000
+invokeai-web
+```
+
+Please check out these
+[Macintosh](https://phoenixnap.com/kb/set-environment-variable-mac)
+and
+[Windows](https://phoenixnap.com/kb/windows-set-environment-variable)
+guides for setting temporary and permanent environment variables.
+
+#### Reading the Command Line
+
+Lastly, InvokeAI takes settings from the command line, which override
+everything else. The command-line settings have the same name as the
+corresponding configuration file settings, preceded by a `--`, for
+example `--port 8000`.
+
+If you are using the launcher (`invoke.sh` or `invoke.bat`) to launch
+InvokeAI, then just pass the command-line arguments to the launcher:
+
+```
+invoke.bat --port 8000 --host 0.0.0.0
+```
+
+The arguments will be applied when you select the web server option
+(and the other options as well).
+
+If, on the other hand, you prefer to launch InvokeAI directly from the
+command line, you would first activate the virtual environment (known
+as the "developer's console" in the launcher), and run `invokeai-web`:
+
+```
+> C:\Users\Fred\invokeai\.venv\scripts\activate
+(.venv) > invokeai-web --port 8000 --host 0.0.0.0
+```
+
+You can get a listing and brief instructions for each of the
+command-line options by giving the `--help` argument:
+
+```
+(.venv) > invokeai-web --help
+usage: InvokeAI [-h] [--host HOST] [--port PORT] [--allow_origins [ALLOW_ORIGINS ...]] [--allow_credentials | --no-allow_credentials]
+                [--allow_methods [ALLOW_METHODS ...]] [--allow_headers [ALLOW_HEADERS ...]] [--esrgan | --no-esrgan]
+                [--internet_available | --no-internet_available] [--log_tokenization | --no-log_tokenization]
+                [--nsfw_checker | --no-nsfw_checker] [--patchmatch | --no-patchmatch] [--restore | --no-restore]
+                [--always_use_cpu | --no-always_use_cpu] [--free_gpu_mem | --no-free_gpu_mem] [--max_cache_size MAX_CACHE_SIZE]
+                [--max_vram_cache_size MAX_VRAM_CACHE_SIZE] [--precision {auto,float16,float32,autocast}]
+                [--sequential_guidance | --no-sequential_guidance] [--xformers_enabled | --no-xformers_enabled]
+                [--tiled_decode | --no-tiled_decode] [--root ROOT] [--autoimport_dir AUTOIMPORT_DIR] [--lora_dir LORA_DIR]
+                [--embedding_dir EMBEDDING_DIR] [--controlnet_dir CONTROLNET_DIR] [--conf_path CONF_PATH] [--models_dir MODELS_DIR]
+                [--legacy_conf_dir LEGACY_CONF_DIR] [--db_dir DB_DIR] [--outdir OUTDIR] [--from_file FROM_FILE]
+                [--use_memory_db | --no-use_memory_db] [--model MODEL] [--log_handlers [LOG_HANDLERS ...]]
+                [--log_format {plain,color,syslog,legacy}] [--log_level {debug,info,warning,error,critical}]
+...
+```
+
+## The Configuration Settings
+
+The configuration settings are divided into several distinct
+groups in `invokeia.yaml`:
+
+### Web Server
+
+| Setting  | Default Value  |  Description |
+|----------|----------------|--------------|
+| `host`     | `localhost`      | Name or IP address of the network interface that the web server will listen on  |
+| `port`     | `9090`           | Network port number that the web server will listen on  |
+| `allow_origins`  | `[]`       | A list of host names or IP addresses that are allowed to connect to the InvokeAI API in the format `['host1','host2',...]` |
+| `allow_credentials | `true`   | Require credentials for a foreign host to access the InvokeAI API (don't change this) |
+| `allow_methods` | `*`         | List of HTTP methods ("GET", "POST") that the web server is allowed to use when accessing the API  |
+| `allow_headers` | `*`         | List of HTTP headers that the web server will accept when accessing the API  |
+
+The documentation for InvokeAI's API can be accessed by browsing to the following URL: [http://localhost:9090/docs].
+
+### Features
+
+These configuration settings allow you to enable and disable various InvokeAI features:
+
+| Setting  | Default Value  |  Description |
+|----------|----------------|--------------|
+| `esrgan`     | `true`      | Activate the ESRGAN upscaling options|
+| `internet_available` | `true`     | When a resource is not available locally, try to fetch it via the internet |
+| `log_tokenization` | `false`      | Before each text2image generation, print a color-coded representation of the prompt to the console; this can help understand why a prompt is not working as expected |
+| `nsfw_checker` | `true`     | Activate the NSFW checker to blur out risque images |
+| `patchmatch` | `true`     | Activate the "patchmatch" algorithm for improved inpainting |
+| `restore`    | `true`     | Activate the facial restoration features (DEPRECATED; restoration features will be removed in 3.0.0) |
+
+### Memory/Performance
+
+These options tune InvokeAI's memory and performance characteristics.
+
+| Setting  | Default Value  |  Description |
+|----------|----------------|--------------|
+| `always_use_cpu`     | `false`      | Use the CPU to generate images, even if a GPU is available |
+| `free_gpu_mem`       | `false`      | Aggressively free up GPU memory after each operation; this will allow you to run in low-VRAM environments with some performance penalties |
+| `max_cache_size`       | `6`      | Amount of CPU RAM (in GB) to reserve for caching models in memory; more cache allows you to keep models in memory and switch among them quickly |
+| `max_vram_cache_size`  | `2.75`   | Amount of GPU VRAM (in GB) to reserve for caching models in VRAM; more cache speeds up generation but reduces the size of the images that can be generated. This can be set to zero to maximize the amount of memory available for generation. |
+| `precision`       | `auto`      | Floating point precision. One of `auto`, `float16` or `float32`. `float16` will consume half the memory of `float32` but produce slightly lower-quality images. The `auto` setting will guess the proper precision based on your video card and operating system |
+| `sequential_guidance`     | `false`      | Calculate guidance in serial rather than in parallel, lowering memory requirements at the cost of some performance loss |
+| `xformers_enabled`        | `true`      | If the x-formers memory-efficient attention module is installed, activate it for better memory usage and generation speed|
+| `tiled_decode`            | `false`     | If true, then during the VAE decoding phase the image will be decoded a section at a time, reducing memory consumption at the cost of a performance hit |
+
+### Paths
+
+These options set the paths of various directories and files used by
+InvokeAI. Relative paths are interpreted relative to INVOKEAI_ROOT, so
+if INVOKEAI_ROOT is `/home/fred/invokeai` and the path is
+`autoimport/main`, then the corresponding directory will be located at
+`/home/fred/invokeai/autoimport/main`.
+
+| Setting  | Default Value  |  Description |
+|----------|----------------|--------------|
+| `autoimport_dir` | `autoimport/main`     | At startup time, read and import any main model files found in this directory |
+| `lora_dir` | `autoimport/lora`     | At startup time, read and import any LoRA/LyCORIS models found in this directory |
+| `embedding_dir` | `autoimport/embedding`  | At startup time, read and import any textual inversion (embedding) models found in this directory |
+| `controlnet_dir` | `autoimport/controlnet`  | At startup time, read and import any ControlNet models found in this directory |
+| `conf_path` | `configs/models.yaml`  | Location of the `models.yaml` model configuration file |
+| `models_dir` | `models`  | Location of the directory containing models installed by InvokeAI's model manager |
+| `legacy_conf_dir` | `configs/stable-diffusion`  | Location of the directory containing the .yaml configuration files for legacy checkpoint models |
+| `db_dir` | `databases`  | Location of the directory containing InvokeAI's image, schema and session database |
+| `outdir` | `outputs`  | Location of the directory in which the gallery of generated and uploaded images will be stored |
+| `use_memory_db` | `false`  | Keep database information in memory rather than on disk; this will not preserve image gallery information across restarts |
+
+Note that the autoimport directories will be searched recursively,
+allowing you to organize the models into folders and subfolders in any
+way you wish. In addition, while we have split up autoimport
+directories by the type of model they contain, this isn't
+necessary. You can combine different model types in the same folder
+and InvokeAI will figure out what they are. So you can easily use just
+one autoimport directory by commenting out the unneeded paths:
+
+```
+Paths:
+  autoimport_dir: autoimport
+#  lora_dir: null
+#  embedding_dir: null
+#  controlnet_dir: null
+```
+
+### Logging
+
+These settings control the information, warning, and debugging
+messages printed to the console log while InvokeAI is running:
+
+| Setting  | Default Value  |  Description |
+|----------|----------------|--------------|
+| `log_handlers` | `console` | This controls where log messages are sent, and can be a list of one or more destinations. Values include `console`, `file`, `syslog` and `http`. These are described in more detail below |
+| `log_format` | `color` | This controls the formatting of the log messages. Values are `plain`, `color`, `legacy` and `syslog` |
+| `log_level`  | `debug` | This filters messages according to the level of severity and can be one of `debug`, `info`, `warning`, `error` and `critical`. For example, setting to `warning` will display all messages at the warning level or higher, but won't display "debug" or "info" messages |
+
+Several different log handler destinations are available, and multiple destinations are supported by providing a list:
+
+```
+  log_handlers:
+     - console
+     - syslog=localhost
+     - file=/var/log/invokeai.log
+```
+
+* `console` is the default. It prints log messages to the command-line window from which InvokeAI was launched.
+
+* `syslog` is only available on Linux and Macintosh systems. It uses
+  the operating system's "syslog" facility to write log file entries
+  locally or to a remote logging machine. `syslog` offers a variety
+  of configuration options:
+
+```
+  syslog=/dev/log`      - log to the /dev/log device
+  syslog=localhost`     - log to the network logger running on the local machine
+  syslog=localhost:512` - same as above, but using a non-standard port
+  syslog=fredserver,facility=LOG_USER,socktype=SOCK_DRAM`
+                        - Log to LAN-connected server "fredserver" using the facility LOG_USER and datagram packets.
+```
+
+* `http` can be used to log to a remote web server. The server must be
+  properly configured to receive and act on log messages. The option
+  accepts the URL to the web server, and a `method` argument
+  indicating whether the message should be submitted using the GET or
+  POST method.
+
+```
+ http=http://my.server/path/to/logger,method=POST
+```
+
+The `log_format` option provides several alternative formats:
+
+* `color`    - default format providing time, date and a message, using text colors to distinguish different log severities
+* `plain`    - same as above, but monochrome text only
+* `syslog`   - the log level and error message only, allowing the syslog system to attach the time and date
+* `legacy`   - a format similar to the one used by the legacy 2.3 InvokeAI releases.

docs/features/NODES.md

@@ -0,0 +1,206 @@
+# Nodes Editor (Experimental)
+
+🚨
+*The node editor is experimental. We've made it accessible because we use it to develop the application, but we have not addressed the many known rough edges. It's very easy to shoot yourself in the foot, and we cannot offer support for it until it sees full release (ETA v3.1). Everything is subject to change without warning.* 
+🚨
+
+The nodes editor is a blank canvas allowing for the use of individual functions and image transformations to control the image generation workflow. The node processing flow is usually done from left (inputs) to right (outputs), though linearity can become abstracted the more complex the node graph becomes. Nodes inputs and outputs are connected by dragging connectors from node to node.
+
+To better understand how nodes are used, think of how an electric power bar works. It takes in one input (electricity from a wall outlet) and passes it to multiple devices through multiple outputs. Similarly, a node could have multiple inputs and outputs functioning at the same (or different) time, but all node outputs pass information onward like a power bar passes electricity. Not all outputs are compatible with all inputs, however - Each node has different constraints on how it is expecting to input/output information. In general, node outputs are colour-coded to match compatible inputs of other nodes.
+
+## Anatomy of a Node
+
+Individual nodes are made up of the following:
+
+- Inputs: Edge points on the left side of the node window where you connect outputs from other nodes.
+- Outputs: Edge points on the right side of the node window where you connect to inputs on other nodes.
+- Options: Various options which are either manually configured, or overridden by connecting an output from another node to the input.
+
+## Diffusion Overview
+
+Taking the time to understand the diffusion process will help you to understand how to set up your nodes in the nodes editor. 
+
+There are two main spaces Stable Diffusion works in: image space and latent space.
+
+Image space represents images in pixel form that you look at. Latent space represents compressed inputs. It’s in latent space that Stable Diffusion processes images. A VAE (Variational Auto Encoder) is responsible for compressing and encoding inputs into latent space, as well as decoding outputs back into image space.
+
+When you generate an image using text-to-image, multiple steps occur in latent space:
+1. Random noise is generated at the chosen height and width. The noise’s characteristics are dictated by the chosen (or not chosen) seed. This noise tensor is passed into latent space. We’ll call this noise A.
+1. Using a model’s U-Net, a noise predictor examines noise A, and the words tokenized by CLIP from your prompt (conditioning). It generates its own noise tensor to predict what the final image might look like in latent space. We’ll call this noise B.
+1. Noise B is subtracted from noise A in an attempt to create a final latent image indicative of the inputs. This step is repeated for the number of sampler steps chosen.
+1. The VAE decodes the final latent image from latent space into image space.
+
+image-to-image is a similar process, with only step 1 being different:
+1. The input image is decoded from image space into latent space by the VAE. Noise is then added to the input latent image. Denoising Strength dictates how much noise is added, 0 being none, and 1 being all-encompassing. We’ll call this noise A. The process is then the same as steps 2-4 in the text-to-image explanation above. 
+
+Furthermore, a model provides the CLIP prompt tokenizer, the VAE, and a U-Net (where noise prediction occurs given a prompt and initial noise tensor).
+
+A noise scheduler (eg. DPM++ 2M Karras) schedules the subtraction of noise from the latent image across the sampler steps chosen (step 3 above). Less noise is usually subtracted at higher sampler steps. 
+
+## Node Types (Base Nodes)
+
+| Node <img width=160 align="right"> | Function                                                                              |
+| ---------------------------------- | --------------------------------------------------------------------------------------|
+| Add                                | Adds two numbers |
+| CannyImageProcessor                | Canny edge detection for ControlNet |
+| ClipSkip                           | Skip layers in clip text_encoder model |
+| Collect                            | Collects values into a collection |
+| Prompt (Compel)                    | Parse prompt using compel package to conditioning |
+| ContentShuffleImageProcessor       | Applies content shuffle processing to image |
+| ControlNet                         | Collects ControlNet info to pass to other nodes |
+| CvInpaint                          | Simple inpaint using opencv |
+| Divide                             | Divides two numbers |
+| DynamicPrompt                      | Parses a prompt using adieyal/dynamic prompt's random or combinatorial generator |
+| FloatLinearRange                   | Creates a range |
+| HedImageProcessor                  | Applies HED edge detection to image |
+| ImageBlur                          | Blurs an image |
+| ImageChannel                       | Gets a channel from an image |
+| ImageCollection                    | Load a collection of images and provide it as output |
+| ImageConvert                       | Converts an image to a different mode |
+| ImageCrop                          | Crops an image to a specified box. The box can be outside of the image. |
+| ImageInverseLerp                   | Inverse linear interpolation of all pixels of an image |
+| ImageLerp                          | Linear interpolation of all pixels of an image |
+| ImageMultiply                      | Multiplies two images together using `PIL.ImageChops.Multiply()` |
+| ImagePaste                         | Pastes an image into another image |
+| ImageProcessor                     | Base class for invocations that reprocess images for ControlNet |
+| ImageResize                        | Resizes an image to specific dimensions |
+| ImageScale                         | Scales an image by a factor |
+| ImageToLatents                     | Scales latents by a given factor |
+| InfillColor                        | Infills transparent areas of an image with a solid color |
+| InfillPatchMatch                   | Infills transparent areas of an image using the PatchMatch algorithm |
+| InfillTile                         | Infills transparent areas of an image with tiles of the image |
+| Inpaint                            | Generates an image using inpaint |
+| Iterate                            | Iterates over a list of items |
+| LatentsToImage                     | Generates an image from latents |
+| LatentsToLatents                   | Generates latents using latents as base image |
+| LeresImageProcessor                | Applies leres processing to image |
+| LineartAnimeImageProcessor         | Applies line art anime processing to image |
+| LineartImageProcessor              | Applies line art processing to image |
+| LoadImage                          | Load an image and provide it as output |
+| Lora Loader                        | Apply selected lora to unet and text_encoder |
+| Model Loader                       | Loads a main model, outputting its submodels |
+| MaskFromAlpha                      | Extracts the alpha channel of an image as a mask |
+| MediapipeFaceProcessor             | Applies mediapipe face processing to image |
+| MidasDepthImageProcessor           | Applies Midas depth processing to image |
+| MlsdImageProcessor                 | Applied MLSD processing to image |
+| Multiply                           | Multiplies two numbers |
+| Noise                              | Generates latent noise |
+| NormalbaeImageProcessor            | Applies NormalBAE processing to image |
+| OpenposeImageProcessor             | Applies Openpose processing to image |
+| ParamFloat                         | A float parameter |
+| ParamInt                           | An integer parameter |
+| PidiImageProcessor                 | Applies PIDI processing to an image |
+| Progress Image                     | Displays the progress image in the Node Editor |
+| RandomInit                         | Outputs a single random integer |
+| RandomRange                        | Creates a collection of random numbers |
+| Range                              | Creates a range of numbers from start to stop with step |
+| RangeOfSize                        | Creates a range from start to start + size with step |
+| ResizeLatents                      | Resizes latents to explicit width/height (in pixels). Provided dimensions are floor-divided by 8. |
+| RestoreFace                        | Restores faces in the image |
+| ScaleLatents                       | Scales latents by a given factor |
+| SegmentAnythingProcessor           | Applies segment anything processing to image |
+| ShowImage                          | Displays a provided image, and passes it forward in the pipeline |
+| StepParamEasing                    | Experimental per-step parameter for easing for denoising steps |
+| Subtract                           | Subtracts two numbers |
+| TextToLatents                      | Generates latents from conditionings |
+| TileResampleProcessor              | Bass class for invocations that preprocess images for ControlNet |
+| Upscale                            | Upscales an image |
+| VAE Loader                         | Loads a VAE model, outputting a VaeLoaderOutput |
+| ZoeDepthImageProcessor             | Applies Zoe depth processing to image |
+
+## Node Grouping Concepts
+
+There are several node grouping concepts that can be examined with a narrow focus. These (and other) groupings can be pieced together to make up functional graph setups, and are important to understanding how groups of nodes work together as part of a whole. Note that the screenshots below aren't examples of complete functioning node graphs (see Examples).
+
+### Noise
+
+As described, an initial noise tensor is necessary for the latent diffusion process. As a result, all non-image *ToLatents nodes require a noise node input.  
+
+<img width="654" alt="groupsnoise" src="https://github.com/ymgenesis/InvokeAI/assets/25252829/2e8d297e-ad55-4d27-bc93-c119dad2a2c5">
+
+### Conditioning
+
+As described, conditioning is necessary for the latent diffusion process, whether empty or not. As a result, all non-image *ToLatents nodes require positive and negative conditioning inputs. Conditioning is reliant on a CLIP tokenizer provided by the Model Loader node.
+
+<img width="1024" alt="groupsconditioning" src="https://github.com/ymgenesis/InvokeAI/assets/25252829/f8f7ad8a-8d9c-418e-b5ad-1437b774b27e">
+
+### Image Space & VAE
+
+The ImageToLatents node doesn't require a noise node input, but requires a VAE input to convert the image from image space into latent space. In reverse, the LatentsToImage node requires a VAE input to convert from latent space back into image space.
+
+<img width="637" alt="groupsimgvae" src="https://github.com/ymgenesis/InvokeAI/assets/25252829/dd99969c-e0a8-4f78-9b17-3ffe179cef9a">
+
+### Defined & Random Seeds
+
+It is common to want to use both the same seed (for continuity) and random seeds (for variance). To define a seed, simply enter it into the 'Seed' field on a noise node. Conversely, the RandomInt node generates a random integer between 'Low' and 'High', and can be used as input to the 'Seed' edge point on a noise node to randomize your seed.
+
+<img width="922" alt="groupsrandseed" src="https://github.com/ymgenesis/InvokeAI/assets/25252829/af55bc20-60f6-438e-aba5-3ec871443710">
+
+### Control
+
+Control means to guide the diffusion process to adhere to a defined input or structure. Control can be provided as input to non-image *ToLatents nodes from ControlNet nodes. ControlNet nodes usually require an image processor which converts an input image for use with ControlNet.
+
+<img width="805" alt="groupscontrol" src="https://github.com/ymgenesis/InvokeAI/assets/25252829/cc9c5de7-23a7-46c8-bbad-1f3609d999a6">
+
+### LoRA
+
+The Lora Loader node lets you load a LoRA (say that ten times fast) and pass it as output to both the Prompt (Compel) and non-image *ToLatents nodes. A model's CLIP tokenizer is passed through the LoRA into Prompt (Compel), where it affects conditioning. A model's U-Net is also passed through the LoRA into a non-image *ToLatents node, where it affects noise prediction.
+
+<img width="993" alt="groupslora" src="https://github.com/ymgenesis/InvokeAI/assets/25252829/630962b0-d914-4505-b3ea-ccae9b0269da">
+
+### Scaling
+
+Use the ImageScale, ScaleLatents, and Upscale nodes to upscale images and/or latent images. The chosen method differs across contexts. However, be aware that latents are already noisy and compressed at their original resolution; scaling an image could produce more detailed results.
+
+<img width="644" alt="groupsallscale" src="https://github.com/ymgenesis/InvokeAI/assets/25252829/99314f05-dd9f-4b6d-b378-31de55346a13">
+
+### Iteration + Multiple Images as Input
+
+Iteration is a common concept in any processing, and means to repeat a process with given input. In nodes, you're able to use the Iterate node to iterate through collections usually gathered by the Collect node. The Iterate node has many potential uses, from processing a collection of images one after another, to varying seeds across multiple image generations and more. This screenshot demonstrates how to collect several images and pass them out one at a time.
+
+<img width="788" alt="groupsiterate" src="https://github.com/ymgenesis/InvokeAI/assets/25252829/4af5ca27-82c9-4018-8c5b-024d3ee0a121">
+
+### Multiple Image Generation + Random Seeds
+
+Multiple image generation in the node editor is done using the RandomRange node. In this case, the 'Size' field represents the number of images to generate. As RandomRange produces a collection of integers, we need to add the Iterate node to iterate through the collection. 
+
+To control seeds across generations takes some care. The first row in the screenshot will generate multiple images with different seeds, but using the same RandomRange parameters across invocations will result in the same group of random seeds being used across the images, producing repeatable results. In the second row, adding the RandomInt node as input to RandomRange's 'Seed' edge point will ensure that seeds are varied across all images across invocations, producing varied results.
+
+<img width="1027" alt="groupsmultigenseeding" src="https://github.com/ymgenesis/InvokeAI/assets/25252829/518d1b2b-fed1-416b-a052-ab06552521b3">
+
+## Examples
+
+With our knowledge of node grouping and the diffusion process, let’s break down some basic graphs in the nodes editor. Note that a node's options can be overridden by inputs from other nodes. These examples aren't strict rules to follow and only demonstrate some basic configurations.
+
+### Basic text-to-image Node Graph
+
+<img width="875" alt="nodest2i" src="https://github.com/ymgenesis/InvokeAI/assets/25252829/17c67720-c376-4db8-94f0-5e00381a61ee">
+
+- Model Loader: A necessity to generating images (as we’ve read above). We choose our model from the dropdown. It outputs a U-Net, CLIP tokenizer, and VAE.
+- Prompt (Compel): Another necessity. Two prompt nodes are created. One will output positive conditioning (what you want, ‘dog’), one will output negative (what you don’t want, ‘cat’). They both input the CLIP tokenizer that the Model Loader node outputs.
+- Noise: Consider this noise A from step one of the text-to-image explanation above. Choose a seed number, width, and height.
+- TextToLatents: This node takes many inputs for converting and processing text & noise from image space into latent space, hence the name TextTo**Latents**. In this setup, it inputs positive and negative conditioning from the prompt nodes for processing (step 2 above). It inputs noise from the noise node for processing (steps 2 & 3 above). Lastly, it inputs a U-Net from the Model Loader node for processing (step 2 above). It outputs latents for use in the next LatentsToImage node. Choose number of sampler steps, CFG scale, and scheduler.
+- LatentsToImage: This node takes in processed latents from the TextToLatents node, and the model’s VAE from the Model Loader node which is responsible for decoding latents back into the image space, hence the name LatentsTo**Image**. This node is the last stop, and once the image is decoded, it is saved to the gallery.
+
+### Basic image-to-image Node Graph
+
+<img width="998" alt="nodesi2i" src="https://github.com/ymgenesis/InvokeAI/assets/25252829/3f2c95d5-cee7-4415-9b79-b46ee60a92fe">
+
+- Model Loader: Choose a model from the dropdown.
+- Prompt (Compel): Two prompt nodes. One positive (dog), one negative (dog). Same CLIP inputs from the Model Loader node as before.
+- ImageToLatents: Upload a source image directly in the node window, via drag'n'drop from the gallery, or passed in as input. The ImageToLatents node inputs the VAE from the Model Loader node to decode the chosen image from image space into latent space, hence the name ImageTo**Latents**. It outputs latents for use in the next LatentsToLatents node. It also outputs the source image's width and height for use in the next Noise node if the final image is to be the same dimensions as the source image.
+- Noise: A noise tensor is created with the width and height of the source image, and connected to the next LatentsToLatents node. Notice the width and height fields are overridden by the input from the ImageToLatents width and height outputs.
+- LatentsToLatents: The inputs and options are nearly identical to TextToLatents, except that LatentsToLatents also takes latents as an input. Considering our source image is already converted to latents in the last ImageToLatents node, and text + noise are no longer the only inputs to process, we use the LatentsToLatents node. 
+- LatentsToImage: Like previously, the LatentsToImage node will use the VAE from the Model Loader as input to decode the latents from LatentsToLatents into image space, and save it to the gallery.
+
+### Basic ControlNet Node Graph
+
+<img width="703" alt="nodescontrol" src="https://github.com/ymgenesis/InvokeAI/assets/25252829/b02ded86-ceb4-44a2-9910-e19ad184d471">
+
+- Model Loader
+- Prompt (Compel)
+- Noise: Width and height of the CannyImageProcessor ControlNet image is passed in to set the dimensions of the noise passed to TextToLatents.
+- CannyImageProcessor: The CannyImageProcessor node is used to process the source image being used as a ControlNet. Each ControlNet processor node applies control in different ways, and has some different options to configure. Width and height are passed to noise, as mentioned. The processed ControlNet image is output to the ControlNet node.
+- ControlNet: Select the type of control model. In this case, canny is chosen as the CannyImageProcessor was used to generate the ControlNet image. Configure the control node options, and pass the control output to TextToLatents.
+- TextToLatents: Similar to the basic text-to-image example, except ControlNet is passed to the control input edge point.
+- LatentsToImage

docs/features/PROMPTS.md

@@ -301,5 +301,48 @@ summoning up the concept of some sort of scifi creature? Let's find out.
 Indeed, removing the word "hybrid" produces an image that is more like what we'd
 expect.
 
-In conclusion, prompt blending is great for exploring creative space,
-but takes some trial and error to achieve the desired effect.
+## Dynamic Prompts
+
+Dynamic Prompts are a powerful feature designed to produce a variety of prompts based on user-defined options. Using a special syntax, you can construct a prompt with multiple possibilities, and the system will automatically generate a series of permutations based on your settings. This is extremely beneficial for ideation, exploring various scenarios, or testing different concepts swiftly and efficiently.
+
+### Structure of a Dynamic Prompt
+
+A Dynamic Prompt comprises of regular text, supplemented with alternatives enclosed within curly braces {} and separated by a vertical bar |. For example: {option1|option2|option3}. The system will then select one of the options to include in the final prompt. This flexible system allows for options to be placed throughout the text as needed.
+
+Furthermore, Dynamic Prompts can designate multiple selections from a single group of options. This feature is triggered by prefixing the options with a numerical value followed by $$. For example, in {2$$option1|option2|option3}, the system will select two distinct options from the set.
+### Creating Dynamic Prompts
+
+To create a Dynamic Prompt, follow these steps:
+
+    Draft your sentence or phrase, identifying words or phrases with multiple possible options.
+    Encapsulate the different options within curly braces {}.
+    Within the braces, separate each option using a vertical bar |.
+    If you want to include multiple options from a single group, prefix with the desired number and $$.
+
+For instance: A {house|apartment|lodge|cottage} in {summer|winter|autumn|spring} designed in {2$$style1|style2|style3}.
+### How Dynamic Prompts Work
+
+Once a Dynamic Prompt is configured, the system generates an array of combinations using the options provided. Each group of options in curly braces is treated independently, with the system selecting one option from each group. For a prefixed set (e.g., 2$$), the system will select two distinct options.
+
+For example, the following prompts could be generated from the above Dynamic Prompt:
+
+    A house in summer designed in style1, style2
+    A lodge in autumn designed in style3, style1
+    A cottage in winter designed in style2, style3
+    And many more!
+
+When the `Combinatorial` setting is on, Invoke will disable the "Images" selection, and generate every combination up until the setting for Max Prompts is reached.
+When the `Combinatorial` setting is off, Invoke will randomly generate combinations up until the setting for Images has been reached.
+
+
+
+### Tips and Tricks for Using Dynamic Prompts
+
+Below are some useful strategies for creating Dynamic Prompts:
+
+    Utilize Dynamic Prompts to generate a wide spectrum of prompts, perfect for brainstorming and exploring diverse ideas.
+    Ensure that the options within a group are contextually relevant to the part of the sentence where they are used. For instance, group building types together, and seasons together.
+    Apply the 2$$ prefix when you want to incorporate more than one option from a single group. This becomes quite handy when mixing and matching different elements.
+    Experiment with different quantities for the prefix. For example, 3$$ will select three distinct options.
+    Be aware of coherence in your prompts. Although the system can generate all possible combinations, not all may semantically make sense. Therefore, carefully choose the options for each group.
+    Always review and fine-tune the generated prompts as needed. While Dynamic Prompts can help you generate a multitude of combinations, the final polishing and refining remain in your hands.

docs/features/TRAINING.md

@@ -1,9 +1,10 @@
 ---
-title: Textual-Inversion
+title: Training
 ---
 
-# :material-file-document: Textual Inversion
+# :material-file-document: Training
 
+# Textual Inversion Training
 ## **Personalizing Text-to-Image Generation**
 
 You may personalize the generated images to provide your own styles or objects
@@ -258,16 +259,6 @@ invokeai-ti \
        --only_save_embeds
 ```
 
-## Using Embeddings
-
-After training completes, the resultant embeddings will be saved into your `$INVOKEAI_ROOT/embeddings/<trigger word>/learned_embeds.bin`.
-
-These will be automatically loaded when you start InvokeAI.
-
-Add the trigger word, surrounded by angle brackets, to use that embedding. For example, if your trigger word was `terence`, use `<terence>` in prompts. This is the same syntax used by the HuggingFace concepts library.
-
-**Note:** `.pt` embeddings do not require the angle brackets.
-
 ## Troubleshooting
 
 ### `Cannot load embedding for <trigger>. It was trained on a model with token dimension 1024, but the current model has token dimension 768`

docs/features/WEB.md

@@ -76,10 +76,10 @@ From top to bottom, these are:
    with outpainting,and modify interior portions of the image with
    inpainting, erase portions of a starting image and have the AI fill in
    the erased region from a text prompt.
-4. Workflow Management (not yet implemented) - this panel will allow you to create
+4. Node Editor - this panel allows you to create
    pipelines of common operations and combine them into workflows.
-5. Training (not yet implemented) - this panel will provide an interface to [textual
-   inversion training](TEXTUAL_INVERSION.md) and fine tuning.
+5. Model Manager - this panel allows you to import and configure new
+   models using URLs, local paths, or HuggingFace diffusers repo_ids.
 
 The inpainting, outpainting and postprocessing tabs are currently in
 development. However, limited versions of their features can already be accessed

docs/features/index.md

@@ -37,7 +37,7 @@ guide also covers optimizing models to load quickly.
 Teach an old model new tricks. Merge 2-3 models together to create a
 new model that combines characteristics of the originals.
 
-## * [Textual Inversion](TEXTUAL_INVERSION.md)
+## * [Textual Inversion](TRAINING.md)
 Personalize models by adding your own style or subjects.
 
 # Other Features

docs/index.md

@@ -146,13 +146,15 @@ This method is recommended for those familiar with running Docker containers
 - [Installing](installation/050_INSTALLING_MODELS.md)
 - [Model Merging](features/MODEL_MERGING.md)
 - [Style/Subject Concepts and Embeddings](features/CONCEPTS.md)
-- [Textual Inversion](features/TEXTUAL_INVERSION.md)
 - [Not Safe for Work (NSFW) Checker](features/NSFW.md)
 <!-- seperator -->
 ### Prompt Engineering
 - [Prompt Syntax](features/PROMPTS.md)
 - [Generating Variations](features/VARIATIONS.md)
 
+### InvokeAI Configuration
+- [Guide to InvokeAI Runtime Settings](features/CONFIGURATION.md)
+
 ## :octicons-log-16: Important Changes Since Version 2.3
 
 ### Nodes

docs/installation/010_INSTALL_AUTOMATED.md

@@ -354,8 +354,8 @@ experimental versions later.
 
 12. **InvokeAI Options**: You can launch InvokeAI with several different command-line arguments that
     customize its behavior. For example, you can change the location of the
-    image output directory, or select your favorite sampler. See the
-    [Command-Line Interface](../features/CLI.md) for a full list of the options.
+    image output directory or balance memory usage vs performance. See
+    [Configuration](../features/CONFIGURATION.md) for a full list of the options.
 
     - To set defaults that will take effect every time you launch InvokeAI,
       use a text editor (e.g. Notepad) to exit the file

docs/installation/020_INSTALL_MANUAL.md

@@ -256,7 +256,7 @@ manager, please follow these steps:
 
 10.  Render away!
 
-    Browse the [features](../features/CLI.md) section to learn about all the
+    Browse the [features](../features/index.md) section to learn about all the
     things you can do with InvokeAI.
 
 
@@ -270,7 +270,7 @@ manager, please follow these steps:
 
 12. Other scripts
 
-    The [Textual Inversion](../features/TEXTUAL_INVERSION.md) script can be launched with the command:
+    The [Textual Inversion](../features/TRAINING.md) script can be launched with the command:
 
     ```bash
     invokeai-ti --gui

docs/installation/050_INSTALLING_MODELS.md

@@ -43,24 +43,7 @@ InvokeAI comes with support for a good set of starter models. You'll
 find them listed in the master models file
 `configs/INITIAL_MODELS.yaml` in the InvokeAI root directory. The
 subset that are currently installed are found in
-`configs/models.yaml`. As of v2.3.1, the list of starter models is:
-
-|Model Name | HuggingFace Repo ID | Description | URL |
-|---------- | ---------- | ----------- | --- |
-|stable-diffusion-1.5|runwayml/stable-diffusion-v1-5|Stable Diffusion version 1.5 diffusers model (4.27 GB)|https://huggingface.co/runwayml/stable-diffusion-v1-5 |
-|sd-inpainting-1.5|runwayml/stable-diffusion-inpainting|RunwayML SD 1.5 model optimized for inpainting, diffusers version (4.27 GB)|https://huggingface.co/runwayml/stable-diffusion-inpainting |
-|stable-diffusion-2.1|stabilityai/stable-diffusion-2-1|Stable Diffusion version 2.1 diffusers model, trained on 768 pixel images (5.21 GB)|https://huggingface.co/stabilityai/stable-diffusion-2-1 |
-|sd-inpainting-2.0|stabilityai/stable-diffusion-2-inpainting|Stable Diffusion version 2.0 inpainting model (5.21 GB)|https://huggingface.co/stabilityai/stable-diffusion-2-inpainting |
-|analog-diffusion-1.0|wavymulder/Analog-Diffusion|An SD-1.5 model trained on diverse analog photographs (2.13 GB)|https://huggingface.co/wavymulder/Analog-Diffusion |
-|deliberate-1.0|XpucT/Deliberate|Versatile model that produces detailed images up to 768px (4.27 GB)|https://huggingface.co/XpucT/Deliberate |
-|d&d-diffusion-1.0|0xJustin/Dungeons-and-Diffusion|Dungeons & Dragons characters (2.13 GB)|https://huggingface.co/0xJustin/Dungeons-and-Diffusion |
-|dreamlike-photoreal-2.0|dreamlike-art/dreamlike-photoreal-2.0|A photorealistic model trained on 768 pixel images based on SD 1.5 (2.13 GB)|https://huggingface.co/dreamlike-art/dreamlike-photoreal-2.0 |
-|inkpunk-1.0|Envvi/Inkpunk-Diffusion|Stylized illustrations inspired by Gorillaz, FLCL and Shinkawa; prompt with "nvinkpunk" (4.27 GB)|https://huggingface.co/Envvi/Inkpunk-Diffusion |
-|openjourney-4.0|prompthero/openjourney|An SD 1.5 model fine tuned on Midjourney; prompt with "mdjrny-v4 style" (2.13 GB)|https://huggingface.co/prompthero/openjourney |
-|portrait-plus-1.0|wavymulder/portraitplus|An SD-1.5 model trained on close range portraits of people; prompt with "portrait+" (2.13 GB)|https://huggingface.co/wavymulder/portraitplus |
-|seek-art-mega-1.0|coreco/seek.art_MEGA|A general use SD-1.5 "anything" model that supports multiple styles (2.1 GB)|https://huggingface.co/coreco/seek.art_MEGA |
-|trinart-2.0|naclbit/trinart_stable_diffusion_v2|An SD-1.5 model finetuned with ~40K assorted high resolution manga/anime-style images (2.13 GB)|https://huggingface.co/naclbit/trinart_stable_diffusion_v2 |
-|waifu-diffusion-1.4|hakurei/waifu-diffusion|An SD-1.5 model trained on 680k anime/manga-style images (2.13 GB)|https://huggingface.co/hakurei/waifu-diffusion |
+`configs/models.yaml`.
 
 Note that these files are covered by an "Ethical AI" license which
 forbids certain uses. When you initially download them, you are asked
@@ -71,8 +54,7 @@ with the model terms by visiting the URLs in the table above.
 
 ## Community-Contributed Models
 
-There are too many to list here and more are being contributed every
-day.  [HuggingFace](https://huggingface.co/models?library=diffusers)
+[HuggingFace](https://huggingface.co/models?library=diffusers)
 is a great resource for diffusers models, and is also the home of a
 [fast-growing repository](https://huggingface.co/sd-concepts-library)
 of embedding (".bin") models that add subjects and/or styles to your
@@ -86,310 +68,106 @@ only `.safetensors` and `.ckpt` models, but they can be easily loaded
 into InvokeAI and/or converted into optimized `diffusers` models. Be
 aware that CIVITAI hosts many models that generate NSFW content.
 
-!!! note
-
-    InvokeAI 2.3.x does not support directly importing and
-    running Stable Diffusion version 2 checkpoint models. You may instead
-    convert them into `diffusers` models using the conversion methods
-    described below.
-
 ## Installation
 
-There are multiple ways to install and manage models:
+There are two ways to install and manage models:
 
-1. The `invokeai-configure` script which will download and install them for you.
+1. The `invokeai-model-install` script which will download and install
+them for you.  In addition to supporting main models, you can install
+ControlNet, LoRA and Textual Inversion models.
 
-2. The command-line tool (CLI) has commands that allows you to import, configure and modify
-   models files.
-
-3. The web interface (WebUI) has a GUI for importing and managing
+2. The web interface (WebUI) has a GUI for importing and managing
    models.
 
-### Installation via `invokeai-configure`
+3. By placing models (or symbolic links to models) inside one of the
+InvokeAI root directory's `autoimport` folder.
 
-From the `invoke` launcher, choose option (6) "re-run the configure
-script to download new models." This will launch the same script that
-prompted you to select models at install time. You can use this to add
-models that you skipped the first time around. It is all right to
-specify a model that was previously downloaded; the script will just
-confirm that the files are complete.
+### Installation via `invokeai-model-install`
 
-### Installation via the CLI
+From the `invoke` launcher, choose option [5] "Download and install
+models." This will launch the same script that prompted you to select
+models at install time. You can use this to add models that you
+skipped the first time around. It is all right to specify a model that
+was previously downloaded; the script will just confirm that the files
+are complete.
 
-You can install a new model, including any of the community-supported ones, via
-the command-line client's `!import_model` command.
+The installer has different panels for installing main models from
+HuggingFace, models from Civitai and other arbitrary web sites,
+ControlNet models, LoRA/LyCORIS models, and Textual Inversion
+embeddings. Each section has a text box in which you can enter a new
+model to install. You can refer to a model using its:
 
-#### Installing individual `.ckpt` and `.safetensors` models
+1. Local path to the .ckpt, .safetensors or diffusers folder on your local machine
+2. A directory on your machine that contains multiple models
+3. A URL that points to a downloadable model
+4. A HuggingFace repo id
 
-If the model is already downloaded to your local disk, use
-`!import_model /path/to/file.ckpt` to load it. For example:
+Previously-installed models are shown with checkboxes. Uncheck a box
+to unregister the model from InvokeAI. Models that are physically
+installed inside the InvokeAI root directory will be deleted and
+purged (after a confirmation warning). Models that are located outside
+the InvokeAI root directory will be unregistered but not deleted.
 
-```bash
-invoke> !import_model C:/Users/fred/Downloads/martians.safetensors
+Note: The installer script uses a console-based text interface that requires
+significant amounts of horizontal and vertical space. If the display
+looks messed up, just enlarge the terminal window and/or relaunch the
+script.
+
+If you wish you can script model addition and deletion, as well as
+listing installed models. Start the "developer's console" and give the
+command `invokeai-model-install --help`. This will give you a series
+of command-line parameters that will let you control model
+installation. Examples:
+
+```
+# (list all controlnet models)
+invokeai-model-install --list controlnet
+
+# (install the model at the indicated URL)
+invokeai-model-install --add http://civitai.com/2860
+
+# (delete the named model)
+invokeai-model-install --delete sd-1/main/analog-diffusion
 ```
 
-!!! tip "Forward Slashes"
-    On Windows systems, use forward slashes rather than backslashes
-    in your file paths.
-    If you do use backslashes,
-    you must double them like this:
-    `C:\\Users\\fred\\Downloads\\martians.safetensors`
+### Installation via the Web GUI
 
-Alternatively you can directly import the file using its URL:
+To install a new model using the Web GUI, do the following:
 
-```bash
-invoke> !import_model https://example.org/sd_models/martians.safetensors
-```
+1. Open the InvokeAI Model Manager (cube at the bottom of the
+left-hand panel) and navigate to *Import Models*
 
-For this to work, the URL must not be password-protected. Otherwise
-you will receive a 404 error.
+2. In the field labeled *Location* type in the path to the model you
+wish to install. You may use a URL, HuggingFace repo id, or a path on
+your local disk.
 
-When you import a legacy model, the CLI will first ask you what type
-of model this is. You can indicate whether it is a model based on
-Stable Diffusion 1.x (1.4 or 1.5), one based on Stable Diffusion 2.x,
-or a 1.x inpainting model. Be careful to indicate the correct model
-type, or it will not load correctly. You can correct the model type
-after the fact using the `!edit_model` command.
+3. Alternatively, the *Scan for Models* button allows you to paste in
+the path to a folder somewhere on your machine. It will be scanned for
+importable models and prompt you to add the ones of your choice.
 
-The system will then ask you a few other questions about the model,
-including what size image it was trained on (usually 512x512), what
-name and description you wish to use for it, and whether you would
-like to install a custom VAE (variable autoencoder) file for the
-model. For recent models, the answer to the VAE question is usually
-"no," but it won't hurt to answer "yes".
+4. Press *Add Model* and wait for confirmation that the model
+was added.
 
-After importing, the model will load. If this is successful, you will
-be asked if you want to keep the model loaded in memory to start
-generating immediately. You'll also be asked if you wish to make this
-the default model on startup. You can change this later using
-`!edit_model`.
+To delete a model, Select *Model Manager* to list all the currently
+installed models. Press the trash can icons to delete any models you
+wish to get rid of. Models whose weights are located inside the
+InvokeAI `models` directory will be purged from disk, while those
+located outside will be unregistered from InvokeAI, but not deleted.
 
-#### Importing a batch of `.ckpt` and `.safetensors` models from a directory
+You can see where model weights are located by clicking on the model name.
+This will bring up an editable info panel showing the model's characteristics,
+including the `Model Location` of its files.
 
-You may also point `!import_model` to a directory containing a set of
-`.ckpt` or `.safetensors` files. They will be imported _en masse_.
+### Installation via the `autoimport` function
 
-!!! example
+In the InvokeAI root directory you will find a series of folders under
+`autoimport`, one each for main models, controlnets, embeddings and
+Loras.  Any models that you add to these directories will be scanned
+at startup time and registered automatically.
 
-    ```console
-    invoke> !import_model C:/Users/fred/Downloads/civitai_models/
-    ```
+You may create symbolic links from these folders to models located
+elsewhere on disk and they will be autoimported. You can also create
+subfolders and organize them as you wish.
 
-You will be given the option to import all models found in the
-directory, or select which ones to import. If there are subfolders
-within the directory, they will be searched for models to import.
-
-#### Installing `diffusers` models
-
-You can install a `diffusers` model from the HuggingFace site using
-`!import_model` and the HuggingFace repo_id for the model:
-
-```bash
-invoke> !import_model andite/anything-v4.0
-```
-
-Alternatively, you can download the model to disk and import it from
-there. The model may be distributed as a ZIP file, or as a Git
-repository:
-
-```bash
-invoke> !import_model C:/Users/fred/Downloads/andite--anything-v4.0
-```
-
-!!! tip "The CLI supports file path autocompletion"
-         Type a bit of the path name and hit ++tab++ in order to get a choice of
-         possible completions.
-
-!!! tip "On Windows, you can drag model files onto the command-line"
-         Once you have typed in `!import_model `, you can drag the
-         model  file or directory onto the command-line to insert the model path. This way, you don't need to
-         type it or copy/paste. However, you will need to reverse or
-         double backslashes as noted above.
-
-Before installing, the CLI will ask you for a short name and
-description for the model, whether to make this the default model that
-is loaded at InvokeAI startup time, and whether to replace its
-VAE. Generally the answer to the latter question is "no".
-
-### Converting legacy models into `diffusers`
-
-The CLI `!convert_model` will convert a `.safetensors` or `.ckpt`
-models file into `diffusers` and install it.This will enable the model
-to load and run faster without loss of image quality.
-
-The usage is identical to `!import_model`. You may point the command
-to either a downloaded model file on disk, or to a (non-password
-protected) URL:
-
-```bash
-invoke> !convert_model C:/Users/fred/Downloads/martians.safetensors
-```
-
-After a successful conversion, the CLI will offer you the option of
-deleting the original `.ckpt` or `.safetensors` file.
-
-### Optimizing a previously-installed model
-
-Lastly, if you have previously installed a `.ckpt` or `.safetensors`
-file and wish to convert it into a `diffusers` model, you can do this
-without re-downloading and converting the original file using the
-`!optimize_model` command. Simply pass the short name of an existing
-installed model:
-
-```bash
-invoke> !optimize_model martians-v1.0
-```
-
-The model will be converted into `diffusers` format and replace the
-previously installed version. You will again be offered the
-opportunity to delete the original `.ckpt` or `.safetensors` file.
-
-### Related CLI Commands
-
-There are a whole series of additional model management commands in
-the CLI that you can read about in [Command-Line
-Interface](../features/CLI.md). These include:
-
-* `!models` - List all installed models
-* `!switch <model name>` - Switch to the indicated model
-* `!edit_model <model name>` - Edit the indicated model to change its name, description or other properties
-* `!del_model <model name>` - Delete the indicated model
-
-### Manually editing `configs/models.yaml`
-
-
-If you are comfortable with a text editor then you may simply edit `models.yaml`
-directly.
-
-You will need to download the desired `.ckpt/.safetensors` file and
-place it somewhere on your machine's filesystem. Alternatively, for a
-`diffusers` model, record the repo_id or download the whole model
-directory. Then using a **text** editor (e.g. the Windows Notepad
-application), open the file `configs/models.yaml`, and add a new
-stanza that follows this model:
-
-#### A legacy model
-
-A legacy `.ckpt` or `.safetensors` entry will look like this:
-
-```yaml
-arabian-nights-1.0:
-  description: A great fine-tune in Arabian Nights style
-  weights: ./path/to/arabian-nights-1.0.ckpt
-  config: ./configs/stable-diffusion/v1-inference.yaml
-  format: ckpt
-  width: 512
-  height: 512
-  default: false
-```
-
-Note that `format` is `ckpt` for both `.ckpt` and `.safetensors` files.
-
-#### A diffusers model
-
-A stanza for a `diffusers` model will look like this for a HuggingFace
-model with a repository ID:
-
-```yaml
-arabian-nights-1.1:
-  description: An even better fine-tune of the Arabian Nights
-  repo_id: captahab/arabian-nights-1.1
-  format: diffusers
-  default: true
-```
-
-And for a downloaded directory:
-
-```yaml
-arabian-nights-1.1:
-  description: An even better fine-tune of the Arabian Nights
-  path: /path/to/captahab-arabian-nights-1.1
-  format: diffusers
-  default: true
-```
-
-There is additional syntax for indicating an external VAE to use with
-this model. See `INITIAL_MODELS.yaml` and `models.yaml` for examples.
-
-After you save the modified `models.yaml` file relaunch
-`invokeai`. The new model will now be available for your use.
-
-### Installation via the WebUI
-
-To access the WebUI Model Manager, click on the button that looks like
-a cube in the upper right side of the browser screen. This will bring
-up a dialogue that lists the models you have already installed, and
-allows you to load, delete or edit them:
-
-<figure markdown>
-
-![model-manager](../assets/installing-models/webui-models-1.png)
-
-</figure>
-
-To add a new model, click on **+ Add New** and select to either a
-checkpoint/safetensors model, or a diffusers model:
-
-<figure markdown>
-
-![model-manager-add-new](../assets/installing-models/webui-models-2.png)
-
-</figure>
-
-In this example, we chose **Add Diffusers**. As shown in the figure
-below, a new dialogue prompts you to enter the name to use for the
-model, its description, and either the location of the `diffusers`
-model on disk, or its Repo ID on the HuggingFace web site. If you
-choose to enter a path to disk, the system will autocomplete for you
-as you type:
-
-<figure markdown>
-
-![model-manager-add-diffusers](../assets/installing-models/webui-models-3.png)
-
-</figure>
-
-Press **Add Model** at the bottom of the dialogue (scrolled out of
-site in the figure), and the model will be downloaded, imported, and
-registered in `models.yaml`.
-
-The **Add Checkpoint/Safetensor Model** option is similar, except that
-in this case you can choose to scan an entire folder for
-checkpoint/safetensors files to import. Simply type in the path of the
-directory and press the "Search" icon. This will display the
-`.ckpt` and `.safetensors` found inside the directory and its
-subfolders, and allow you to choose which ones to import:
-
-<figure markdown>
-
-![model-manager-add-checkpoint](../assets/installing-models/webui-models-4.png)
-
-</figure>
-
-## Model Management Startup Options
-
-The `invoke` launcher and the `invokeai` script accept a series of
-command-line arguments that modify InvokeAI's behavior when loading
-models. These can be provided on the command line, or added to the
-InvokeAI root directory's `invokeai.init` initialization file.
-
-The arguments are:
-
-* `--model <model name>` -- Start up with the indicated model loaded
-* `--ckpt_convert` -- When a checkpoint/safetensors model is loaded, convert it into a `diffusers` model in memory. This does not permanently save the converted model to disk.
-* `--autoconvert <path/to/directory>` -- Scan the indicated directory path for new checkpoint/safetensors files, convert them into `diffusers` models, and import them into InvokeAI.
-
-Here is an example of providing an argument on the command line using
-the `invoke.sh` launch script:
-
-```bash
-invoke.sh --autoconvert /home/fred/stable-diffusion-checkpoints
-```
-
-And here is what the same argument looks like in `invokeai.init`:
-
-```bash
---outdir="/home/fred/invokeai/outputs
---no-nsfw_checker
---autoconvert /home/fred/stable-diffusion-checkpoints
-```
+The location of the autoimport directories are controlled by settings
+in `invokeai.yaml`. See [Configuration](../features/CONFIGURATION.md).

installer/create_installer.sh

@@ -24,7 +24,8 @@ read -e -p "Tag this repo with '${VERSION}' and '${LATEST_TAG}'? [n]: " input
 RESPONSE=${input:='n'}
 if [ "$RESPONSE" == 'y' ]; then
 
-    if ! git tag $VERSION ; then
+    git push origin :refs/tags/$VERSION
+    if ! git tag -fa $VERSION ; then
 	    echo "Existing/invalid tag"
 	    exit -1
     fi

installer/install.bat.in

@@ -38,7 +38,7 @@ echo    https://learn.microsoft.com/en-US/cpp/windows/latest-supported-vc-redist
 echo.
 echo See %INSTRUCTIONS% for more details.
 echo.
-echo "For the best user experience we suggest enlarging or maximizing this window now."
+echo FOR THE BEST USER EXPERIENCE WE SUGGEST MAXIMIZING THIS WINDOW NOW.
 pause
 
 @rem ---------------------------- check Python version ---------------

installer/lib/installer.py

@@ -248,6 +248,7 @@ class InvokeAiInstance:
                 "install",
                 "--require-virtualenv",
                 "torch~=2.0.0",
+                "torchmetrics==0.11.4",
                 "torchvision>=0.14.1",
                 "--force-reinstall",
                 "--find-links" if find_links is not None else None,

installer/templates/invoke.bat.in

@@ -19,8 +19,8 @@ echo 8. Open the developer console
 echo 9. Update InvokeAI
 echo 10. Command-line help
 echo Q - Quit
-set /P choice="Please enter 1-10, Q: [2] "
-if not defined choice set choice=2
+set /P choice="Please enter 1-10, Q: [1] "
+if not defined choice set choice=1
 IF /I "%choice%" == "1" (
     echo Starting the InvokeAI browser-based UI..
     python .venv\Scripts\invokeai-web.exe %*
@@ -56,7 +56,7 @@ IF /I "%choice%" == "1" (
     call cmd /k
 ) ELSE IF /I "%choice%" == "9" (
    echo Running invokeai-update...
-   python .venv\Scripts\invokeai-update.exe %*
+   python -m invokeai.frontend.install.invokeai_update
 ) ELSE IF /I "%choice%" == "10" (
     echo Displaying command line help...
     python .venv\Scripts\invokeai.exe --help %*

installer/templates/invoke.sh.in

@@ -93,7 +93,7 @@ do_choice() {
     9)
         clear
         printf "Update InvokeAI\n"
-        invokeai-update
+        python -m invokeai.frontend.install.invokeai_update
         ;;
     10)
         clear

invokeai/app/api/dependencies.py

@@ -11,16 +11,16 @@ from invokeai.app.services.board_images import (
 )
 from invokeai.app.services.board_record_storage import SqliteBoardRecordStorage
 from invokeai.app.services.boards import BoardService, BoardServiceDependencies
+from invokeai.app.services.config import InvokeAIAppConfig
 from invokeai.app.services.image_record_storage import SqliteImageRecordStorage
 from invokeai.app.services.images import ImageService, ImageServiceDependencies
-from invokeai.app.services.metadata import CoreMetadataService
 from invokeai.app.services.resource_name import SimpleNameService
 from invokeai.app.services.urls import LocalUrlService
 from invokeai.backend.util.logging import InvokeAILogger
+from invokeai.version.invokeai_version import __version__
 
 from ..services.default_graphs import create_system_graphs
 from ..services.latent_storage import DiskLatentsStorage, ForwardCacheLatentsStorage
-from ..services.restoration_services import RestorationServices
 from ..services.graph import GraphExecutionState, LibraryGraph
 from ..services.image_file_storage import DiskImageFileStorage
 from ..services.invocation_queue import MemoryInvocationQueue
@@ -57,8 +57,9 @@ class ApiDependencies:
     invoker: Invoker = None
 
     @staticmethod
-    def initialize(config, event_handler_id: int, logger: Logger = logger):
-        logger.info(f"Internet connectivity is {config.internet_available}")
+    def initialize(config: InvokeAIAppConfig, event_handler_id: int, logger: Logger = logger):
+        logger.debug(f"InvokeAI version {__version__}")
+        logger.debug(f"Internet connectivity is {config.internet_available}")
 
         events = FastAPIEventService(event_handler_id)
 
@@ -73,7 +74,6 @@ class ApiDependencies:
         )
 
         urls = LocalUrlService()
-        metadata = CoreMetadataService()
         image_record_storage = SqliteImageRecordStorage(db_location)
         image_file_storage = DiskImageFileStorage(f"{output_folder}/images")
         names = SimpleNameService()
@@ -109,7 +109,6 @@ class ApiDependencies:
                 board_image_record_storage=board_image_record_storage,
                 image_record_storage=image_record_storage,
                 image_file_storage=image_file_storage,
-                metadata=metadata,
                 url=urls,
                 logger=logger,
                 names=names,
@@ -118,7 +117,7 @@ class ApiDependencies:
         )
 
         services = InvocationServices(
-            model_manager=ModelManagerService(config,logger),
+            model_manager=ModelManagerService(config, logger),
             events=events,
             latents=latents,
             images=images,
@@ -130,7 +129,6 @@ class ApiDependencies:
             ),
             graph_execution_manager=graph_execution_manager,
             processor=DefaultInvocationProcessor(),
-            restoration=RestorationServices(config, logger),
             configuration=config,
             logger=logger,
         )

invokeai/app/api/routers/app_info.py

@@ -1,18 +1,36 @@
 from fastapi.routing import APIRouter
-from pydantic import BaseModel
+from pydantic import BaseModel, Field
 
+from invokeai.backend.image_util.patchmatch import PatchMatch
 from invokeai.version import __version__
 
-app_router = APIRouter(prefix="/v1/app", tags=['app'])
+app_router = APIRouter(prefix="/v1/app", tags=["app"])
 
 
 class AppVersion(BaseModel):
     """App Version Response"""
-    version: str
+
+    version: str = Field(description="App version")
 
 
-@app_router.get('/version', operation_id="app_version",
-                status_code=200,
-                response_model=AppVersion)
+class AppConfig(BaseModel):
+    """App Config Response"""
+
+    infill_methods: list[str] = Field(description="List of available infill methods")
+
+
+@app_router.get(
+    "/version", operation_id="app_version", status_code=200, response_model=AppVersion
+)
 async def get_version() -> AppVersion:
     return AppVersion(version=__version__)
+
+
+@app_router.get(
+    "/config", operation_id="get_config", status_code=200, response_model=AppConfig
+)
+async def get_config() -> AppConfig:
+    infill_methods = ['tile']
+    if PatchMatch.patchmatch_available():
+        infill_methods.append('patchmatch')
+    return AppConfig(infill_methods=infill_methods)

invokeai/app/api/routers/images.py

@@ -1,25 +1,27 @@
 import io
 from typing import Optional
-from fastapi import Body, HTTPException, Path, Query, Request, Response, UploadFile
-from fastapi.routing import APIRouter
+
+from fastapi import (Body, HTTPException, Path, Query, Request, Response,
+                     UploadFile)
 from fastapi.responses import FileResponse
+from fastapi.routing import APIRouter
 from PIL import Image
-from invokeai.app.models.image import (
-    ImageCategory,
-    ResourceOrigin,
-)
+
+from invokeai.app.invocations.metadata import ImageMetadata
+from invokeai.app.models.image import ImageCategory, ResourceOrigin
 from invokeai.app.services.image_record_storage import OffsetPaginatedResults
-from invokeai.app.services.models.image_record import (
-    ImageDTO,
-    ImageRecordChanges,
-    ImageUrlsDTO,
-)
 from invokeai.app.services.item_storage import PaginatedResults
+from invokeai.app.services.models.image_record import (ImageDTO,
+                                                       ImageRecordChanges,
+                                                       ImageUrlsDTO)
 
 from ..dependencies import ApiDependencies
 
 images_router = APIRouter(prefix="/v1/images", tags=["images"])
 
+# images are immutable; set a high max-age
+IMAGE_MAX_AGE = 31536000
+
 
 @images_router.post(
     "/",
@@ -103,23 +105,38 @@ async def update_image(
 
 
 @images_router.get(
-    "/{image_name}/metadata",
-    operation_id="get_image_metadata",
+    "/{image_name}",
+    operation_id="get_image_dto",
     response_model=ImageDTO,
 )
-async def get_image_metadata(
+async def get_image_dto(
     image_name: str = Path(description="The name of image to get"),
 ) -> ImageDTO:
-    """Gets an image's metadata"""
+    """Gets an image's DTO"""
 
     try:
         return ApiDependencies.invoker.services.images.get_dto(image_name)
     except Exception as e:
         raise HTTPException(status_code=404)
 
+@images_router.get(
+    "/{image_name}/metadata",
+    operation_id="get_image_metadata",
+    response_model=ImageMetadata,
+)
+async def get_image_metadata(
+    image_name: str = Path(description="The name of image to get"),
+) -> ImageMetadata:
+    """Gets an image's metadata"""
+
+    try:
+        return ApiDependencies.invoker.services.images.get_metadata(image_name)
+    except Exception as e:
+        raise HTTPException(status_code=404)
+
 
 @images_router.get(
-    "/{image_name}",
+    "/{image_name}/full",
     operation_id="get_image_full",
     response_class=Response,
     responses={
@@ -141,12 +158,14 @@ async def get_image_full(
         if not ApiDependencies.invoker.services.images.validate_path(path):
             raise HTTPException(status_code=404)
 
-        return FileResponse(
+        response = FileResponse(
             path,
             media_type="image/png",
             filename=image_name,
             content_disposition_type="inline",
         )
+        response.headers["Cache-Control"] = f"max-age={IMAGE_MAX_AGE}"
+        return response
     except Exception as e:
         raise HTTPException(status_code=404)
 
@@ -175,9 +194,11 @@ async def get_image_thumbnail(
         if not ApiDependencies.invoker.services.images.validate_path(path):
             raise HTTPException(status_code=404)
 
-        return FileResponse(
+        response = FileResponse(
             path, media_type="image/webp", content_disposition_type="inline"
         )
+        response.headers["Cache-Control"] = f"max-age={IMAGE_MAX_AGE}"
+        return response
     except Exception as e:
         raise HTTPException(status_code=404)
 
@@ -208,10 +229,10 @@ async def get_image_urls(
 
 @images_router.get(
     "/",
-    operation_id="list_images_with_metadata",
+    operation_id="list_image_dtos",
     response_model=OffsetPaginatedResults[ImageDTO],
 )
-async def list_images_with_metadata(
+async def list_image_dtos(
     image_origin: Optional[ResourceOrigin] = Query(
         default=None, description="The origin of images to list"
     ),
@@ -227,7 +248,7 @@ async def list_images_with_metadata(
     offset: int = Query(default=0, description="The page offset"),
     limit: int = Query(default=10, description="The number of images per page"),
 ) -> OffsetPaginatedResults[ImageDTO]:
-    """Gets a list of images"""
+    """Gets a list of image DTOs"""
 
     image_dtos = ApiDependencies.invoker.services.images.get_many(
         offset,

invokeai/app/api/routers/models.py

@@ -1,6 +1,7 @@
-# Copyright (c) 2023 Kyle Schouviller (https://github.com/kyle0654), 2023 Kent Keirsey (https://github.com/hipsterusername), 2024 Lincoln Stein
+# Copyright (c) 2023 Kyle Schouviller (https://github.com/kyle0654), 2023 Kent Keirsey (https://github.com/hipsterusername), 2023 Lincoln D. Stein
 
 
+import pathlib
 from typing import Literal, List, Optional, Union
 
 from fastapi import Body, Path, Query, Response
@@ -12,8 +13,11 @@ from invokeai.backend import BaseModelType, ModelType
 from invokeai.backend.model_management.models import (
     OPENAPI_MODEL_CONFIGS,
     SchedulerPredictionType,
+    ModelNotFoundException,
+    InvalidModelException,
 )
 from invokeai.backend.model_management import MergeInterpolationMethod
+
 from ..dependencies import ApiDependencies
 
 models_router = APIRouter(prefix="/v1/models", tags=["models"])
@@ -22,6 +26,7 @@ UpdateModelResponse = Union[tuple(OPENAPI_MODEL_CONFIGS)]
 ImportModelResponse = Union[tuple(OPENAPI_MODEL_CONFIGS)]
 ConvertModelResponse = Union[tuple(OPENAPI_MODEL_CONFIGS)]
 MergeModelResponse = Union[tuple(OPENAPI_MODEL_CONFIGS)]
+ImportModelAttributes = Union[tuple(OPENAPI_MODEL_CONFIGS)]
 
 class ModelsList(BaseModel):
     models: list[Union[tuple(OPENAPI_MODEL_CONFIGS)]]
@@ -32,11 +37,16 @@ class ModelsList(BaseModel):
     responses={200: {"model": ModelsList }},
 )
 async def list_models(
-    base_model: Optional[BaseModelType] = Query(default=None, description="Base model"),
+    base_models: Optional[List[BaseModelType]] = Query(default=None, description="Base models to include"),
     model_type: Optional[ModelType] = Query(default=None, description="The type of model to get"),
 ) -> ModelsList:
     """Gets a list of models"""
-    models_raw = ApiDependencies.invoker.services.model_manager.list_models(base_model, model_type)
+    if base_models and len(base_models)>0:
+        models_raw = list()
+        for base_model in base_models:
+            models_raw.extend(ApiDependencies.invoker.services.model_manager.list_models(base_model, model_type))
+    else:
+        models_raw = ApiDependencies.invoker.services.model_manager.list_models(None, model_type)
     models = parse_obj_as(ModelsList, { "models": models_raw })
     return models
 
@@ -44,8 +54,9 @@ async def list_models(
     "/{base_model}/{model_type}/{model_name}",
     operation_id="update_model",
     responses={200: {"description" : "The model was updated successfully"},
+               400: {"description" : "Bad request"},
                404: {"description" : "The model could not be found"},
-               400: {"description" : "Bad request"}
+               409: {"description" : "There is already a model corresponding to the new name"},
                },
     status_code = 200,
     response_model = UpdateModelResponse,
@@ -56,33 +67,69 @@ async def update_model(
         model_name: str = Path(description="model name"),
         info: Union[tuple(OPENAPI_MODEL_CONFIGS)] = Body(description="Model configuration"),
 ) -> UpdateModelResponse:
-    """ Add Model """
+    """ Update model contents with a new config. If the model name or base fields are changed, then the model is renamed. """
+    logger = ApiDependencies.invoker.services.logger
+
+    
     try:
+        previous_info = ApiDependencies.invoker.services.model_manager.list_model(
+            model_name=model_name,
+            base_model=base_model,
+            model_type=model_type,
+        )
+
+        # rename operation requested
+        if info.model_name != model_name or info.base_model != base_model:
+            ApiDependencies.invoker.services.model_manager.rename_model(
+                base_model = base_model,
+                model_type = model_type,
+                model_name = model_name,
+                new_name = info.model_name,
+                new_base = info.base_model,
+            )
+            logger.info(f'Successfully renamed {base_model}/{model_name}=>{info.base_model}/{info.model_name}')
+            # update information to support an update of attributes
+            model_name = info.model_name
+            base_model = info.base_model
+            new_info = ApiDependencies.invoker.services.model_manager.list_model(
+                model_name=model_name,
+                base_model=base_model,
+                model_type=model_type,
+            )
+            if new_info.get('path') != previous_info.get('path'):  # model manager moved model path during rename - don't overwrite it
+                info.path = new_info.get('path')
+            
         ApiDependencies.invoker.services.model_manager.update_model(
             model_name=model_name,
             base_model=base_model,
             model_type=model_type,
             model_attributes=info.dict()
         )
+            
         model_raw = ApiDependencies.invoker.services.model_manager.list_model(
             model_name=model_name,
             base_model=base_model,
             model_type=model_type,
         )
         model_response = parse_obj_as(UpdateModelResponse, model_raw)
-    except KeyError as e:
+    except ModelNotFoundException as e:
         raise HTTPException(status_code=404, detail=str(e))
     except ValueError as e:
+        logger.error(str(e))
+        raise HTTPException(status_code=409, detail=str(e))
+    except Exception as e:
+        logger.error(str(e))
         raise HTTPException(status_code=400, detail=str(e))
 
     return model_response
 
 @models_router.post(
-    "/",
+    "/import",
     operation_id="import_model",
     responses= {
         201: {"description" : "The model imported successfully"},
         404: {"description" : "The model could not be found"},
+        415: {"description" : "Unrecognized file/folder format"},
         424: {"description" : "The model appeared to import successfully, but could not be found in the model manager"},
         409: {"description" : "There is already a model corresponding to this path or repo_id"},
     },
@@ -94,7 +141,7 @@ async def import_model(
         prediction_type: Optional[Literal['v_prediction','epsilon','sample']] = \
                 Body(description='Prediction type for SDv2 checkpoint files', default="v_prediction"),
 ) -> ImportModelResponse:
-    """ Add a model using its local path, repo_id, or remote URL """
+    """ Add a model using its local path, repo_id, or remote URL. Model characteristics will be probed and configured automatically """
     
     items_to_import = {location}
     prediction_types = { x.value: x for x in SchedulerPredictionType }
@@ -109,7 +156,7 @@ async def import_model(
 
         if not info:
             logger.error("Import failed")
-            raise HTTPException(status_code=424)
+            raise HTTPException(status_code=415)
         
         logger.info(f'Successfully imported {location}, got {info}')
         model_raw = ApiDependencies.invoker.services.model_manager.list_model(
@@ -119,25 +166,66 @@ async def import_model(
         )
         return parse_obj_as(ImportModelResponse, model_raw)
     
-    except KeyError as e:
+    except ModelNotFoundException as e:
+        logger.error(str(e))
+        raise HTTPException(status_code=404, detail=str(e))
+    except InvalidModelException as e:
+        logger.error(str(e))
+        raise HTTPException(status_code=415)
+    except ValueError as e:
+        logger.error(str(e))
+        raise HTTPException(status_code=409, detail=str(e))
+        
+@models_router.post(
+    "/add",
+    operation_id="add_model",
+    responses= {
+        201: {"description" : "The model added successfully"},
+        404: {"description" : "The model could not be found"},
+        424: {"description" : "The model appeared to add successfully, but could not be found in the model manager"},
+        409: {"description" : "There is already a model corresponding to this path or repo_id"},
+    },
+    status_code=201,
+    response_model=ImportModelResponse
+)
+async def add_model(
+        info: Union[tuple(OPENAPI_MODEL_CONFIGS)] = Body(description="Model configuration"),
+) -> ImportModelResponse:
+    """ Add a model using the configuration information appropriate for its type. Only local models can be added by path"""
+    
+    logger = ApiDependencies.invoker.services.logger
+
+    try:
+        ApiDependencies.invoker.services.model_manager.add_model(
+            info.model_name,
+            info.base_model,
+            info.model_type,
+            model_attributes = info.dict()
+        )
+        logger.info(f'Successfully added {info.model_name}')
+        model_raw = ApiDependencies.invoker.services.model_manager.list_model(
+            model_name=info.model_name,
+            base_model=info.base_model,
+            model_type=info.model_type
+        )
+        return parse_obj_as(ImportModelResponse, model_raw)
+    except ModelNotFoundException as e:
         logger.error(str(e))
         raise HTTPException(status_code=404, detail=str(e))
     except ValueError as e:
         logger.error(str(e))
         raise HTTPException(status_code=409, detail=str(e))
-        
 
+    
 @models_router.delete(
     "/{base_model}/{model_type}/{model_name}",
     operation_id="del_model",
     responses={
-        204: {
-        "description": "Model deleted successfully"
-        }, 
-        404: {
-        "description": "Model not found"
-        }
+        204: { "description": "Model deleted successfully" }, 
+        404: { "description": "Model not found" }
     },
+    status_code = 204,
+    response_model = None,
 )
 async def delete_model(
         base_model: BaseModelType = Path(description="Base model"),
@@ -154,9 +242,9 @@ async def delete_model(
                                                                  )
         logger.info(f"Deleted model: {model_name}")
         return Response(status_code=204)
-    except KeyError:
-        logger.error(f"Model not found: {model_name}")
-        raise HTTPException(status_code=404, detail=f"Model '{model_name}' not found")
+    except ModelNotFoundException as e:
+        logger.error(str(e))
+        raise HTTPException(status_code=404, detail=str(e))
 
 @models_router.put(
     "/convert/{base_model}/{model_type}/{model_name}",
@@ -173,24 +261,74 @@ async def convert_model(
         base_model: BaseModelType = Path(description="Base model"),
         model_type: ModelType = Path(description="The type of model"),
         model_name: str = Path(description="model name"),
+        convert_dest_directory: Optional[str] = Query(default=None, description="Save the converted model to the designated directory"),
 ) -> ConvertModelResponse:
-    """Convert a checkpoint model into a diffusers model"""
+    """Convert a checkpoint model into a diffusers model, optionally saving to the indicated destination directory, or `models` if none."""
     logger = ApiDependencies.invoker.services.logger
     try:
         logger.info(f"Converting model: {model_name}")
+        dest = pathlib.Path(convert_dest_directory) if convert_dest_directory else None
         ApiDependencies.invoker.services.model_manager.convert_model(model_name,
                                                                      base_model = base_model,
-                                                                     model_type = model_type
+                                                                     model_type = model_type,
+                                                                     convert_dest_directory = dest,
                                                                      )
         model_raw = ApiDependencies.invoker.services.model_manager.list_model(model_name,
                                                                               base_model = base_model,
                                                                               model_type = model_type)
         response = parse_obj_as(ConvertModelResponse, model_raw)
-    except KeyError:
-        raise HTTPException(status_code=404, detail=f"Model '{model_name}' not found")
+    except ModelNotFoundException as e:
+        raise HTTPException(status_code=404, detail=f"Model '{model_name}' not found: {str(e)}")
     except ValueError as e:
         raise HTTPException(status_code=400, detail=str(e))
     return response
+
+@models_router.get(
+    "/search",
+    operation_id="search_for_models",
+    responses={
+        200: { "description": "Directory searched successfully" },
+        404: { "description": "Invalid directory path"  },
+    },
+    status_code = 200,
+    response_model = List[pathlib.Path]
+)
+async def search_for_models(
+        search_path: pathlib.Path = Query(description="Directory path to search for models")
+)->List[pathlib.Path]:
+    if not search_path.is_dir():
+        raise HTTPException(status_code=404, detail=f"The search path '{search_path}' does not exist or is not directory")
+    return ApiDependencies.invoker.services.model_manager.search_for_models([search_path])
+
+@models_router.get(
+    "/ckpt_confs",
+    operation_id="list_ckpt_configs",
+    responses={
+        200: { "description" : "paths retrieved successfully" },
+    },
+    status_code = 200,
+    response_model = List[pathlib.Path]
+)
+async def list_ckpt_configs(
+)->List[pathlib.Path]:
+    """Return a list of the legacy checkpoint configuration files stored in `ROOT/configs/stable-diffusion`, relative to ROOT."""
+    return ApiDependencies.invoker.services.model_manager.list_checkpoint_configs()
+    
+        
+@models_router.get(
+    "/sync",
+    operation_id="sync_to_config",
+    responses={
+        201: { "description": "synchronization successful" },
+    },
+    status_code = 201,
+    response_model = None
+)
+async def sync_to_config(
+)->None:
+    """Call after making changes to models.yaml, autoimport directories or models directory to synchronize
+    in-memory data structures with disk data structures."""
+    return ApiDependencies.invoker.services.model_manager.sync_to_config()
         
 @models_router.put(
     "/merge/{base_model}",
@@ -210,24 +348,75 @@ async def merge_models(
         alpha: Optional[float]                     = Body(description="Alpha weighting strength to apply to 2d and 3d models", default=0.5),
         interp: Optional[MergeInterpolationMethod] = Body(description="Interpolation method"),
         force: Optional[bool]                      = Body(description="Force merging of models created with different versions of diffusers", default=False),
+        merge_dest_directory: Optional[str]       = Body(description="Save the merged model to the designated directory (with 'merged_model_name' appended)", default=None)
 ) -> MergeModelResponse:
     """Convert a checkpoint model into a diffusers model"""
     logger = ApiDependencies.invoker.services.logger
     try:
-        logger.info(f"Merging models: {model_names}")
+        logger.info(f"Merging models: {model_names} into {merge_dest_directory or '<MODELS>'}/{merged_model_name}")
+        dest = pathlib.Path(merge_dest_directory) if merge_dest_directory else None
         result = ApiDependencies.invoker.services.model_manager.merge_models(model_names,
                                                                              base_model,
-                                                                             merged_model_name or "+".join(model_names),
-                                                                             alpha,
-                                                                             interp,
-                                                                             force)
+                                                                             merged_model_name=merged_model_name or "+".join(model_names),
+                                                                             alpha=alpha,
+                                                                             interp=interp,
+                                                                             force=force,
+                                                                             merge_dest_directory = dest
+                                                                             )
         model_raw = ApiDependencies.invoker.services.model_manager.list_model(result.name,
                                                                               base_model = base_model,
                                                                               model_type = ModelType.Main,
                                                                               )
         response = parse_obj_as(ConvertModelResponse, model_raw)
-    except KeyError:
+    except ModelNotFoundException:
         raise HTTPException(status_code=404, detail=f"One or more of the models '{model_names}' not found")
     except ValueError as e:
         raise HTTPException(status_code=400, detail=str(e))
     return response
+
+# The rename operation is now supported by update_model and no longer needs to be
+# a standalone route.
+# @models_router.post(
+#     "/rename/{base_model}/{model_type}/{model_name}",
+#     operation_id="rename_model",
+#     responses= {
+#         201: {"description" : "The model was renamed successfully"},
+#         404: {"description" : "The model could not be found"},
+#         409: {"description" : "There is already a model corresponding to the new name"},
+#     },
+#     status_code=201,
+#     response_model=ImportModelResponse
+# )
+# async def rename_model(
+#         base_model: BaseModelType = Path(description="Base model"),
+#         model_type: ModelType = Path(description="The type of model"),
+#         model_name: str = Path(description="current model name"),
+#         new_name: Optional[str] = Query(description="new model name", default=None),
+#         new_base: Optional[BaseModelType] = Query(description="new model base", default=None),
+# ) -> ImportModelResponse:
+#     """ Rename a model"""
+    
+#     logger = ApiDependencies.invoker.services.logger
+
+#     try:
+#         result = ApiDependencies.invoker.services.model_manager.rename_model(
+#             base_model = base_model,
+#             model_type = model_type,
+#             model_name = model_name,
+#             new_name = new_name,
+#             new_base = new_base,
+#         )
+#         logger.debug(result)
+#         logger.info(f'Successfully renamed {model_name}=>{new_name}')
+#         model_raw = ApiDependencies.invoker.services.model_manager.list_model(
+#             model_name=new_name or model_name,
+#             base_model=new_base or base_model,
+#             model_type=model_type
+#         )
+#         return parse_obj_as(ImportModelResponse, model_raw)
+#     except ModelNotFoundException as e:
+#         logger.error(str(e))
+#         raise HTTPException(status_code=404, detail=str(e))
+#     except ValueError as e:
+#         logger.error(str(e))
+#         raise HTTPException(status_code=409, detail=str(e))

invokeai/app/api_app.py

@@ -1,5 +1,6 @@
-# Copyright (c) 2022 Kyle Schouviller (https://github.com/kyle0654)
+# Copyright (c) 2022-2023 Kyle Schouviller (https://github.com/kyle0654) and the InvokeAI Team
 import asyncio
+import sys
 from inspect import signature
 
 import uvicorn
@@ -20,6 +21,13 @@ from ..backend.util.logging import InvokeAILogger
 app_config = InvokeAIAppConfig.get_config()
 app_config.parse_args()
 logger = InvokeAILogger.getLogger(config=app_config)
+from invokeai.version.invokeai_version import __version__
+
+# we call this early so that the message appears before
+# other invokeai initialization messages
+if app_config.version:
+    print(f'InvokeAI version {__version__}')
+    sys.exit(0)
 
 import invokeai.frontend.web as web_dir
 import mimetypes
@@ -28,8 +36,10 @@ from .api.dependencies import ApiDependencies
 from .api.routers import sessions, models, images, boards, board_images, app_info
 from .api.sockets import SocketIO
 from .invocations.baseinvocation import BaseInvocation
+    
 
 import torch
+import invokeai.backend.util.hotfixes
 if torch.backends.mps.is_available():
     import invokeai.backend.util.mps_fixes
 

invokeai/app/cli_app.py

@@ -16,6 +16,12 @@ from invokeai.backend.util.logging import InvokeAILogger
 config = InvokeAIAppConfig.get_config()
 config.parse_args()
 logger = InvokeAILogger().getLogger(config=config)
+from invokeai.version.invokeai_version import __version__
+
+# we call this early so that the message appears before other invokeai initialization messages
+if config.version:
+    print(f'InvokeAI version {__version__}')
+    sys.exit(0)
 
 from invokeai.app.services.board_image_record_storage import (
     SqliteBoardImageRecordStorage,
@@ -28,7 +34,6 @@ from invokeai.app.services.board_record_storage import SqliteBoardRecordStorage
 from invokeai.app.services.boards import BoardService, BoardServiceDependencies
 from invokeai.app.services.image_record_storage import SqliteImageRecordStorage
 from invokeai.app.services.images import ImageService, ImageServiceDependencies
-from invokeai.app.services.metadata import CoreMetadataService
 from invokeai.app.services.resource_name import SimpleNameService
 from invokeai.app.services.urls import LocalUrlService
 from .services.default_graphs import (default_text_to_image_graph_id,
@@ -49,10 +54,10 @@ from .services.invocation_services import InvocationServices
 from .services.invoker import Invoker
 from .services.model_manager_service import ModelManagerService
 from .services.processor import DefaultInvocationProcessor
-from .services.restoration_services import RestorationServices
 from .services.sqlite import SqliteItemStorage
 
 import torch
+import invokeai.backend.util.hotfixes
 if torch.backends.mps.is_available():
     import invokeai.backend.util.mps_fixes
 
@@ -208,6 +213,7 @@ def invoke_all(context: CliContext):
         raise SessionError()
 
 def invoke_cli():
+    logger.info(f'InvokeAI version {__version__}')
     # get the optional list of invocations to execute on the command line
     parser = config.get_parser()
     parser.add_argument('commands',nargs='*')
@@ -237,7 +243,6 @@ def invoke_cli():
         )
 
     urls = LocalUrlService()
-    metadata = CoreMetadataService()
     image_record_storage = SqliteImageRecordStorage(db_location)
     image_file_storage = DiskImageFileStorage(f"{output_folder}/images")
     names = SimpleNameService()
@@ -270,7 +275,6 @@ def invoke_cli():
             board_image_record_storage=board_image_record_storage,
             image_record_storage=image_record_storage,
             image_file_storage=image_file_storage,
-            metadata=metadata,
             url=urls,
             logger=logger,
             names=names,
@@ -291,7 +295,6 @@ def invoke_cli():
         ),
         graph_execution_manager=graph_execution_manager,
         processor=DefaultInvocationProcessor(),
-        restoration=RestorationServices(config,logger=logger),
         logger=logger,
         configuration=config,
     )

invokeai/app/invocations/collections.py

@@ -4,17 +4,12 @@ from typing import Literal
 
 import numpy as np
 from pydantic import Field, validator
-from invokeai.app.models.image import ImageField
 
+from invokeai.app.models.image import ImageField
 from invokeai.app.util.misc import SEED_MAX, get_random_seed
 
-from .baseinvocation import (
-    BaseInvocation,
-    InvocationConfig,
-    InvocationContext,
-    BaseInvocationOutput,
-    UIConfig,
-)
+from .baseinvocation import (BaseInvocation, BaseInvocationOutput,
+                             InvocationConfig, InvocationContext, UIConfig)
 
 
 class IntCollectionOutput(BaseInvocationOutput):
@@ -32,7 +27,8 @@ class FloatCollectionOutput(BaseInvocationOutput):
     type: Literal["float_collection"] = "float_collection"
 
     # Outputs
-    collection: list[float] = Field(default=[], description="The float collection")
+    collection: list[float] = Field(
+        default=[], description="The float collection")
 
 
 class ImageCollectionOutput(BaseInvocationOutput):
@@ -41,7 +37,8 @@ class ImageCollectionOutput(BaseInvocationOutput):
     type: Literal["image_collection"] = "image_collection"
 
     # Outputs
-    collection: list[ImageField] = Field(default=[], description="The output images")
+    collection: list[ImageField] = Field(
+        default=[], description="The output images")
 
     class Config:
         schema_extra = {"required": ["type", "collection"]}
@@ -57,6 +54,14 @@ class RangeInvocation(BaseInvocation):
     stop: int = Field(default=10, description="The stop of the range")
     step: int = Field(default=1, description="The step of the range")
 
+    class Config(InvocationConfig):
+        schema_extra = {
+            "ui": {
+                "title": "Range",
+                "tags": ["range", "integer", "collection"]
+            },
+        }
+
     @validator("stop")
     def stop_gt_start(cls, v, values):
         if "start" in values and v <= values["start"]:
@@ -79,10 +84,20 @@ class RangeOfSizeInvocation(BaseInvocation):
     size: int = Field(default=1, description="The number of values")
     step: int = Field(default=1, description="The step of the range")
 
+    class Config(InvocationConfig):
+        schema_extra = {
+            "ui": {
+                "title": "Sized Range",
+                "tags": ["range", "integer", "size", "collection"]
+            },
+        }
+
     def invoke(self, context: InvocationContext) -> IntCollectionOutput:
         return IntCollectionOutput(
-            collection=list(range(self.start, self.start + self.size, self.step))
-        )
+            collection=list(
+                range(
+                    self.start, self.start + self.size,
+                    self.step)))
 
 
 class RandomRangeInvocation(BaseInvocation):
@@ -103,11 +118,21 @@ class RandomRangeInvocation(BaseInvocation):
         default_factory=get_random_seed,
     )
 
+    class Config(InvocationConfig):
+        schema_extra = {
+            "ui": {
+                "title": "Random Range",
+                "tags": ["range", "integer", "random", "collection"]
+            },
+        }
+
     def invoke(self, context: InvocationContext) -> IntCollectionOutput:
         rng = np.random.default_rng(self.seed)
         return IntCollectionOutput(
-            collection=list(rng.integers(low=self.low, high=self.high, size=self.size))
-        )
+            collection=list(
+                rng.integers(
+                    low=self.low, high=self.high,
+                    size=self.size)))
 
 
 class ImageCollectionInvocation(BaseInvocation):
@@ -121,6 +146,7 @@ class ImageCollectionInvocation(BaseInvocation):
         default=[], description="The image collection to load"
     )
     # fmt: on
+
     def invoke(self, context: InvocationContext) -> ImageCollectionOutput:
         return ImageCollectionOutput(collection=self.images)
 
@@ -128,6 +154,7 @@ class ImageCollectionInvocation(BaseInvocation):
         schema_extra = {
             "ui": {
                 "type_hints": {
+                    "title": "Image Collection",
                     "images": "image_collection",
                 }
             },

invokeai/app/invocations/compel.py

@@ -1,8 +1,16 @@
-from typing import Literal, Optional, Union, List
+from typing import Literal, Optional, Union, List, Annotated
 from pydantic import BaseModel, Field
 import re
+
+from .baseinvocation import BaseInvocation, BaseInvocationOutput, InvocationContext, InvocationConfig
+from .model import ClipField
+
+from ...backend.util.devices import torch_dtype
+from ...backend.stable_diffusion.diffusion import InvokeAIDiffuserComponent
+from ...backend.model_management import BaseModelType, ModelType, SubModelType, ModelPatcher
+
 import torch
-from compel import Compel
+from compel import Compel, ReturnedEmbeddingsType
 from compel.prompt_parser import (Blend, Conjunction,
                                   CrossAttentionControlSubstitute,
                                   FlattenedPrompt, Fragment)
@@ -14,6 +22,7 @@ from ...backend.stable_diffusion.diffusion import InvokeAIDiffuserComponent
 from .baseinvocation import (BaseInvocation, BaseInvocationOutput,
                              InvocationConfig, InvocationContext)
 from .model import ClipField
+from dataclasses import dataclass
 
 
 class ConditioningField(BaseModel):
@@ -23,6 +32,34 @@ class ConditioningField(BaseModel):
     class Config:
         schema_extra = {"required": ["conditioning_name"]}
 
+@dataclass
+class BasicConditioningInfo:
+    #type: Literal["basic_conditioning"] = "basic_conditioning"
+    embeds: torch.Tensor
+    extra_conditioning: Optional[InvokeAIDiffuserComponent.ExtraConditioningInfo]
+    # weight: float
+    # mode: ConditioningAlgo
+
+@dataclass
+class SDXLConditioningInfo(BasicConditioningInfo):
+    #type: Literal["sdxl_conditioning"] = "sdxl_conditioning"
+    pooled_embeds: torch.Tensor
+    add_time_ids: torch.Tensor
+
+ConditioningInfoType = Annotated[
+    Union[BasicConditioningInfo, SDXLConditioningInfo],
+    Field(discriminator="type")
+]
+
+@dataclass
+class ConditioningFieldData:
+    conditionings: List[Union[BasicConditioningInfo, SDXLConditioningInfo]]
+    #unconditioned: Optional[torch.Tensor]
+
+#class ConditioningAlgo(str, Enum):
+#    Compose = "compose"
+#    ComposeEx = "compose_ex"
+#    PerpNeg = "perp_neg"
 
 class CompelOutput(BaseInvocationOutput):
     """Compel parser output"""
@@ -57,10 +94,10 @@ class CompelInvocation(BaseInvocation):
     @torch.no_grad()
     def invoke(self, context: InvocationContext) -> CompelOutput:
         tokenizer_info = context.services.model_manager.get_model(
-            **self.clip.tokenizer.dict(),
+            **self.clip.tokenizer.dict(), context=context,
         )
         text_encoder_info = context.services.model_manager.get_model(
-            **self.clip.text_encoder.dict(),
+            **self.clip.text_encoder.dict(), context=context,
         )
 
         def _lora_loader():
@@ -82,6 +119,7 @@ class CompelInvocation(BaseInvocation):
                         model_name=name,
                         base_model=self.clip.text_encoder.base_model,
                         model_type=ModelType.TextualInversion,
+                        context=context,
                     ).context.model
                 )
             except ModelNotFoundException:
@@ -100,7 +138,7 @@ class CompelInvocation(BaseInvocation):
                 text_encoder=text_encoder,
                 textual_inversion_manager=ti_manager,
                 dtype_for_device_getter=torch_dtype,
-                truncate_long_prompts=True,  # TODO:
+                truncate_long_prompts=True,
             )
 
             conjunction = Compel.parse_prompt_string(self.prompt)
@@ -112,19 +150,25 @@ class CompelInvocation(BaseInvocation):
             c, options = compel.build_conditioning_tensor_for_prompt_object(
                 prompt)
 
-            # TODO: long prompt support
-            # if not self.truncate_long_prompts:
-            #    [c, uc] = compel.pad_conditioning_tensors_to_same_length([c, uc])
             ec = InvokeAIDiffuserComponent.ExtraConditioningInfo(
                 tokens_count_including_eos_bos=get_max_token_count(
                     tokenizer, conjunction),
                 cross_attention_control_args=options.get(
                     "cross_attention_control", None),)
 
-        conditioning_name = f"{context.graph_execution_state_id}_{self.id}_conditioning"
+        c = c.detach().to("cpu")
 
-        # TODO: hacky but works ;D maybe rename latents somehow?
-        context.services.latents.save(conditioning_name, (c, ec))
+        conditioning_data = ConditioningFieldData(
+            conditionings=[
+                BasicConditioningInfo(
+                    embeds=c,
+                    extra_conditioning=ec,
+                )
+            ]
+        )
+
+        conditioning_name = f"{context.graph_execution_state_id}_{self.id}_conditioning"
+        context.services.latents.save(conditioning_name, conditioning_data)
 
         return CompelOutput(
             conditioning=ConditioningField(
@@ -132,6 +176,397 @@ class CompelInvocation(BaseInvocation):
             ),
         )
 
+class SDXLPromptInvocationBase:
+    def run_clip_raw(self, context, clip_field, prompt, get_pooled):
+        tokenizer_info = context.services.model_manager.get_model(
+            **clip_field.tokenizer.dict(),
+        )
+        text_encoder_info = context.services.model_manager.get_model(
+            **clip_field.text_encoder.dict(),
+        )
+
+        def _lora_loader():
+            for lora in clip_field.loras:
+                lora_info = context.services.model_manager.get_model(
+                    **lora.dict(exclude={"weight"}))
+                yield (lora_info.context.model, lora.weight)
+                del lora_info
+            return
+
+        #loras = [(context.services.model_manager.get_model(**lora.dict(exclude={"weight"})).context.model, lora.weight) for lora in self.clip.loras]
+
+        ti_list = []
+        for trigger in re.findall(r"<[a-zA-Z0-9., _-]+>", prompt):
+            name = trigger[1:-1]
+            try:
+                ti_list.append(
+                    context.services.model_manager.get_model(
+                        model_name=name,
+                        base_model=clip_field.text_encoder.base_model,
+                        model_type=ModelType.TextualInversion,
+                    ).context.model
+                )
+            except ModelNotFoundException:
+                # print(e)
+                #import traceback
+                #print(traceback.format_exc())
+                print(f"Warn: trigger: \"{trigger}\" not found")
+
+        with ModelPatcher.apply_lora_text_encoder(text_encoder_info.context.model, _lora_loader()),\
+                ModelPatcher.apply_ti(tokenizer_info.context.model, text_encoder_info.context.model, ti_list) as (tokenizer, ti_manager),\
+                ModelPatcher.apply_clip_skip(text_encoder_info.context.model, clip_field.skipped_layers),\
+                text_encoder_info as text_encoder:
+
+            text_inputs = tokenizer(
+                prompt,
+                padding="max_length",
+                max_length=tokenizer.model_max_length,
+                truncation=True,
+                return_tensors="pt",
+            )
+            text_input_ids = text_inputs.input_ids
+            prompt_embeds = text_encoder(
+                text_input_ids.to(text_encoder.device),
+                output_hidden_states=True,
+            )
+            if get_pooled:
+                c_pooled = prompt_embeds[0]
+            else:
+                c_pooled = None
+            c = prompt_embeds.hidden_states[-2]
+
+        del tokenizer
+        del text_encoder
+        del tokenizer_info
+        del text_encoder_info
+
+        c = c.detach().to("cpu")
+        if c_pooled is not None:
+            c_pooled = c_pooled.detach().to("cpu")
+
+        return c, c_pooled, None
+
+    def run_clip_compel(self, context, clip_field, prompt, get_pooled):
+        tokenizer_info = context.services.model_manager.get_model(
+            **clip_field.tokenizer.dict(),
+        )
+        text_encoder_info = context.services.model_manager.get_model(
+            **clip_field.text_encoder.dict(),
+        )
+
+        def _lora_loader():
+            for lora in clip_field.loras:
+                lora_info = context.services.model_manager.get_model(
+                    **lora.dict(exclude={"weight"}))
+                yield (lora_info.context.model, lora.weight)
+                del lora_info
+            return
+
+        #loras = [(context.services.model_manager.get_model(**lora.dict(exclude={"weight"})).context.model, lora.weight) for lora in self.clip.loras]
+
+        ti_list = []
+        for trigger in re.findall(r"<[a-zA-Z0-9., _-]+>", prompt):
+            name = trigger[1:-1]
+            try:
+                ti_list.append(
+                    context.services.model_manager.get_model(
+                        model_name=name,
+                        base_model=clip_field.text_encoder.base_model,
+                        model_type=ModelType.TextualInversion,
+                    ).context.model
+                )
+            except ModelNotFoundException:
+                # print(e)
+                #import traceback
+                #print(traceback.format_exc())
+                print(f"Warn: trigger: \"{trigger}\" not found")
+
+        with ModelPatcher.apply_lora_text_encoder(text_encoder_info.context.model, _lora_loader()),\
+                ModelPatcher.apply_ti(tokenizer_info.context.model, text_encoder_info.context.model, ti_list) as (tokenizer, ti_manager),\
+                ModelPatcher.apply_clip_skip(text_encoder_info.context.model, clip_field.skipped_layers),\
+                text_encoder_info as text_encoder:
+
+            compel = Compel(
+                tokenizer=tokenizer,
+                text_encoder=text_encoder,
+                textual_inversion_manager=ti_manager,
+                dtype_for_device_getter=torch_dtype,
+                truncate_long_prompts=True,  # TODO:
+                returned_embeddings_type=ReturnedEmbeddingsType.PENULTIMATE_HIDDEN_STATES_NON_NORMALIZED, # TODO: clip skip
+                requires_pooled=True,
+            )
+
+            conjunction = Compel.parse_prompt_string(prompt)
+
+            if context.services.configuration.log_tokenization:
+                # TODO: better logging for and syntax
+                for prompt_obj in conjunction.prompts:
+                    log_tokenization_for_prompt_object(prompt_obj, tokenizer)
+
+            # TODO: ask for optimizations? to not run text_encoder twice
+            c, options = compel.build_conditioning_tensor_for_conjunction(conjunction)
+            if get_pooled:
+                c_pooled = compel.conditioning_provider.get_pooled_embeddings([prompt])
+            else:
+                c_pooled = None
+
+            ec = InvokeAIDiffuserComponent.ExtraConditioningInfo(
+                tokens_count_including_eos_bos=get_max_token_count(tokenizer, conjunction),
+                cross_attention_control_args=options.get("cross_attention_control", None),
+            )
+
+        del tokenizer
+        del text_encoder
+        del tokenizer_info
+        del text_encoder_info
+
+        c = c.detach().to("cpu")
+        if c_pooled is not None:
+            c_pooled = c_pooled.detach().to("cpu")
+
+        return c, c_pooled, ec
+
+class SDXLCompelPromptInvocation(BaseInvocation, SDXLPromptInvocationBase):
+    """Parse prompt using compel package to conditioning."""
+
+    type: Literal["sdxl_compel_prompt"] = "sdxl_compel_prompt"
+
+    prompt: str = Field(default="", description="Prompt")
+    style: str = Field(default="", description="Style prompt")
+    original_width: int = Field(1024, description="")
+    original_height: int = Field(1024, description="")
+    crop_top: int = Field(0, description="")
+    crop_left: int = Field(0, description="")
+    target_width: int = Field(1024, description="")
+    target_height: int = Field(1024, description="")
+    clip: ClipField = Field(None, description="Clip to use")
+    clip2: ClipField = Field(None, description="Clip2 to use")
+
+    # Schema customisation
+    class Config(InvocationConfig):
+        schema_extra = {
+            "ui": {
+                "title": "SDXL Prompt (Compel)",
+                "tags": ["prompt", "compel"],
+                "type_hints": {
+                    "model": "model"
+                }
+            },
+        }
+
+    @torch.no_grad()
+    def invoke(self, context: InvocationContext) -> CompelOutput:
+        c1, c1_pooled, ec1 = self.run_clip_compel(context, self.clip, self.prompt, False)
+        if self.style.strip() == "":
+            c2, c2_pooled, ec2 = self.run_clip_compel(context, self.clip2, self.prompt, True)
+        else:
+            c2, c2_pooled, ec2 = self.run_clip_compel(context, self.clip2, self.style, True)
+
+        original_size = (self.original_height, self.original_width)
+        crop_coords = (self.crop_top, self.crop_left)
+        target_size = (self.target_height, self.target_width)
+
+        add_time_ids = torch.tensor([
+            original_size + crop_coords + target_size
+        ])
+
+        conditioning_data = ConditioningFieldData(
+            conditionings=[
+                SDXLConditioningInfo(
+                    embeds=torch.cat([c1, c2], dim=-1),
+                    pooled_embeds=c2_pooled,
+                    add_time_ids=add_time_ids,
+                    extra_conditioning=ec1,
+                )
+            ]
+        )
+
+        conditioning_name = f"{context.graph_execution_state_id}_{self.id}_conditioning"
+        context.services.latents.save(conditioning_name, conditioning_data)
+
+        return CompelOutput(
+            conditioning=ConditioningField(
+                conditioning_name=conditioning_name,
+            ),
+        )
+
+class SDXLRefinerCompelPromptInvocation(BaseInvocation, SDXLPromptInvocationBase):
+    """Parse prompt using compel package to conditioning."""
+
+    type: Literal["sdxl_refiner_compel_prompt"] = "sdxl_refiner_compel_prompt"
+
+    style: str = Field(default="", description="Style prompt") # TODO: ?
+    original_width: int = Field(1024, description="")
+    original_height: int = Field(1024, description="")
+    crop_top: int = Field(0, description="")
+    crop_left: int = Field(0, description="")
+    aesthetic_score: float = Field(6.0, description="")
+    clip2: ClipField = Field(None, description="Clip to use")
+
+    # Schema customisation
+    class Config(InvocationConfig):
+        schema_extra = {
+            "ui": {
+                "title": "SDXL Refiner Prompt (Compel)",
+                "tags": ["prompt", "compel"],
+                "type_hints": {
+                    "model": "model"
+                }
+            },
+        }
+
+    @torch.no_grad()
+    def invoke(self, context: InvocationContext) -> CompelOutput:
+        c2, c2_pooled, ec2 = self.run_clip_compel(context, self.clip2, self.style, True)
+
+        original_size = (self.original_height, self.original_width)
+        crop_coords = (self.crop_top, self.crop_left)
+
+        add_time_ids = torch.tensor([
+            original_size + crop_coords + (self.aesthetic_score,)
+        ])
+
+        conditioning_data = ConditioningFieldData(
+            conditionings=[
+                SDXLConditioningInfo(
+                    embeds=c2,
+                    pooled_embeds=c2_pooled,
+                    add_time_ids=add_time_ids,
+                    extra_conditioning=ec2, # or None
+                )
+            ]
+        )
+
+        conditioning_name = f"{context.graph_execution_state_id}_{self.id}_conditioning"
+        context.services.latents.save(conditioning_name, conditioning_data)
+
+        return CompelOutput(
+            conditioning=ConditioningField(
+                conditioning_name=conditioning_name,
+            ),
+        )
+
+class SDXLRawPromptInvocation(BaseInvocation, SDXLPromptInvocationBase):
+    """Pass unmodified prompt to conditioning without compel processing."""
+
+    type: Literal["sdxl_raw_prompt"] = "sdxl_raw_prompt"
+
+    prompt: str = Field(default="", description="Prompt")
+    style: str = Field(default="", description="Style prompt")
+    original_width: int = Field(1024, description="")
+    original_height: int = Field(1024, description="")
+    crop_top: int = Field(0, description="")
+    crop_left: int = Field(0, description="")
+    target_width: int = Field(1024, description="")
+    target_height: int = Field(1024, description="")
+    clip: ClipField = Field(None, description="Clip to use")
+    clip2: ClipField = Field(None, description="Clip2 to use")
+
+    # Schema customisation
+    class Config(InvocationConfig):
+        schema_extra = {
+            "ui": {
+                "title": "SDXL Prompt (Raw)",
+                "tags": ["prompt", "compel"],
+                "type_hints": {
+                    "model": "model"
+                }
+            },
+        }
+
+    @torch.no_grad()
+    def invoke(self, context: InvocationContext) -> CompelOutput:
+        c1, c1_pooled, ec1 = self.run_clip_raw(context, self.clip, self.prompt, False)
+        if self.style.strip() == "":
+            c2, c2_pooled, ec2 = self.run_clip_raw(context, self.clip2, self.prompt, True)
+        else:
+            c2, c2_pooled, ec2 = self.run_clip_raw(context, self.clip2, self.style, True)
+
+        original_size = (self.original_height, self.original_width)
+        crop_coords = (self.crop_top, self.crop_left)
+        target_size = (self.target_height, self.target_width)
+
+        add_time_ids = torch.tensor([
+            original_size + crop_coords + target_size
+        ])
+
+        conditioning_data = ConditioningFieldData(
+            conditionings=[
+                SDXLConditioningInfo(
+                    embeds=torch.cat([c1, c2], dim=-1),
+                    pooled_embeds=c2_pooled,
+                    add_time_ids=add_time_ids,
+                    extra_conditioning=ec1,
+                )
+            ]
+        )
+
+        conditioning_name = f"{context.graph_execution_state_id}_{self.id}_conditioning"
+        context.services.latents.save(conditioning_name, conditioning_data)
+
+        return CompelOutput(
+            conditioning=ConditioningField(
+                conditioning_name=conditioning_name,
+            ),
+        )
+
+class SDXLRefinerRawPromptInvocation(BaseInvocation, SDXLPromptInvocationBase):
+    """Parse prompt using compel package to conditioning."""
+
+    type: Literal["sdxl_refiner_raw_prompt"] = "sdxl_refiner_raw_prompt"
+
+    style: str = Field(default="", description="Style prompt") # TODO: ?
+    original_width: int = Field(1024, description="")
+    original_height: int = Field(1024, description="")
+    crop_top: int = Field(0, description="")
+    crop_left: int = Field(0, description="")
+    aesthetic_score: float = Field(6.0, description="")
+    clip2: ClipField = Field(None, description="Clip to use")
+
+    # Schema customisation
+    class Config(InvocationConfig):
+        schema_extra = {
+            "ui": {
+                "title": "SDXL Refiner Prompt (Raw)",
+                "tags": ["prompt", "compel"],
+                "type_hints": {
+                    "model": "model"
+                }
+            },
+        }
+
+    @torch.no_grad()
+    def invoke(self, context: InvocationContext) -> CompelOutput:
+        c2, c2_pooled, ec2 = self.run_clip_raw(context, self.clip2, self.style, True)
+
+        original_size = (self.original_height, self.original_width)
+        crop_coords = (self.crop_top, self.crop_left)
+
+        add_time_ids = torch.tensor([
+            original_size + crop_coords + (self.aesthetic_score,)
+        ])
+
+        conditioning_data = ConditioningFieldData(
+            conditionings=[
+                SDXLConditioningInfo(
+                    embeds=c2,
+                    pooled_embeds=c2_pooled,
+                    add_time_ids=add_time_ids,
+                    extra_conditioning=ec2, # or None
+                )
+            ]
+        )
+
+        conditioning_name = f"{context.graph_execution_state_id}_{self.id}_conditioning"
+        context.services.latents.save(conditioning_name, conditioning_data)
+
+        return CompelOutput(
+            conditioning=ConditioningField(
+                conditioning_name=conditioning_name,
+            ),
+        )
+
+
 class ClipSkipInvocationOutput(BaseInvocationOutput):
     """Clip skip node output"""
     type: Literal["clip_skip_output"] = "clip_skip_output"
@@ -144,6 +579,14 @@ class ClipSkipInvocation(BaseInvocation):
     clip: ClipField = Field(None, description="Clip to use")
     skipped_layers: int = Field(0, description="Number of layers to skip in text_encoder")
 
+    class Config(InvocationConfig):
+        schema_extra = {
+            "ui": {
+                "title": "CLIP Skip",
+                "tags": ["clip", "skip"]
+            },
+        }
+
     def invoke(self, context: InvocationContext) -> ClipSkipInvocationOutput:
         self.clip.skipped_layers += self.skipped_layers
         return ClipSkipInvocationOutput(

invokeai/app/invocations/controlnet_image_processors.py

@@ -1,42 +1,25 @@
 # Invocations for ControlNet image preprocessors
 # initial implementation by Gregg Helt, 2023
 # heavily leverages controlnet_aux package: https://github.com/patrickvonplaten/controlnet_aux
-from builtins import float, bool
+from builtins import bool, float
+from typing import Dict, List, Literal, Optional, Union
 
 import cv2
 import numpy as np
-from typing import Literal, Optional, Union, List, Dict
+from controlnet_aux import (CannyDetector, ContentShuffleDetector, HEDdetector,
+                            LeresDetector, LineartAnimeDetector,
+                            LineartDetector, MediapipeFaceDetector,
+                            MidasDetector, MLSDdetector, NormalBaeDetector,
+                            OpenposeDetector, PidiNetDetector, SamDetector,
+                            ZoeDetector)
+from controlnet_aux.util import HWC3, ade_palette
 from PIL import Image
 from pydantic import BaseModel, Field, validator
 
-from ..models.image import ImageField, ImageCategory, ResourceOrigin
-from .baseinvocation import (
-    BaseInvocation,
-    BaseInvocationOutput,
-    InvocationContext,
-    InvocationConfig,
-)
-
-from controlnet_aux import (
-    CannyDetector,
-    HEDdetector,
-    LineartDetector,
-    LineartAnimeDetector,
-    MidasDetector,
-    MLSDdetector,
-    NormalBaeDetector,
-    OpenposeDetector,
-    PidiNetDetector,
-    ContentShuffleDetector,
-    ZoeDetector,
-    MediapipeFaceDetector,
-    SamDetector,
-    LeresDetector,
-)
-
-from controlnet_aux.util import HWC3, ade_palette
-
-
+from ...backend.model_management import BaseModelType, ModelType
+from ..models.image import ImageCategory, ImageField, ResourceOrigin
+from .baseinvocation import (BaseInvocation, BaseInvocationOutput,
+                             InvocationConfig, InvocationContext)
 from .image import ImageOutput, PILInvocationConfig
 
 CONTROLNET_DEFAULT_MODELS = [
@@ -74,66 +57,82 @@ CONTROLNET_DEFAULT_MODELS = [
     "lllyasviel/control_v11e_sd15_ip2p",
     "lllyasviel/control_v11f1e_sd15_tile",
 
-     #################################################
-     #  thibaud sd v2.1 models (ControlNet v1.0? or v1.1?
-     ##################################################
-     "thibaud/controlnet-sd21-openpose-diffusers",
-     "thibaud/controlnet-sd21-canny-diffusers",
-     "thibaud/controlnet-sd21-depth-diffusers",
-     "thibaud/controlnet-sd21-scribble-diffusers",
-     "thibaud/controlnet-sd21-hed-diffusers",
-     "thibaud/controlnet-sd21-zoedepth-diffusers",
-     "thibaud/controlnet-sd21-color-diffusers",
-     "thibaud/controlnet-sd21-openposev2-diffusers",
-     "thibaud/controlnet-sd21-lineart-diffusers",
-     "thibaud/controlnet-sd21-normalbae-diffusers",
-     "thibaud/controlnet-sd21-ade20k-diffusers",
+    #################################################
+    #  thibaud sd v2.1 models (ControlNet v1.0? or v1.1?
+    ##################################################
+    "thibaud/controlnet-sd21-openpose-diffusers",
+    "thibaud/controlnet-sd21-canny-diffusers",
+    "thibaud/controlnet-sd21-depth-diffusers",
+    "thibaud/controlnet-sd21-scribble-diffusers",
+    "thibaud/controlnet-sd21-hed-diffusers",
+    "thibaud/controlnet-sd21-zoedepth-diffusers",
+    "thibaud/controlnet-sd21-color-diffusers",
+    "thibaud/controlnet-sd21-openposev2-diffusers",
+    "thibaud/controlnet-sd21-lineart-diffusers",
+    "thibaud/controlnet-sd21-normalbae-diffusers",
+    "thibaud/controlnet-sd21-ade20k-diffusers",
 
-     ##############################################
-     #  ControlNetMediaPipeface, ControlNet v1.1
-     ##############################################
-     # ["CrucibleAI/ControlNetMediaPipeFace", "diffusion_sd15"],  # SD 1.5
-     #    diffusion_sd15 needs to be passed to from_pretrained() as subfolder arg
-     #    hacked t2l to split to model & subfolder if format is "model,subfolder"
-     "CrucibleAI/ControlNetMediaPipeFace,diffusion_sd15",  # SD 1.5
-     "CrucibleAI/ControlNetMediaPipeFace",  # SD 2.1?
+    ##############################################
+    #  ControlNetMediaPipeface, ControlNet v1.1
+    ##############################################
+    # ["CrucibleAI/ControlNetMediaPipeFace", "diffusion_sd15"],  # SD 1.5
+    #    diffusion_sd15 needs to be passed to from_pretrained() as subfolder arg
+    #    hacked t2l to split to model & subfolder if format is "model,subfolder"
+    "CrucibleAI/ControlNetMediaPipeFace,diffusion_sd15",  # SD 1.5
+    "CrucibleAI/ControlNetMediaPipeFace",  # SD 2.1?
 ]
 
 CONTROLNET_NAME_VALUES = Literal[tuple(CONTROLNET_DEFAULT_MODELS)]
-CONTROLNET_MODE_VALUES = Literal[tuple(["balanced", "more_prompt", "more_control", "unbalanced"])]
+CONTROLNET_MODE_VALUES = Literal[tuple(
+    ["balanced", "more_prompt", "more_control", "unbalanced"])]
 # crop and fill options not ready yet
 # CONTROLNET_RESIZE_VALUES = Literal[tuple(["just_resize", "crop_resize", "fill_resize"])]
 
 
+class ControlNetModelField(BaseModel):
+    """ControlNet model field"""
+
+    model_name: str = Field(description="Name of the ControlNet model")
+    base_model: BaseModelType = Field(description="Base model")
+
+
 class ControlField(BaseModel):
     image: ImageField = Field(default=None, description="The control image")
-    control_model: Optional[str] = Field(default=None, description="The ControlNet model to use")
+    control_model: Optional[ControlNetModelField] = Field(
+        default=None, description="The ControlNet model to use")
     # control_weight: Optional[float] = Field(default=1, description="weight given to controlnet")
-    control_weight: Union[float, List[float]] = Field(default=1, description="The weight given to the ControlNet")
-    begin_step_percent: float = Field(default=0, ge=0, le=1,
-                                      description="When the ControlNet is first applied (% of total steps)")
-    end_step_percent: float = Field(default=1, ge=0, le=1,
-                                    description="When the ControlNet is last applied (% of total steps)")
-    control_mode: CONTROLNET_MODE_VALUES = Field(default="balanced", description="The control mode to use")
+    control_weight: Union[float, List[float]] = Field(
+        default=1, description="The weight given to the ControlNet")
+    begin_step_percent: float = Field(
+        default=0, ge=0, le=1,
+        description="When the ControlNet is first applied (% of total steps)")
+    end_step_percent: float = Field(
+        default=1, ge=0, le=1,
+        description="When the ControlNet is last applied (% of total steps)")
+    control_mode: CONTROLNET_MODE_VALUES = Field(
+        default="balanced", description="The control mode to use")
     # resize_mode: CONTROLNET_RESIZE_VALUES = Field(default="just_resize", description="The resize mode to use")
 
     @validator("control_weight")
-    def abs_le_one(cls, v):
-        """validate that all abs(values) are <=1"""
+    def validate_control_weight(cls, v):
+        """Validate that all control weights in the valid range"""
         if isinstance(v, list):
             for i in v:
-                if abs(i) > 1:
-                    raise ValueError('all abs(control_weight) must be <= 1')
+                if i < -1 or i > 2:
+                    raise ValueError(
+                        'Control weights must be within -1 to 2 range')
         else:
-            if abs(v) > 1:
-                raise ValueError('abs(control_weight) must be <= 1')
+            if v < -1 or v > 2:
+                raise ValueError('Control weights must be within -1 to 2 range')
         return v
+
     class Config:
         schema_extra = {
             "required": ["image", "control_model", "control_weight", "begin_step_percent", "end_step_percent"],
             "ui": {
                 "type_hints": {
                     "control_weight": "float",
+                    "control_model": "controlnet_model",
                     # "control_weight": "number",
                 }
             }
@@ -154,10 +153,10 @@ class ControlNetInvocation(BaseInvocation):
     type: Literal["controlnet"] = "controlnet"
     # Inputs
     image: ImageField = Field(default=None, description="The control image")
-    control_model: CONTROLNET_NAME_VALUES = Field(default="lllyasviel/sd-controlnet-canny",
+    control_model: ControlNetModelField = Field(default="lllyasviel/sd-controlnet-canny",
                                                   description="control model used")
     control_weight: Union[float, List[float]] = Field(default=1.0, description="The weight given to the ControlNet")
-    begin_step_percent: float = Field(default=0, ge=0, le=1,
+    begin_step_percent: float = Field(default=0, ge=-1, le=2,
                                       description="When the ControlNet is first applied (% of total steps)")
     end_step_percent: float = Field(default=1, ge=0, le=1,
                                     description="When the ControlNet is last applied (% of total steps)")
@@ -167,13 +166,14 @@ class ControlNetInvocation(BaseInvocation):
     class Config(InvocationConfig):
         schema_extra = {
             "ui": {
-                "tags": ["latents"],
+                "title": "ControlNet",
+                "tags": ["controlnet", "latents"],
                 "type_hints": {
-                  "model": "model",
-                  "control": "control",
-                  # "cfg_scale": "float",
-                  "cfg_scale": "number",
-                  "control_weight": "float",
+                    "model": "model",
+                    "control": "control",
+                    # "cfg_scale": "float",
+                    "cfg_scale": "number",
+                    "control_weight": "float",
                 }
             },
         }
@@ -200,6 +200,13 @@ class ImageProcessorInvocation(BaseInvocation, PILInvocationConfig):
     image: ImageField = Field(default=None, description="The image to process")
     # fmt: on
 
+    class Config(InvocationConfig):
+        schema_extra = {
+            "ui": {
+                "title": "Image Processor",
+                "tags": ["image", "processor"]
+            },
+        }
 
     def run_processor(self, image):
         # superclass just passes through image without processing
@@ -231,14 +238,15 @@ class ImageProcessorInvocation(BaseInvocation, PILInvocationConfig):
         return ImageOutput(
             image=processed_image_field,
             # width=processed_image.width,
-            width = image_dto.width,
+            width=image_dto.width,
             # height=processed_image.height,
-            height = image_dto.height,
+            height=image_dto.height,
             # mode=processed_image.mode,
         )
 
 
-class CannyImageProcessorInvocation(ImageProcessorInvocation, PILInvocationConfig):
+class CannyImageProcessorInvocation(
+        ImageProcessorInvocation, PILInvocationConfig):
     """Canny edge detection for ControlNet"""
     # fmt: off
     type: Literal["canny_image_processor"] = "canny_image_processor"
@@ -247,13 +255,23 @@ class CannyImageProcessorInvocation(ImageProcessorInvocation, PILInvocationConfi
     high_threshold: int = Field(default=200, ge=0, le=255, description="The high threshold of the Canny pixel gradient (0-255)")
     # fmt: on
 
+    class Config(InvocationConfig):
+        schema_extra = {
+            "ui": {
+                "title": "Canny Processor",
+                "tags": ["controlnet", "canny", "image", "processor"]
+            },
+        }
+
     def run_processor(self, image):
         canny_processor = CannyDetector()
-        processed_image = canny_processor(image, self.low_threshold, self.high_threshold)
+        processed_image = canny_processor(
+            image, self.low_threshold, self.high_threshold)
         return processed_image
 
 
-class HedImageProcessorInvocation(ImageProcessorInvocation, PILInvocationConfig):
+class HedImageProcessorInvocation(
+        ImageProcessorInvocation, PILInvocationConfig):
     """Applies HED edge detection to image"""
     # fmt: off
     type: Literal["hed_image_processor"] = "hed_image_processor"
@@ -265,6 +283,14 @@ class HedImageProcessorInvocation(ImageProcessorInvocation, PILInvocationConfig)
     scribble: bool = Field(default=False, description="Whether to use scribble mode")
     # fmt: on
 
+    class Config(InvocationConfig):
+        schema_extra = {
+            "ui": {
+                "title": "Softedge(HED) Processor",
+                "tags": ["controlnet", "softedge", "hed", "image", "processor"]
+            },
+        }
+
     def run_processor(self, image):
         hed_processor = HEDdetector.from_pretrained("lllyasviel/Annotators")
         processed_image = hed_processor(image,
@@ -277,7 +303,8 @@ class HedImageProcessorInvocation(ImageProcessorInvocation, PILInvocationConfig)
         return processed_image
 
 
-class LineartImageProcessorInvocation(ImageProcessorInvocation, PILInvocationConfig):
+class LineartImageProcessorInvocation(
+        ImageProcessorInvocation, PILInvocationConfig):
     """Applies line art processing to image"""
     # fmt: off
     type: Literal["lineart_image_processor"] = "lineart_image_processor"
@@ -287,16 +314,25 @@ class LineartImageProcessorInvocation(ImageProcessorInvocation, PILInvocationCon
     coarse: bool = Field(default=False, description="Whether to use coarse mode")
     # fmt: on
 
+    class Config(InvocationConfig):
+        schema_extra = {
+            "ui": {
+                "title": "Lineart Processor",
+                "tags": ["controlnet", "lineart", "image", "processor"]
+            },
+        }
+
     def run_processor(self, image):
-        lineart_processor = LineartDetector.from_pretrained("lllyasviel/Annotators")
-        processed_image = lineart_processor(image,
-                                            detect_resolution=self.detect_resolution,
-                                            image_resolution=self.image_resolution,
-                                            coarse=self.coarse)
+        lineart_processor = LineartDetector.from_pretrained(
+            "lllyasviel/Annotators")
+        processed_image = lineart_processor(
+            image, detect_resolution=self.detect_resolution,
+            image_resolution=self.image_resolution, coarse=self.coarse)
         return processed_image
 
 
-class LineartAnimeImageProcessorInvocation(ImageProcessorInvocation, PILInvocationConfig):
+class LineartAnimeImageProcessorInvocation(
+        ImageProcessorInvocation, PILInvocationConfig):
     """Applies line art anime processing to image"""
     # fmt: off
     type: Literal["lineart_anime_image_processor"] = "lineart_anime_image_processor"
@@ -305,8 +341,17 @@ class LineartAnimeImageProcessorInvocation(ImageProcessorInvocation, PILInvocati
     image_resolution: int = Field(default=512, ge=0, description="The pixel resolution for the output image")
     # fmt: on
 
+    class Config(InvocationConfig):
+        schema_extra = {
+            "ui": {
+                "title": "Lineart Anime Processor",
+                "tags": ["controlnet", "lineart", "anime", "image", "processor"]
+            },
+        }
+
     def run_processor(self, image):
-        processor = LineartAnimeDetector.from_pretrained("lllyasviel/Annotators")
+        processor = LineartAnimeDetector.from_pretrained(
+            "lllyasviel/Annotators")
         processed_image = processor(image,
                                     detect_resolution=self.detect_resolution,
                                     image_resolution=self.image_resolution,
@@ -314,7 +359,8 @@ class LineartAnimeImageProcessorInvocation(ImageProcessorInvocation, PILInvocati
         return processed_image
 
 
-class OpenposeImageProcessorInvocation(ImageProcessorInvocation, PILInvocationConfig):
+class OpenposeImageProcessorInvocation(
+        ImageProcessorInvocation, PILInvocationConfig):
     """Applies Openpose processing to image"""
     # fmt: off
     type: Literal["openpose_image_processor"] = "openpose_image_processor"
@@ -324,17 +370,26 @@ class OpenposeImageProcessorInvocation(ImageProcessorInvocation, PILInvocationCo
     image_resolution: int = Field(default=512, ge=0, description="The pixel resolution for the output image")
     # fmt: on
 
+    class Config(InvocationConfig):
+        schema_extra = {
+            "ui": {
+                "title": "Openpose Processor",
+                "tags": ["controlnet", "openpose", "image", "processor"]
+            },
+        }
+
     def run_processor(self, image):
-        openpose_processor = OpenposeDetector.from_pretrained("lllyasviel/Annotators")
-        processed_image = openpose_processor(image,
-                                             detect_resolution=self.detect_resolution,
-                                             image_resolution=self.image_resolution,
-                                             hand_and_face=self.hand_and_face,
-                                             )
+        openpose_processor = OpenposeDetector.from_pretrained(
+            "lllyasviel/Annotators")
+        processed_image = openpose_processor(
+            image, detect_resolution=self.detect_resolution,
+            image_resolution=self.image_resolution,
+            hand_and_face=self.hand_and_face,)
         return processed_image
 
 
-class MidasDepthImageProcessorInvocation(ImageProcessorInvocation, PILInvocationConfig):
+class MidasDepthImageProcessorInvocation(
+        ImageProcessorInvocation, PILInvocationConfig):
     """Applies Midas depth processing to image"""
     # fmt: off
     type: Literal["midas_depth_image_processor"] = "midas_depth_image_processor"
@@ -345,6 +400,14 @@ class MidasDepthImageProcessorInvocation(ImageProcessorInvocation, PILInvocation
     # depth_and_normal: bool = Field(default=False, description="whether to use depth and normal mode")
     # fmt: on
 
+    class Config(InvocationConfig):
+        schema_extra = {
+            "ui": {
+                "title": "Midas (Depth) Processor",
+                "tags": ["controlnet", "midas", "depth", "image", "processor"]
+            },
+        }
+
     def run_processor(self, image):
         midas_processor = MidasDetector.from_pretrained("lllyasviel/Annotators")
         processed_image = midas_processor(image,
@@ -356,7 +419,8 @@ class MidasDepthImageProcessorInvocation(ImageProcessorInvocation, PILInvocation
         return processed_image
 
 
-class NormalbaeImageProcessorInvocation(ImageProcessorInvocation, PILInvocationConfig):
+class NormalbaeImageProcessorInvocation(
+        ImageProcessorInvocation, PILInvocationConfig):
     """Applies NormalBae processing to image"""
     # fmt: off
     type: Literal["normalbae_image_processor"] = "normalbae_image_processor"
@@ -365,15 +429,25 @@ class NormalbaeImageProcessorInvocation(ImageProcessorInvocation, PILInvocationC
     image_resolution: int = Field(default=512, ge=0, description="The pixel resolution for the output image")
     # fmt: on
 
+    class Config(InvocationConfig):
+        schema_extra = {
+            "ui": {
+                "title": "Normal BAE Processor",
+                "tags": ["controlnet", "normal", "bae", "image", "processor"]
+            },
+        }
+
     def run_processor(self, image):
-        normalbae_processor = NormalBaeDetector.from_pretrained("lllyasviel/Annotators")
-        processed_image = normalbae_processor(image,
-                                              detect_resolution=self.detect_resolution,
-                                              image_resolution=self.image_resolution)
+        normalbae_processor = NormalBaeDetector.from_pretrained(
+            "lllyasviel/Annotators")
+        processed_image = normalbae_processor(
+            image, detect_resolution=self.detect_resolution,
+            image_resolution=self.image_resolution)
         return processed_image
 
 
-class MlsdImageProcessorInvocation(ImageProcessorInvocation, PILInvocationConfig):
+class MlsdImageProcessorInvocation(
+        ImageProcessorInvocation, PILInvocationConfig):
     """Applies MLSD processing to image"""
     # fmt: off
     type: Literal["mlsd_image_processor"] = "mlsd_image_processor"
@@ -384,17 +458,25 @@ class MlsdImageProcessorInvocation(ImageProcessorInvocation, PILInvocationConfig
     thr_d: float = Field(default=0.1, ge=0, description="MLSD parameter `thr_d`")
     # fmt: on
 
+    class Config(InvocationConfig):
+        schema_extra = {
+            "ui": {
+                "title": "MLSD Processor",
+                "tags": ["controlnet", "mlsd", "image", "processor"]
+            },
+        }
+
     def run_processor(self, image):
         mlsd_processor = MLSDdetector.from_pretrained("lllyasviel/Annotators")
-        processed_image = mlsd_processor(image,
-                                         detect_resolution=self.detect_resolution,
-                                         image_resolution=self.image_resolution,
-                                         thr_v=self.thr_v,
-                                         thr_d=self.thr_d)
+        processed_image = mlsd_processor(
+            image, detect_resolution=self.detect_resolution,
+            image_resolution=self.image_resolution, thr_v=self.thr_v,
+            thr_d=self.thr_d)
         return processed_image
 
 
-class PidiImageProcessorInvocation(ImageProcessorInvocation, PILInvocationConfig):
+class PidiImageProcessorInvocation(
+        ImageProcessorInvocation, PILInvocationConfig):
     """Applies PIDI processing to image"""
     # fmt: off
     type: Literal["pidi_image_processor"] = "pidi_image_processor"
@@ -405,17 +487,26 @@ class PidiImageProcessorInvocation(ImageProcessorInvocation, PILInvocationConfig
     scribble: bool = Field(default=False, description="Whether to use scribble mode")
     # fmt: on
 
+    class Config(InvocationConfig):
+        schema_extra = {
+            "ui": {
+                "title": "PIDI Processor",
+                "tags": ["controlnet", "pidi", "image", "processor"]
+            },
+        }
+
     def run_processor(self, image):
-        pidi_processor = PidiNetDetector.from_pretrained("lllyasviel/Annotators")
-        processed_image = pidi_processor(image,
-                                         detect_resolution=self.detect_resolution,
-                                         image_resolution=self.image_resolution,
-                                         safe=self.safe,
-                                         scribble=self.scribble)
+        pidi_processor = PidiNetDetector.from_pretrained(
+            "lllyasviel/Annotators")
+        processed_image = pidi_processor(
+            image, detect_resolution=self.detect_resolution,
+            image_resolution=self.image_resolution, safe=self.safe,
+            scribble=self.scribble)
         return processed_image
 
 
-class ContentShuffleImageProcessorInvocation(ImageProcessorInvocation, PILInvocationConfig):
+class ContentShuffleImageProcessorInvocation(
+        ImageProcessorInvocation, PILInvocationConfig):
     """Applies content shuffle processing to image"""
     # fmt: off
     type: Literal["content_shuffle_image_processor"] = "content_shuffle_image_processor"
@@ -427,6 +518,14 @@ class ContentShuffleImageProcessorInvocation(ImageProcessorInvocation, PILInvoca
     f: Optional[int] = Field(default=256, ge=0, description="Content shuffle `f` parameter")
     # fmt: on
 
+    class Config(InvocationConfig):
+        schema_extra = {
+            "ui": {
+                "title": "Content Shuffle Processor",
+                "tags": ["controlnet", "contentshuffle", "image", "processor"]
+            },
+        }
+
     def run_processor(self, image):
         content_shuffle_processor = ContentShuffleDetector()
         processed_image = content_shuffle_processor(image,
@@ -440,19 +539,30 @@ class ContentShuffleImageProcessorInvocation(ImageProcessorInvocation, PILInvoca
 
 
 # should work with controlnet_aux >= 0.0.4 and timm <= 0.6.13
-class ZoeDepthImageProcessorInvocation(ImageProcessorInvocation, PILInvocationConfig):
+class ZoeDepthImageProcessorInvocation(
+        ImageProcessorInvocation, PILInvocationConfig):
     """Applies Zoe depth processing to image"""
     # fmt: off
     type: Literal["zoe_depth_image_processor"] = "zoe_depth_image_processor"
     # fmt: on
 
+    class Config(InvocationConfig):
+        schema_extra = {
+            "ui": {
+                "title": "Zoe (Depth) Processor",
+                "tags": ["controlnet", "zoe", "depth", "image", "processor"]
+            },
+        }
+
     def run_processor(self, image):
-        zoe_depth_processor = ZoeDetector.from_pretrained("lllyasviel/Annotators")
+        zoe_depth_processor = ZoeDetector.from_pretrained(
+            "lllyasviel/Annotators")
         processed_image = zoe_depth_processor(image)
         return processed_image
 
 
-class MediapipeFaceProcessorInvocation(ImageProcessorInvocation, PILInvocationConfig):
+class MediapipeFaceProcessorInvocation(
+        ImageProcessorInvocation, PILInvocationConfig):
     """Applies mediapipe face processing to image"""
     # fmt: off
     type: Literal["mediapipe_face_processor"] = "mediapipe_face_processor"
@@ -461,16 +571,27 @@ class MediapipeFaceProcessorInvocation(ImageProcessorInvocation, PILInvocationCo
     min_confidence: float = Field(default=0.5, ge=0, le=1, description="Minimum confidence for face detection")
     # fmt: on
 
+    class Config(InvocationConfig):
+        schema_extra = {
+            "ui": {
+                "title": "Mediapipe Processor",
+                "tags": ["controlnet", "mediapipe", "image", "processor"]
+            },
+        }
+
     def run_processor(self, image):
         # MediaPipeFaceDetector throws an error if image has alpha channel
         #     so convert to RGB if needed
         if image.mode == 'RGBA':
             image = image.convert('RGB')
         mediapipe_face_processor = MediapipeFaceDetector()
-        processed_image = mediapipe_face_processor(image, max_faces=self.max_faces, min_confidence=self.min_confidence)
+        processed_image = mediapipe_face_processor(
+            image, max_faces=self.max_faces, min_confidence=self.min_confidence)
         return processed_image
 
-class LeresImageProcessorInvocation(ImageProcessorInvocation, PILInvocationConfig):
+
+class LeresImageProcessorInvocation(
+        ImageProcessorInvocation, PILInvocationConfig):
     """Applies leres processing to image"""
     # fmt: off
     type: Literal["leres_image_processor"] = "leres_image_processor"
@@ -482,18 +603,25 @@ class LeresImageProcessorInvocation(ImageProcessorInvocation, PILInvocationConfi
     image_resolution: int = Field(default=512, ge=0, description="The pixel resolution for the output image")
     # fmt: on
 
+    class Config(InvocationConfig):
+        schema_extra = {
+            "ui": {
+                "title": "Leres (Depth) Processor",
+                "tags": ["controlnet", "leres", "depth", "image", "processor"]
+            },
+        }
+
     def run_processor(self, image):
         leres_processor = LeresDetector.from_pretrained("lllyasviel/Annotators")
-        processed_image = leres_processor(image,
-                                          thr_a=self.thr_a,
-                                          thr_b=self.thr_b,
-                                          boost=self.boost,
-                                          detect_resolution=self.detect_resolution,
-                                          image_resolution=self.image_resolution)
+        processed_image = leres_processor(
+            image, thr_a=self.thr_a, thr_b=self.thr_b, boost=self.boost,
+            detect_resolution=self.detect_resolution,
+            image_resolution=self.image_resolution)
         return processed_image
 
 
-class TileResamplerProcessorInvocation(ImageProcessorInvocation, PILInvocationConfig):
+class TileResamplerProcessorInvocation(
+        ImageProcessorInvocation, PILInvocationConfig):
 
     # fmt: off
     type: Literal["tile_image_processor"] = "tile_image_processor"
@@ -502,6 +630,14 @@ class TileResamplerProcessorInvocation(ImageProcessorInvocation, PILInvocationCo
     down_sampling_rate: float = Field(default=1.0, ge=1.0, le=8.0, description="Down sampling rate")
     # fmt: on
 
+    class Config(InvocationConfig):
+        schema_extra = {
+            "ui": {
+                "title": "Tile Resample Processor",
+                "tags": ["controlnet", "tile", "resample", "image", "processor"]
+            },
+        }
+
     # tile_resample copied from sd-webui-controlnet/scripts/processor.py
     def tile_resample(self,
                       np_img: np.ndarray,
@@ -520,28 +656,33 @@ class TileResamplerProcessorInvocation(ImageProcessorInvocation, PILInvocationCo
     def run_processor(self, img):
         np_img = np.array(img, dtype=np.uint8)
         processed_np_image = self.tile_resample(np_img,
-                                                #res=self.tile_size,
+                                                # res=self.tile_size,
                                                 down_sampling_rate=self.down_sampling_rate
                                                 )
         processed_image = Image.fromarray(processed_np_image)
         return processed_image
 
 
-
-
-class SegmentAnythingProcessorInvocation(ImageProcessorInvocation, PILInvocationConfig):
+class SegmentAnythingProcessorInvocation(
+        ImageProcessorInvocation, PILInvocationConfig):
     """Applies segment anything processing to image"""
     # fmt: off
     type: Literal["segment_anything_processor"] = "segment_anything_processor"
     # fmt: on
 
+    class Config(InvocationConfig):
+        schema_extra = {"ui": {"title": "Segment Anything Processor", "tags": [
+            "controlnet", "segment", "anything", "sam", "image", "processor"]}, }
+
     def run_processor(self, image):
         # segment_anything_processor = SamDetector.from_pretrained("ybelkada/segment-anything", subfolder="checkpoints")
-        segment_anything_processor = SamDetectorReproducibleColors.from_pretrained("ybelkada/segment-anything", subfolder="checkpoints")
+        segment_anything_processor = SamDetectorReproducibleColors.from_pretrained(
+            "ybelkada/segment-anything", subfolder="checkpoints")
         np_img = np.array(image, dtype=np.uint8)
         processed_image = segment_anything_processor(np_img)
         return processed_image
 
+
 class SamDetectorReproducibleColors(SamDetector):
 
     # overriding SamDetector.show_anns() method to use reproducible colors for segmentation image
@@ -553,7 +694,8 @@ class SamDetectorReproducibleColors(SamDetector):
             return
         sorted_anns = sorted(anns, key=(lambda x: x['area']), reverse=True)
         h, w = anns[0]['segmentation'].shape
-        final_img = Image.fromarray(np.zeros((h, w, 3), dtype=np.uint8), mode="RGB")
+        final_img = Image.fromarray(
+            np.zeros((h, w, 3), dtype=np.uint8), mode="RGB")
         palette = ade_palette()
         for i, ann in enumerate(sorted_anns):
             m = ann['segmentation']
@@ -561,5 +703,8 @@ class SamDetectorReproducibleColors(SamDetector):
             # doing modulo just in case number of annotated regions exceeds number of colors in palette
             ann_color = palette[i % len(palette)]
             img[:, :] = ann_color
-            final_img.paste(Image.fromarray(img, mode="RGB"), (0, 0), Image.fromarray(np.uint8(m * 255)))
+            final_img.paste(
+                Image.fromarray(img, mode="RGB"),
+                (0, 0),
+                Image.fromarray(np.uint8(m * 255)))
         return np.array(final_img, dtype=np.uint8)

invokeai/app/invocations/cv.py

@@ -35,6 +35,14 @@ class CvInpaintInvocation(BaseInvocation, CvInvocationConfig):
     mask: ImageField = Field(default=None, description="The mask to use when inpainting")
     # fmt: on
 
+    class Config(InvocationConfig):
+        schema_extra = {
+            "ui": {
+                "title": "OpenCV Inpaint",
+                "tags": ["opencv", "inpaint"]
+            },
+        }
+
     def invoke(self, context: InvocationContext) -> ImageOutput:
         image = context.services.images.get_pil_image(self.image.image_name)
         mask = context.services.images.get_pil_image(self.mask.image_name)

invokeai/app/invocations/generate.py

@@ -130,6 +130,7 @@ class InpaintInvocation(BaseInvocation):
         schema_extra = {
             "ui": {
                 "tags": ["stable-diffusion", "image"],
+                "title": "Inpaint"
             },
         }
 
@@ -146,48 +147,54 @@ class InpaintInvocation(BaseInvocation):
             source_node_id=source_node_id,
         )
 
-    def get_conditioning(self, context):
-        c, extra_conditioning_info = context.services.latents.get(self.positive_conditioning.conditioning_name)
-        uc, _ = context.services.latents.get(self.negative_conditioning.conditioning_name)
+    def get_conditioning(self, context, unet):
+        positive_cond_data = context.services.latents.get(self.positive_conditioning.conditioning_name)
+        c = positive_cond_data.conditionings[0].embeds.to(device=unet.device, dtype=unet.dtype)
+        extra_conditioning_info = positive_cond_data.conditionings[0].extra_conditioning
+
+        negative_cond_data = context.services.latents.get(self.negative_conditioning.conditioning_name)
+        uc = negative_cond_data.conditionings[0].embeds.to(device=unet.device, dtype=unet.dtype)
 
         return (uc, c, extra_conditioning_info)
 
     @contextmanager
     def load_model_old_way(self, context, scheduler):
-        unet_info = context.services.model_manager.get_model(**self.unet.unet.dict())
-        vae_info = context.services.model_manager.get_model(**self.vae.vae.dict())
+        def _lora_loader():
+            for lora in self.unet.loras:
+                lora_info = context.services.model_manager.get_model(
+                    **lora.dict(exclude={"weight"}), context=context,)
+                yield (lora_info.context.model, lora.weight)
+                del lora_info
+            return
+        
+        unet_info = context.services.model_manager.get_model(**self.unet.unet.dict(), context=context,)
+        vae_info = context.services.model_manager.get_model(**self.vae.vae.dict(), context=context,)
 
-        #unet = unet_info.context.model
-        #vae = vae_info.context.model
+        with vae_info as vae,\
+                ModelPatcher.apply_lora_unet(unet_info.context.model, _lora_loader()),\
+                unet_info as unet:
 
-        with ExitStack() as stack:
-            loras = [(stack.enter_context(context.services.model_manager.get_model(**lora.dict(exclude={"weight"}))), lora.weight) for lora in self.unet.loras]
+            device = context.services.model_manager.mgr.cache.execution_device
+            dtype = context.services.model_manager.mgr.cache.precision
 
-            with vae_info as vae,\
-                 unet_info as unet,\
-                 ModelPatcher.apply_lora_unet(unet, loras):
+            pipeline = StableDiffusionGeneratorPipeline(
+                vae=vae,
+                text_encoder=None,
+                tokenizer=None,
+                unet=unet,
+                scheduler=scheduler,
+                safety_checker=None,
+                feature_extractor=None,
+                requires_safety_checker=False,
+                precision="float16" if dtype == torch.float16 else "float32",
+                execution_device=device,
+            )
 
-                device = context.services.model_manager.mgr.cache.execution_device
-                dtype = context.services.model_manager.mgr.cache.precision
-
-                pipeline = StableDiffusionGeneratorPipeline(
-                    vae=vae,
-                    text_encoder=None,
-                    tokenizer=None,
-                    unet=unet,
-                    scheduler=scheduler,
-                    safety_checker=None,
-                    feature_extractor=None,
-                    requires_safety_checker=False,
-                    precision="float16" if dtype == torch.float16 else "float32",
-                    execution_device=device,
-                )
-
-                yield OldModelInfo(
-                    name=self.unet.unet.model_name,
-                    hash="<NO-HASH>",
-                    model=pipeline,
-                )
+            yield OldModelInfo(
+                name=self.unet.unet.model_name,
+                hash="<NO-HASH>",
+                model=pipeline,
+            )
 
     def invoke(self, context: InvocationContext) -> ImageOutput:
         image = (
@@ -207,7 +214,6 @@ class InpaintInvocation(BaseInvocation):
         )
         source_node_id = graph_execution_state.prepared_source_mapping[self.id]
 
-        conditioning = self.get_conditioning(context)
         scheduler = get_scheduler(
             context=context,
             scheduler_info=self.unet.scheduler,
@@ -215,6 +221,8 @@ class InpaintInvocation(BaseInvocation):
         )
 
         with self.load_model_old_way(context, scheduler) as model:
+            conditioning = self.get_conditioning(context, model.context.model.unet)
+
             outputs = Inpaint(model).generate(
                 conditioning=conditioning,
                 scheduler=scheduler,
@@ -226,21 +234,21 @@ class InpaintInvocation(BaseInvocation):
                 ),  # Shorthand for passing all of the parameters above manually
             )
 
-        # Outputs is an infinite iterator that will return a new InvokeAIGeneratorOutput object
-        # each time it is called. We only need the first one.
-        generator_output = next(outputs)
+            # Outputs is an infinite iterator that will return a new InvokeAIGeneratorOutput object
+            # each time it is called. We only need the first one.
+            generator_output = next(outputs)
 
-        image_dto = context.services.images.create(
-            image=generator_output.image,
-            image_origin=ResourceOrigin.INTERNAL,
-            image_category=ImageCategory.GENERAL,
-            session_id=context.graph_execution_state_id,
-            node_id=self.id,
-            is_intermediate=self.is_intermediate,
-        )
+            image_dto = context.services.images.create(
+                image=generator_output.image,
+                image_origin=ResourceOrigin.INTERNAL,
+                image_category=ImageCategory.GENERAL,
+                session_id=context.graph_execution_state_id,
+                node_id=self.id,
+                is_intermediate=self.is_intermediate,
+            )
 
-        return ImageOutput(
-            image=ImageField(image_name=image_dto.image_name),
-            width=image_dto.width,
-            height=image_dto.height,
-        )
+            return ImageOutput(
+                image=ImageField(image_name=image_dto.image_name),
+                width=image_dto.width,
+                height=image_dto.height,
+            )

invokeai/app/invocations/image.py

@@ -5,6 +5,7 @@ from typing import Literal, Optional
 import numpy
 from PIL import Image, ImageFilter, ImageOps, ImageChops
 from pydantic import BaseModel, Field
+from typing import Union
 
 from ..models.image import ImageCategory, ImageField, ResourceOrigin
 from .baseinvocation import (
@@ -70,6 +71,15 @@ class LoadImageInvocation(BaseInvocation):
         default=None, description="The image to load"
     )
     # fmt: on
+
+    class Config(InvocationConfig):
+        schema_extra = {
+            "ui": {
+                "title": "Load Image",
+                "tags": ["image", "load"]
+            },
+        }
+
     def invoke(self, context: InvocationContext) -> ImageOutput:
         image = context.services.images.get_pil_image(self.image.image_name)
 
@@ -90,6 +100,14 @@ class ShowImageInvocation(BaseInvocation):
         default=None, description="The image to show"
     )
 
+    class Config(InvocationConfig):
+        schema_extra = {
+            "ui": {
+                "title": "Show Image",
+                "tags": ["image", "show"]
+            },
+        }
+
     def invoke(self, context: InvocationContext) -> ImageOutput:
         image = context.services.images.get_pil_image(self.image.image_name)
         if image:
@@ -118,6 +136,14 @@ class ImageCropInvocation(BaseInvocation, PILInvocationConfig):
     height: int = Field(default=512, gt=0, description="The height of the crop rectangle")
     # fmt: on
 
+    class Config(InvocationConfig):
+        schema_extra = {
+            "ui": {
+                "title": "Crop Image",
+                "tags": ["image", "crop"]
+            },
+        }
+
     def invoke(self, context: InvocationContext) -> ImageOutput:
         image = context.services.images.get_pil_image(self.image.image_name)
 
@@ -156,6 +182,14 @@ class ImagePasteInvocation(BaseInvocation, PILInvocationConfig):
     y:                     int = Field(default=0, description="The top y coordinate at which to paste the image")
     # fmt: on
 
+    class Config(InvocationConfig):
+        schema_extra = {
+            "ui": {
+                "title": "Paste Image",
+                "tags": ["image", "paste"]
+            },
+        }
+
     def invoke(self, context: InvocationContext) -> ImageOutput:
         base_image = context.services.images.get_pil_image(self.base_image.image_name)
         image = context.services.images.get_pil_image(self.image.image_name)
@@ -206,6 +240,14 @@ class MaskFromAlphaInvocation(BaseInvocation, PILInvocationConfig):
     invert:      bool = Field(default=False, description="Whether or not to invert the mask")
     # fmt: on
 
+    class Config(InvocationConfig):
+        schema_extra = {
+            "ui": {
+                "title": "Mask From Alpha",
+                "tags": ["image", "mask", "alpha"]
+            },
+        }
+
     def invoke(self, context: InvocationContext) -> MaskOutput:
         image = context.services.images.get_pil_image(self.image.image_name)
 
@@ -240,6 +282,14 @@ class ImageMultiplyInvocation(BaseInvocation, PILInvocationConfig):
     image2: Optional[ImageField]  = Field(default=None, description="The second image to multiply")
     # fmt: on
 
+    class Config(InvocationConfig):
+        schema_extra = {
+            "ui": {
+                "title": "Multiply Images",
+                "tags": ["image", "multiply"]
+            },
+        }
+
     def invoke(self, context: InvocationContext) -> ImageOutput:
         image1 = context.services.images.get_pil_image(self.image1.image_name)
         image2 = context.services.images.get_pil_image(self.image2.image_name)
@@ -276,6 +326,14 @@ class ImageChannelInvocation(BaseInvocation, PILInvocationConfig):
     channel: IMAGE_CHANNELS  = Field(default="A", description="The channel to get")
     # fmt: on
 
+    class Config(InvocationConfig):
+        schema_extra = {
+            "ui": {
+                "title": "Image Channel",
+                "tags": ["image", "channel"]
+            },
+        }
+
     def invoke(self, context: InvocationContext) -> ImageOutput:
         image = context.services.images.get_pil_image(self.image.image_name)
 
@@ -311,6 +369,14 @@ class ImageConvertInvocation(BaseInvocation, PILInvocationConfig):
     mode: IMAGE_MODES  = Field(default="L", description="The mode to convert to")
     # fmt: on
 
+    class Config(InvocationConfig):
+        schema_extra = {
+            "ui": {
+                "title": "Convert Image",
+                "tags": ["image", "convert"]
+            },
+        }
+
     def invoke(self, context: InvocationContext) -> ImageOutput:
         image = context.services.images.get_pil_image(self.image.image_name)
 
@@ -344,6 +410,14 @@ class ImageBlurInvocation(BaseInvocation, PILInvocationConfig):
     blur_type: Literal["gaussian", "box"] = Field(default="gaussian", description="The type of blur")
     # fmt: on
 
+    class Config(InvocationConfig):
+        schema_extra = {
+            "ui": {
+                "title": "Blur Image",
+                "tags": ["image", "blur"]
+            },
+        }
+
     def invoke(self, context: InvocationContext) -> ImageOutput:
         image = context.services.images.get_pil_image(self.image.image_name)
 
@@ -398,11 +472,19 @@ class ImageResizeInvocation(BaseInvocation, PILInvocationConfig):
 
     # Inputs
     image: Optional[ImageField]  = Field(default=None, description="The image to resize")
-    width:                         int = Field(ge=64, multiple_of=8, description="The width to resize to (px)")
-    height:                        int = Field(ge=64, multiple_of=8, description="The height to resize to (px)")
+    width:                         Union[int, None] = Field(ge=64, multiple_of=8, description="The width to resize to (px)")
+    height:                        Union[int, None] = Field(ge=64, multiple_of=8, description="The height to resize to (px)")
     resample_mode:  PIL_RESAMPLING_MODES = Field(default="bicubic", description="The resampling mode")
     # fmt: on
 
+    class Config(InvocationConfig):
+        schema_extra = {
+            "ui": {
+                "title": "Resize Image",
+                "tags": ["image", "resize"]
+            },
+        }
+
     def invoke(self, context: InvocationContext) -> ImageOutput:
         image = context.services.images.get_pil_image(self.image.image_name)
 
@@ -436,11 +518,19 @@ class ImageScaleInvocation(BaseInvocation, PILInvocationConfig):
     type: Literal["img_scale"] = "img_scale"
 
     # Inputs
-    image:       Optional[ImageField] = Field(default=None, description="The image to scale")
-    scale_factor:                  float = Field(gt=0, description="The factor by which to scale the image")
+    image:          Optional[ImageField] = Field(default=None, description="The image to scale")
+    scale_factor:        Optional[float] = Field(default=2.0, gt=0, description="The factor by which to scale the image")
     resample_mode:  PIL_RESAMPLING_MODES = Field(default="bicubic", description="The resampling mode")
     # fmt: on
 
+    class Config(InvocationConfig):
+        schema_extra = {
+            "ui": {
+                "title": "Scale Image",
+                "tags": ["image", "scale"]
+            },
+        }
+
     def invoke(self, context: InvocationContext) -> ImageOutput:
         image = context.services.images.get_pil_image(self.image.image_name)
 
@@ -481,6 +571,14 @@ class ImageLerpInvocation(BaseInvocation, PILInvocationConfig):
     max: int = Field(default=255, ge=0, le=255, description="The maximum output value")
     # fmt: on
 
+    class Config(InvocationConfig):
+        schema_extra = {
+            "ui": {
+                "title": "Image Linear Interpolation",
+                "tags": ["image", "linear", "interpolation", "lerp"]
+            },
+        }
+
     def invoke(self, context: InvocationContext) -> ImageOutput:
         image = context.services.images.get_pil_image(self.image.image_name)
 
@@ -517,6 +615,14 @@ class ImageInverseLerpInvocation(BaseInvocation, PILInvocationConfig):
     max: int = Field(default=255, ge=0, le=255, description="The maximum input value")
     # fmt: on
 
+    class Config(InvocationConfig):
+        schema_extra = {
+            "ui": {
+                "title": "Image Inverse Linear Interpolation",
+                "tags": ["image", "linear", "interpolation", "inverse"]
+            },
+        }
+
     def invoke(self, context: InvocationContext) -> ImageOutput:
         image = context.services.images.get_pil_image(self.image.image_name)
 

invokeai/app/invocations/infill.py

@@ -14,6 +14,7 @@ from invokeai.backend.image_util.patchmatch import PatchMatch
 from ..models.image import ColorField, ImageCategory, ImageField, ResourceOrigin
 from .baseinvocation import (
     BaseInvocation,
+    InvocationConfig,
     InvocationContext,
 )
 
@@ -133,6 +134,14 @@ class InfillColorInvocation(BaseInvocation):
         description="The color to use to infill",
     )
 
+    class Config(InvocationConfig):
+        schema_extra = {
+            "ui": {
+                "title": "Color Infill",
+                "tags": ["image", "inpaint", "color", "infill"]
+            },
+        }
+
     def invoke(self, context: InvocationContext) -> ImageOutput:
         image = context.services.images.get_pil_image(self.image.image_name)
 
@@ -173,6 +182,14 @@ class InfillTileInvocation(BaseInvocation):
         default_factory=get_random_seed,
     )
 
+    class Config(InvocationConfig):
+        schema_extra = {
+            "ui": {
+                "title": "Tile Infill",
+                "tags": ["image", "inpaint", "tile", "infill"]
+            },
+        }
+
     def invoke(self, context: InvocationContext) -> ImageOutput:
         image = context.services.images.get_pil_image(self.image.image_name)
 
@@ -206,6 +223,14 @@ class InfillPatchMatchInvocation(BaseInvocation):
         default=None, description="The image to infill"
     )
 
+    class Config(InvocationConfig):
+        schema_extra = {
+            "ui": {
+                "title": "Patch Match Infill",
+                "tags": ["image", "inpaint", "patchmatch", "infill"]
+            },
+        }
+
     def invoke(self, context: InvocationContext) -> ImageOutput:
         image = context.services.images.get_pil_image(self.image.image_name)
 

invokeai/app/invocations/latent.py

@@ -1,5 +1,6 @@
 # Copyright (c) 2023 Kyle Schouviller (https://github.com/kyle0654)
 
+from contextlib import ExitStack
 from typing import List, Literal, Optional, Union
 
 import einops
@@ -9,9 +10,10 @@ from diffusers.image_processor import VaeImageProcessor
 from diffusers.schedulers import SchedulerMixin as Scheduler
 from pydantic import BaseModel, Field, validator
 
+from invokeai.app.invocations.metadata import CoreMetadata
 from invokeai.app.util.step_callback import stable_diffusion_step_callback
+from invokeai.backend.model_management.models.base import ModelType
 
-from ..models.image import ImageCategory, ImageField, ResourceOrigin
 from ...backend.model_management.lora import ModelPatcher
 from ...backend.stable_diffusion import PipelineIntermediateState
 from ...backend.stable_diffusion.diffusers_pipeline import (
@@ -20,7 +22,9 @@ from ...backend.stable_diffusion.diffusers_pipeline import (
 from ...backend.stable_diffusion.diffusion.shared_invokeai_diffusion import \
     PostprocessingSettings
 from ...backend.stable_diffusion.schedulers import SCHEDULER_MAP
-from ...backend.util.devices import torch_dtype
+from ...backend.util.devices import choose_torch_device, torch_dtype
+from ...backend.model_management import ModelPatcher
+from ..models.image import ImageCategory, ImageField, ResourceOrigin
 from .baseinvocation import (BaseInvocation, BaseInvocationOutput,
                              InvocationConfig, InvocationContext)
 from .compel import ConditioningField
@@ -28,6 +32,13 @@ from .controlnet_image_processors import ControlField
 from .image import ImageOutput
 from .model import ModelInfo, UNetField, VaeField
 
+from diffusers.models.attention_processor import (
+    AttnProcessor2_0,
+    LoRAAttnProcessor2_0,
+    LoRAXFormersAttnProcessor,
+    XFormersAttnProcessor,
+)
+
 
 class LatentsField(BaseModel):
     """A latents field used for passing latents between invocations"""
@@ -70,16 +81,21 @@ def get_scheduler(
     scheduler_name: str,
 ) -> Scheduler:
     scheduler_class, scheduler_extra_config = SCHEDULER_MAP.get(
-        scheduler_name, SCHEDULER_MAP['ddim'])
+        scheduler_name, SCHEDULER_MAP['ddim']
+    )
     orig_scheduler_info = context.services.model_manager.get_model(
-        **scheduler_info.dict())
+        **scheduler_info.dict(), context=context,
+    )
     with orig_scheduler_info as orig_scheduler:
         scheduler_config = orig_scheduler.config
 
     if "_backup" in scheduler_config:
         scheduler_config = scheduler_config["_backup"]
-    scheduler_config = {**scheduler_config, **
-                        scheduler_extra_config, "_backup": scheduler_config}
+    scheduler_config = {
+        **scheduler_config,
+        **scheduler_extra_config,
+        "_backup": scheduler_config,
+    }
     scheduler = scheduler_class.from_config(scheduler_config)
 
     # hack copied over from generate.py
@@ -124,6 +140,7 @@ class TextToLatentsInvocation(BaseInvocation):
     class Config(InvocationConfig):
         schema_extra = {
             "ui": {
+                "title": "Text To Latents",
                 "tags": ["latents"],
                 "type_hints": {
                     "model": "model",
@@ -136,8 +153,11 @@ class TextToLatentsInvocation(BaseInvocation):
 
     # TODO: pass this an emitter method or something? or a session for dispatching?
     def dispatch_progress(
-            self, context: InvocationContext, source_node_id: str,
-            intermediate_state: PipelineIntermediateState) -> None:
+        self,
+        context: InvocationContext,
+        source_node_id: str,
+        intermediate_state: PipelineIntermediateState,
+    ) -> None:
         stable_diffusion_step_callback(
             context=context,
             intermediate_state=intermediate_state,
@@ -146,11 +166,17 @@ class TextToLatentsInvocation(BaseInvocation):
         )
 
     def get_conditioning_data(
-            self, context: InvocationContext, scheduler) -> ConditioningData:
-        c, extra_conditioning_info = context.services.latents.get(
-            self.positive_conditioning.conditioning_name)
-        uc, _ = context.services.latents.get(
-            self.negative_conditioning.conditioning_name)
+        self,
+        context: InvocationContext,
+        scheduler,
+        unet,
+    ) -> ConditioningData:
+        positive_cond_data = context.services.latents.get(self.positive_conditioning.conditioning_name)
+        c = positive_cond_data.conditionings[0].embeds.to(device=unet.device, dtype=unet.dtype)
+        extra_conditioning_info = positive_cond_data.conditionings[0].extra_conditioning
+
+        negative_cond_data = context.services.latents.get(self.negative_conditioning.conditioning_name)
+        uc = negative_cond_data.conditionings[0].embeds.to(device=unet.device, dtype=unet.dtype)
 
         conditioning_data = ConditioningData(
             unconditioned_embeddings=uc,
@@ -172,12 +198,15 @@ class TextToLatentsInvocation(BaseInvocation):
             eta=0.0,  # ddim_eta
 
             # for ancestral and sde schedulers
-            generator=torch.Generator(device=uc.device).manual_seed(0),
+            generator=torch.Generator(device=unet.device).manual_seed(0),
         )
         return conditioning_data
 
     def create_pipeline(
-            self, unet, scheduler) -> StableDiffusionGeneratorPipeline:
+        self,
+        unet,
+        scheduler,
+    ) -> StableDiffusionGeneratorPipeline:
         # TODO:
         # configure_model_padding(
         #    unet,
@@ -212,6 +241,7 @@ class TextToLatentsInvocation(BaseInvocation):
         model: StableDiffusionGeneratorPipeline,
         control_input: List[ControlField],
         latents_shape: List[int],
+        exit_stack: ExitStack,
         do_classifier_free_guidance: bool = True,
     ) -> List[ControlNetData]:
 
@@ -237,25 +267,20 @@ class TextToLatentsInvocation(BaseInvocation):
             control_data = []
             control_models = []
             for control_info in control_list:
-                # handle control models
-                if ("," in control_info.control_model):
-                    control_model_split = control_info.control_model.split(",")
-                    control_name = control_model_split[0]
-                    control_subfolder = control_model_split[1]
-                    print("Using HF model subfolders")
-                    print("    control_name: ", control_name)
-                    print("    control_subfolder: ", control_subfolder)
-                    control_model = ControlNetModel.from_pretrained(
-                        control_name, subfolder=control_subfolder,
-                        torch_dtype=model.unet.dtype).to(
-                        model.device)
-                else:
-                    control_model = ControlNetModel.from_pretrained(
-                        control_info.control_model, torch_dtype=model.unet.dtype).to(model.device)
+                control_model = exit_stack.enter_context(
+                    context.services.model_manager.get_model(
+                        model_name=control_info.control_model.model_name,
+                        model_type=ModelType.ControlNet,
+                        base_model=control_info.control_model.base_model,
+                        context=context,
+                    )
+                )
+
                 control_models.append(control_model)
                 control_image_field = control_info.image
                 input_image = context.services.images.get_pil_image(
-                    control_image_field.image_name)
+                    control_image_field.image_name
+                )
                 # self.image.image_type, self.image.image_name
                 # FIXME: still need to test with different widths, heights, devices, dtypes
                 #        and add in batch_size, num_images_per_prompt?
@@ -277,7 +302,8 @@ class TextToLatentsInvocation(BaseInvocation):
                     weight=control_info.control_weight,
                     begin_step_percent=control_info.begin_step_percent,
                     end_step_percent=control_info.end_step_percent,
-                    control_mode=control_info.control_mode,)
+                    control_mode=control_info.control_mode,
+                )
                 control_data.append(control_item)
                 # MultiControlNetModel has been refactored out, just need list[ControlNetData]
         return control_data
@@ -288,7 +314,8 @@ class TextToLatentsInvocation(BaseInvocation):
 
         # Get the source node id (we are invoking the prepared node)
         graph_execution_state = context.services.graph_execution_manager.get(
-            context.graph_execution_state_id)
+            context.graph_execution_state_id
+        )
         source_node_id = graph_execution_state.prepared_source_mapping[self.id]
 
         def step_callback(state: PipelineIntermediateState):
@@ -297,16 +324,21 @@ class TextToLatentsInvocation(BaseInvocation):
         def _lora_loader():
             for lora in self.unet.loras:
                 lora_info = context.services.model_manager.get_model(
-                    **lora.dict(exclude={"weight"}))
+                    **lora.dict(exclude={"weight"}), context=context,
+                )
                 yield (lora_info.context.model, lora.weight)
                 del lora_info
             return
 
         unet_info = context.services.model_manager.get_model(
-            **self.unet.unet.dict())
-        with ModelPatcher.apply_lora_unet(unet_info.context.model, _lora_loader()),\
+            **self.unet.unet.dict(), context=context,
+        )
+        with ExitStack() as exit_stack,\
+                ModelPatcher.apply_lora_unet(unet_info.context.model, _lora_loader()),\
                 unet_info as unet:
 
+            noise = noise.to(device=unet.device, dtype=unet.dtype)
+
             scheduler = get_scheduler(
                 context=context,
                 scheduler_info=self.unet.scheduler,
@@ -314,13 +346,14 @@ class TextToLatentsInvocation(BaseInvocation):
             )
 
             pipeline = self.create_pipeline(unet, scheduler)
-            conditioning_data = self.get_conditioning_data(context, scheduler)
+            conditioning_data = self.get_conditioning_data(context, scheduler, unet)
 
             control_data = self.prep_control_data(
                 model=pipeline, context=context, control_input=self.control,
                 latents_shape=noise.shape,
                 # do_classifier_free_guidance=(self.cfg_scale >= 1.0))
                 do_classifier_free_guidance=True,
+                exit_stack=exit_stack,
             )
 
             # TODO: Verify the noise is the right size
@@ -334,6 +367,7 @@ class TextToLatentsInvocation(BaseInvocation):
             )
 
         # https://discuss.huggingface.co/t/memory-usage-by-later-pipeline-stages/23699
+        result_latents = result_latents.to("cpu")
         torch.cuda.empty_cache()
 
         name = f'{context.graph_execution_state_id}__{self.id}'
@@ -357,6 +391,7 @@ class LatentsToLatentsInvocation(TextToLatentsInvocation):
     class Config(InvocationConfig):
         schema_extra = {
             "ui": {
+                "title": "Latent To Latents",
                 "tags": ["latents"],
                 "type_hints": {
                     "model": "model",
@@ -373,7 +408,8 @@ class LatentsToLatentsInvocation(TextToLatentsInvocation):
 
         # Get the source node id (we are invoking the prepared node)
         graph_execution_state = context.services.graph_execution_manager.get(
-            context.graph_execution_state_id)
+            context.graph_execution_state_id
+        )
         source_node_id = graph_execution_state.prepared_source_mapping[self.id]
 
         def step_callback(state: PipelineIntermediateState):
@@ -382,16 +418,22 @@ class LatentsToLatentsInvocation(TextToLatentsInvocation):
         def _lora_loader():
             for lora in self.unet.loras:
                 lora_info = context.services.model_manager.get_model(
-                    **lora.dict(exclude={"weight"}))
+                    **lora.dict(exclude={"weight"}), context=context,
+                )
                 yield (lora_info.context.model, lora.weight)
                 del lora_info
             return
 
         unet_info = context.services.model_manager.get_model(
-            **self.unet.unet.dict())
-        with ModelPatcher.apply_lora_unet(unet_info.context.model, _lora_loader()),\
+            **self.unet.unet.dict(), context=context,
+        )
+        with ExitStack() as exit_stack,\
+                ModelPatcher.apply_lora_unet(unet_info.context.model, _lora_loader()),\
                 unet_info as unet:
 
+            noise = noise.to(device=unet.device, dtype=unet.dtype)
+            latent = latent.to(device=unet.device, dtype=unet.dtype)
+
             scheduler = get_scheduler(
                 context=context,
                 scheduler_info=self.unet.scheduler,
@@ -399,18 +441,20 @@ class LatentsToLatentsInvocation(TextToLatentsInvocation):
             )
 
             pipeline = self.create_pipeline(unet, scheduler)
-            conditioning_data = self.get_conditioning_data(context, scheduler)
+            conditioning_data = self.get_conditioning_data(context, scheduler, unet)
 
             control_data = self.prep_control_data(
                 model=pipeline, context=context, control_input=self.control,
                 latents_shape=noise.shape,
                 # do_classifier_free_guidance=(self.cfg_scale >= 1.0))
                 do_classifier_free_guidance=True,
+                exit_stack=exit_stack,
             )
 
             # TODO: Verify the noise is the right size
             initial_latents = latent if self.strength < 1.0 else torch.zeros_like(
-                latent, device=unet.device, dtype=latent.dtype)
+                latent, device=unet.device, dtype=latent.dtype
+            )
 
             timesteps, _ = pipeline.get_img2img_timesteps(
                 self.steps,
@@ -429,6 +473,7 @@ class LatentsToLatentsInvocation(TextToLatentsInvocation):
             )
 
         # https://discuss.huggingface.co/t/memory-usage-by-later-pipeline-stages/23699
+        result_latents = result_latents.to("cpu")
         torch.cuda.empty_cache()
 
         name = f'{context.graph_execution_state_id}__{self.id}'
@@ -449,11 +494,14 @@ class LatentsToImageInvocation(BaseInvocation):
     tiled: bool = Field(
         default=False,
         description="Decode latents by overlaping tiles(less memory consumption)")
+    fp32: bool = Field(False, description="Decode in full precision")
+    metadata: Optional[CoreMetadata] = Field(default=None, description="Optional core metadata to be written to the image")
 
     # Schema customisation
     class Config(InvocationConfig):
         schema_extra = {
             "ui": {
+                "title": "Latents To Image",
                 "tags": ["latents", "image"],
             },
         }
@@ -463,10 +511,36 @@ class LatentsToImageInvocation(BaseInvocation):
         latents = context.services.latents.get(self.latents.latents_name)
 
         vae_info = context.services.model_manager.get_model(
-            **self.vae.vae.dict(),
+            **self.vae.vae.dict(), context=context,
         )
 
         with vae_info as vae:
+            latents = latents.to(vae.device)
+            if self.fp32:
+                vae.to(dtype=torch.float32)
+
+                use_torch_2_0_or_xformers = isinstance(
+                    vae.decoder.mid_block.attentions[0].processor,
+                    (
+                        AttnProcessor2_0,
+                        XFormersAttnProcessor,
+                        LoRAXFormersAttnProcessor,
+                        LoRAAttnProcessor2_0,
+                    ),
+                )
+                # if xformers or torch_2_0 is used attention block does not need
+                # to be in float32 which can save lots of memory
+                if use_torch_2_0_or_xformers:
+                    vae.post_quant_conv.to(latents.dtype)
+                    vae.decoder.conv_in.to(latents.dtype)
+                    vae.decoder.mid_block.to(latents.dtype)
+                else:
+                    latents = latents.float()
+
+            else:
+                vae.to(dtype=torch.float16)
+                latents = latents.half()
+
             if self.tiled or context.services.configuration.tiled_decode:
                 vae.enable_tiling()
             else:
@@ -493,7 +567,8 @@ class LatentsToImageInvocation(BaseInvocation):
             image_category=ImageCategory.GENERAL,
             node_id=self.id,
             session_id=context.graph_execution_state_id,
-            is_intermediate=self.is_intermediate
+            is_intermediate=self.is_intermediate,
+            metadata=self.metadata.dict() if self.metadata else None,
         )
 
         return ImageOutput(
@@ -515,25 +590,38 @@ class ResizeLatentsInvocation(BaseInvocation):
     # Inputs
     latents: Optional[LatentsField] = Field(
         description="The latents to resize")
-    width:                         int = Field(
+    width:                         Union[int, None] = Field(default=512,
         ge=64, multiple_of=8, description="The width to resize to (px)")
-    height:                        int = Field(
+    height:                        Union[int, None] = Field(default=512,
         ge=64, multiple_of=8, description="The height to resize to (px)")
     mode:   LATENTS_INTERPOLATION_MODE = Field(
         default="bilinear", description="The interpolation mode")
     antialias: bool = Field(
         default=False,
         description="Whether or not to antialias (applied in bilinear and bicubic modes only)")
+    
+    class Config(InvocationConfig):
+        schema_extra = {
+            "ui": {
+                "title": "Resize Latents",
+                "tags": ["latents", "resize"]
+            },
+        }
 
     def invoke(self, context: InvocationContext) -> LatentsOutput:
         latents = context.services.latents.get(self.latents.latents_name)
 
+        # TODO:
+        device=choose_torch_device()
+
         resized_latents = torch.nn.functional.interpolate(
-            latents, size=(self.height // 8, self.width // 8),
+            latents.to(device), size=(self.height // 8, self.width // 8),
             mode=self.mode, antialias=self.antialias
-            if self.mode in ["bilinear", "bicubic"] else False,)
+            if self.mode in ["bilinear", "bicubic"] else False,
+        )
 
         # https://discuss.huggingface.co/t/memory-usage-by-later-pipeline-stages/23699
+        resized_latents = resized_latents.to("cpu")
         torch.cuda.empty_cache()
 
         name = f"{context.graph_execution_state_id}__{self.id}"
@@ -557,17 +645,30 @@ class ScaleLatentsInvocation(BaseInvocation):
     antialias: bool = Field(
         default=False,
         description="Whether or not to antialias (applied in bilinear and bicubic modes only)")
+    
+    class Config(InvocationConfig):
+        schema_extra = {
+            "ui": {
+                "title": "Scale Latents",
+                "tags": ["latents", "scale"]
+            },
+        }
 
     def invoke(self, context: InvocationContext) -> LatentsOutput:
         latents = context.services.latents.get(self.latents.latents_name)
 
+        # TODO:
+        device=choose_torch_device()
+
         # resizing
         resized_latents = torch.nn.functional.interpolate(
-            latents, scale_factor=self.scale_factor, mode=self.mode,
+            latents.to(device), scale_factor=self.scale_factor, mode=self.mode,
             antialias=self.antialias
-            if self.mode in ["bilinear", "bicubic"] else False,)
+            if self.mode in ["bilinear", "bicubic"] else False,
+        )
 
         # https://discuss.huggingface.co/t/memory-usage-by-later-pipeline-stages/23699
+        resized_latents = resized_latents.to("cpu")
         torch.cuda.empty_cache()
 
         name = f"{context.graph_execution_state_id}__{self.id}"
@@ -587,12 +688,15 @@ class ImageToLatentsInvocation(BaseInvocation):
     tiled: bool = Field(
         default=False,
         description="Encode latents by overlaping tiles(less memory consumption)")
+    fp32: bool = Field(False, description="Decode in full precision")
+
 
     # Schema customisation
     class Config(InvocationConfig):
         schema_extra = {
             "ui": {
-                "tags": ["latents", "image"],
+                "title": "Image To Latents",
+                "tags": ["latents", "image"]
             },
         }
 
@@ -605,7 +709,7 @@ class ImageToLatentsInvocation(BaseInvocation):
 
         #vae_info = context.services.model_manager.get_model(**self.vae.vae.dict())
         vae_info = context.services.model_manager.get_model(
-            **self.vae.vae.dict(),
+            **self.vae.vae.dict(), context=context,
         )
 
         image_tensor = image_resized_to_grid_as_tensor(image.convert("RGB"))
@@ -613,6 +717,32 @@ class ImageToLatentsInvocation(BaseInvocation):
             image_tensor = einops.rearrange(image_tensor, "c h w -> 1 c h w")
 
         with vae_info as vae:
+            orig_dtype = vae.dtype
+            if self.fp32:
+                vae.to(dtype=torch.float32)
+
+                use_torch_2_0_or_xformers = isinstance(
+                    vae.decoder.mid_block.attentions[0].processor,
+                    (
+                        AttnProcessor2_0,
+                        XFormersAttnProcessor,
+                        LoRAXFormersAttnProcessor,
+                        LoRAAttnProcessor2_0,
+                    ),
+                )
+                # if xformers or torch_2_0 is used attention block does not need
+                # to be in float32 which can save lots of memory
+                if use_torch_2_0_or_xformers:
+                    vae.post_quant_conv.to(orig_dtype)
+                    vae.decoder.conv_in.to(orig_dtype)
+                    vae.decoder.mid_block.to(orig_dtype)
+                #else:
+                #    latents = latents.float()
+
+            else:
+                vae.to(dtype=torch.float16)
+                #latents = latents.half()
+
             if self.tiled:
                 vae.enable_tiling()
             else:
@@ -627,8 +757,9 @@ class ImageToLatentsInvocation(BaseInvocation):
                 )  # FIXME: uses torch.randn. make reproducible!
 
             latents = 0.18215 * latents
+            latents = latents.to(dtype=orig_dtype)
 
         name = f"{context.graph_execution_state_id}__{self.id}"
-        # context.services.latents.set(name, latents)
+        latents = latents.to("cpu")
         context.services.latents.save(name, latents)
         return build_latents_output(latents_name=name, latents=latents)

invokeai/app/invocations/math.py

@@ -52,6 +52,14 @@ class AddInvocation(BaseInvocation, MathInvocationConfig):
     b: int = Field(default=0, description="The second number")
     # fmt: on
 
+    class Config(InvocationConfig):
+        schema_extra = {
+            "ui": {
+                "title": "Add",
+                "tags": ["math", "add"]
+            },
+        }
+
     def invoke(self, context: InvocationContext) -> IntOutput:
         return IntOutput(a=self.a + self.b)
 
@@ -65,6 +73,14 @@ class SubtractInvocation(BaseInvocation, MathInvocationConfig):
     b: int = Field(default=0, description="The second number")
     # fmt: on
 
+    class Config(InvocationConfig):
+        schema_extra = {
+            "ui": {
+                "title": "Subtract",
+                "tags": ["math", "subtract"]
+            },
+        }
+
     def invoke(self, context: InvocationContext) -> IntOutput:
         return IntOutput(a=self.a - self.b)
 
@@ -78,6 +94,14 @@ class MultiplyInvocation(BaseInvocation, MathInvocationConfig):
     b: int = Field(default=0, description="The second number")
     # fmt: on
 
+    class Config(InvocationConfig):
+        schema_extra = {
+            "ui": {
+                "title": "Multiply",
+                "tags": ["math", "multiply"]
+            },
+        }
+
     def invoke(self, context: InvocationContext) -> IntOutput:
         return IntOutput(a=self.a * self.b)
 
@@ -91,6 +115,14 @@ class DivideInvocation(BaseInvocation, MathInvocationConfig):
     b: int = Field(default=0, description="The second number")
     # fmt: on
 
+    class Config(InvocationConfig):
+        schema_extra = {
+            "ui": {
+                "title": "Divide",
+                "tags": ["math", "divide"]
+            },
+        }
+
     def invoke(self, context: InvocationContext) -> IntOutput:
         return IntOutput(a=int(self.a / self.b))
 
@@ -105,5 +137,14 @@ class RandomIntInvocation(BaseInvocation):
         default=np.iinfo(np.int32).max, description="The exclusive high value"
     )
     # fmt: on
+
+    class Config(InvocationConfig):
+        schema_extra = {
+            "ui": {
+                "title": "Random Integer",
+                "tags": ["math", "random", "integer"]
+            },
+        }
+
     def invoke(self, context: InvocationContext) -> IntOutput:
         return IntOutput(a=np.random.randint(self.low, self.high))

invokeai/app/invocations/metadata.py

@@ -0,0 +1,132 @@
+from typing import Literal, Optional, Union
+
+from pydantic import BaseModel, Field
+
+from invokeai.app.invocations.baseinvocation import (BaseInvocation,
+                                                     BaseInvocationOutput, InvocationConfig,
+                                                     InvocationContext)
+from invokeai.app.invocations.controlnet_image_processors import ControlField
+from invokeai.app.invocations.model import (LoRAModelField, MainModelField,
+                                            VAEModelField)
+
+
+class LoRAMetadataField(BaseModel):
+    """LoRA metadata for an image generated in InvokeAI."""
+    lora: LoRAModelField = Field(description="The LoRA model")
+    weight: float = Field(description="The weight of the LoRA model")
+
+
+class CoreMetadata(BaseModel):
+    """Core generation metadata for an image generated in InvokeAI."""
+
+    generation_mode: str = Field(description="The generation mode that output this image",)
+    positive_prompt: str = Field(description="The positive prompt parameter")
+    negative_prompt: str = Field(description="The negative prompt parameter")
+    width: int = Field(description="The width parameter")
+    height: int = Field(description="The height parameter")
+    seed: int = Field(description="The seed used for noise generation")
+    rand_device: str = Field(description="The device used for random number generation")
+    cfg_scale: float = Field(description="The classifier-free guidance scale parameter")
+    steps: int = Field(description="The number of steps used for inference")
+    scheduler: str = Field(description="The scheduler used for inference")
+    clip_skip: int = Field(description="The number of skipped CLIP layers",)
+    model: MainModelField = Field(description="The main model used for inference")
+    controlnets: list[ControlField]= Field(description="The ControlNets used for inference")
+    loras: list[LoRAMetadataField] = Field(description="The LoRAs used for inference")
+    strength: Union[float, None] = Field(
+        default=None,
+        description="The strength used for latents-to-latents",
+    )
+    init_image: Union[str, None] = Field(
+        default=None, description="The name of the initial image"
+    )
+    vae: Union[VAEModelField, None] = Field(
+        default=None,
+        description="The VAE used for decoding, if the main model's default was not used",
+    )
+
+
+class ImageMetadata(BaseModel):
+    """An image's generation metadata"""
+
+    metadata: Optional[dict] = Field(
+        default=None,
+        description="The image's core metadata, if it was created in the Linear or Canvas UI",
+    )
+    graph: Optional[dict] = Field(
+        default=None, description="The graph that created the image"
+    )
+
+
+class MetadataAccumulatorOutput(BaseInvocationOutput):
+    """The output of the MetadataAccumulator node"""
+
+    type: Literal["metadata_accumulator_output"] = "metadata_accumulator_output"
+
+    metadata: CoreMetadata = Field(description="The core metadata for the image")
+
+
+class MetadataAccumulatorInvocation(BaseInvocation):
+    """Outputs a Core Metadata Object"""
+
+    type: Literal["metadata_accumulator"] = "metadata_accumulator"
+
+    generation_mode: str = Field(description="The generation mode that output this image",)
+    positive_prompt: str = Field(description="The positive prompt parameter")
+    negative_prompt: str = Field(description="The negative prompt parameter")
+    width: int = Field(description="The width parameter")
+    height: int = Field(description="The height parameter")
+    seed: int = Field(description="The seed used for noise generation")
+    rand_device: str = Field(description="The device used for random number generation")
+    cfg_scale: float = Field(description="The classifier-free guidance scale parameter")
+    steps: int = Field(description="The number of steps used for inference")
+    scheduler: str = Field(description="The scheduler used for inference")
+    clip_skip: int = Field(description="The number of skipped CLIP layers",)
+    model: MainModelField = Field(description="The main model used for inference")
+    controlnets: list[ControlField]= Field(description="The ControlNets used for inference")
+    loras: list[LoRAMetadataField] = Field(description="The LoRAs used for inference")
+    strength: Union[float, None] = Field(
+        default=None,
+        description="The strength used for latents-to-latents",
+    )
+    init_image: Union[str, None] = Field(
+        default=None, description="The name of the initial image"
+    )
+    vae: Union[VAEModelField, None] = Field(
+        default=None,
+        description="The VAE used for decoding, if the main model's default was not used",
+    )
+
+    class Config(InvocationConfig):
+        schema_extra = {
+            "ui": {
+                "title": "Metadata Accumulator",
+                "tags": ["image", "metadata", "generation"]
+            },
+        }
+
+
+    def invoke(self, context: InvocationContext) -> MetadataAccumulatorOutput:
+        """Collects and outputs a CoreMetadata object"""
+
+        return MetadataAccumulatorOutput(
+            metadata=CoreMetadata(
+                generation_mode=self.generation_mode,
+                positive_prompt=self.positive_prompt,
+                negative_prompt=self.negative_prompt,
+                width=self.width,
+                height=self.height,
+                seed=self.seed,
+                rand_device=self.rand_device,
+                cfg_scale=self.cfg_scale,
+                steps=self.steps,
+                scheduler=self.scheduler,
+                model=self.model,
+                strength=self.strength,
+                init_image=self.init_image,
+                vae=self.vae,
+                controlnets=self.controlnets,
+                loras=self.loras,
+                clip_skip=self.clip_skip,
+            )
+        )

invokeai/app/invocations/model.py

@@ -33,7 +33,6 @@ class ClipField(BaseModel):
     skipped_layers: int = Field(description="Number of skipped layers in text_encoder")
     loras: List[LoraInfo] = Field(description="Loras to apply on model loading")
 
-
 class VaeField(BaseModel):
     # TODO: better naming?
     vae: ModelInfo = Field(description="Info to load vae submodel")
@@ -50,12 +49,12 @@ class ModelLoaderOutput(BaseInvocationOutput):
     vae: VaeField = Field(default=None, description="Vae submodel")
     # fmt: on
 
-
 class MainModelField(BaseModel):
     """Main model field"""
 
     model_name: str = Field(description="Name of the model")
     base_model: BaseModelType = Field(description="Base model")
+    model_type: ModelType = Field(description="Model Type")
 
 
 class LoRAModelField(BaseModel):
@@ -64,7 +63,6 @@ class LoRAModelField(BaseModel):
     model_name: str = Field(description="Name of the LoRA model")
     base_model: BaseModelType = Field(description="Base model")
 
-
 class MainModelLoaderInvocation(BaseInvocation):
     """Loads a main model, outputting its submodels."""
 
@@ -157,6 +155,22 @@ class MainModelLoaderInvocation(BaseInvocation):
                 loras=[],
                 skipped_layers=0,
             ),
+            clip2=ClipField(
+                tokenizer=ModelInfo(
+                    model_name=model_name,
+                    base_model=base_model,
+                    model_type=model_type,
+                    submodel=SubModelType.Tokenizer2,
+                ),
+                text_encoder=ModelInfo(
+                    model_name=model_name,
+                    base_model=base_model,
+                    model_type=model_type,
+                    submodel=SubModelType.TextEncoder2,
+                ),
+                loras=[],
+                skipped_layers=0,
+            ),
             vae=VaeField(
                 vae=ModelInfo(
                     model_name=model_name,
@@ -167,7 +181,7 @@ class MainModelLoaderInvocation(BaseInvocation):
             ),
         )
 
-
+    
 class LoraLoaderOutput(BaseInvocationOutput):
     """Model loader output"""
 
@@ -208,6 +222,9 @@ class LoraLoaderInvocation(BaseInvocation):
         base_model = self.lora.base_model
         lora_name = self.lora.model_name
 
+        # TODO: ui rewrite
+        base_model = BaseModelType.StableDiffusion1
+
         if not context.services.model_manager.model_exists(
             base_model=base_model,
             model_name=lora_name,

invokeai/app/invocations/noise.py

@@ -48,7 +48,7 @@ def get_noise(
         dtype=torch_dtype(device),
         device=noise_device_type,
         generator=generator,
-    ).to(device)
+    ).to("cpu")
 
     return noise_tensor
 
@@ -112,6 +112,7 @@ class NoiseInvocation(BaseInvocation):
     class Config(InvocationConfig):
         schema_extra = {
             "ui": {
+                "title": "Noise",
                 "tags": ["latents", "noise"],
             },
         }

invokeai/app/invocations/onnx.py

@@ -0,0 +1,591 @@
+# Copyright (c) 2023 Borisov Sergey (https://github.com/StAlKeR7779)
+
+from contextlib import ExitStack
+from typing import List, Literal, Optional, Union
+
+import re
+import inspect
+
+from pydantic import BaseModel, Field, validator
+import torch
+import numpy as np
+from diffusers import ControlNetModel, DPMSolverMultistepScheduler
+from diffusers.image_processor import VaeImageProcessor
+from diffusers.schedulers import SchedulerMixin as Scheduler
+
+from ..models.image import ImageCategory, ImageField, ResourceOrigin
+from ...backend.model_management import ONNXModelPatcher
+from ...backend.util import choose_torch_device
+from .baseinvocation import (BaseInvocation, BaseInvocationOutput,
+                             InvocationConfig, InvocationContext)
+from .compel import ConditioningField
+from .controlnet_image_processors import ControlField
+from .image import ImageOutput
+from .model import ModelInfo, UNetField, VaeField
+
+from invokeai.app.invocations.metadata import CoreMetadata
+from invokeai.backend import BaseModelType, ModelType, SubModelType
+from invokeai.app.util.step_callback import stable_diffusion_step_callback
+from ...backend.stable_diffusion import PipelineIntermediateState
+
+from tqdm import tqdm
+from .model import ClipField
+from .latent import LatentsField, LatentsOutput, build_latents_output, get_scheduler, SAMPLER_NAME_VALUES
+from .compel import CompelOutput
+
+
+ORT_TO_NP_TYPE = {
+    "tensor(bool)": np.bool_,
+    "tensor(int8)": np.int8,
+    "tensor(uint8)": np.uint8,
+    "tensor(int16)": np.int16,
+    "tensor(uint16)": np.uint16,
+    "tensor(int32)": np.int32,
+    "tensor(uint32)": np.uint32,
+    "tensor(int64)": np.int64,
+    "tensor(uint64)": np.uint64,
+    "tensor(float16)": np.float16,
+    "tensor(float)": np.float32,
+    "tensor(double)": np.float64,
+}
+
+
+class ONNXPromptInvocation(BaseInvocation):
+    type: Literal["prompt_onnx"] = "prompt_onnx"
+
+    prompt: str = Field(default="", description="Prompt")
+    clip: ClipField = Field(None, description="Clip to use")
+
+    def invoke(self, context: InvocationContext) -> CompelOutput:
+        tokenizer_info = context.services.model_manager.get_model(
+            **self.clip.tokenizer.dict(),
+        )
+        text_encoder_info = context.services.model_manager.get_model(
+            **self.clip.text_encoder.dict(),
+        )
+        with tokenizer_info as orig_tokenizer,\
+             text_encoder_info as text_encoder,\
+             ExitStack() as stack:
+
+            #loras = [(stack.enter_context(context.services.model_manager.get_model(**lora.dict(exclude={"weight"}))), lora.weight) for lora in self.clip.loras]
+            loras = [(context.services.model_manager.get_model(**lora.dict(exclude={"weight"})).context.model, lora.weight) for lora in self.clip.loras]
+
+            ti_list = []
+            for trigger in re.findall(r"<[a-zA-Z0-9., _-]+>", self.prompt):
+                name = trigger[1:-1]
+                try:
+                    ti_list.append(
+                        #stack.enter_context(
+                        #    context.services.model_manager.get_model(
+                        #        model_name=name,
+                        #        base_model=self.clip.text_encoder.base_model,
+                        #        model_type=ModelType.TextualInversion,
+                        #    )
+                        #)
+                        context.services.model_manager.get_model(
+                            model_name=name,
+                            base_model=self.clip.text_encoder.base_model,
+                            model_type=ModelType.TextualInversion,
+                        ).context.model
+                    )
+                except Exception:
+                    #print(e)
+                    #import traceback
+                    #print(traceback.format_exc())
+                    print(f"Warn: trigger: \"{trigger}\" not found")
+
+            with ONNXModelPatcher.apply_lora_text_encoder(text_encoder, loras),\
+                 ONNXModelPatcher.apply_ti(orig_tokenizer, text_encoder, ti_list) as (tokenizer, ti_manager):
+
+                text_encoder.create_session()
+
+                # copy from
+                # https://github.com/huggingface/diffusers/blob/3ebbaf7c96801271f9e6c21400033b6aa5ffcf29/src/diffusers/pipelines/stable_diffusion/pipeline_onnx_stable_diffusion.py#L153
+                text_inputs = tokenizer(
+                    self.prompt,
+                    padding="max_length",
+                    max_length=tokenizer.model_max_length,
+                    truncation=True,
+                    return_tensors="np",
+                )
+                text_input_ids = text_inputs.input_ids
+                """
+                untruncated_ids = tokenizer(prompt, padding="max_length", return_tensors="np").input_ids
+
+                if not np.array_equal(text_input_ids, untruncated_ids):
+                    removed_text = self.tokenizer.batch_decode(
+                        untruncated_ids[:, self.tokenizer.model_max_length - 1 : -1]
+                    )
+                    logger.warning(
+                        "The following part of your input was truncated because CLIP can only handle sequences up to"
+                        f" {self.tokenizer.model_max_length} tokens: {removed_text}"
+                    )
+                """
+
+                prompt_embeds = text_encoder(input_ids=text_input_ids.astype(np.int32))[0]
+
+                text_encoder.release_session()
+
+        conditioning_name = f"{context.graph_execution_state_id}_{self.id}_conditioning"
+
+        # TODO: hacky but works ;D maybe rename latents somehow?
+        context.services.latents.save(conditioning_name, (prompt_embeds, None))
+
+        return CompelOutput(
+            conditioning=ConditioningField(
+                conditioning_name=conditioning_name,
+            ),
+        )
+
+# Text to image
+class ONNXTextToLatentsInvocation(BaseInvocation):
+    """Generates latents from conditionings."""
+
+    type: Literal["t2l_onnx"] = "t2l_onnx"
+
+    # Inputs
+    # fmt: off
+    positive_conditioning: Optional[ConditioningField] = Field(description="Positive conditioning for generation")
+    negative_conditioning: Optional[ConditioningField] = Field(description="Negative conditioning for generation")
+    noise: Optional[LatentsField] = Field(description="The noise to use")
+    steps:       int = Field(default=10, gt=0, description="The number of steps to use to generate the image")
+    cfg_scale: Union[float, List[float]] = Field(default=7.5, ge=1, description="The Classifier-Free Guidance, higher values may result in a result closer to the prompt", )
+    scheduler: SAMPLER_NAME_VALUES = Field(default="euler", description="The scheduler to use" )
+    unet: UNetField = Field(default=None, description="UNet submodel")
+    #control: Union[ControlField, list[ControlField]] = Field(default=None, description="The control to use")
+    #seamless:   bool = Field(default=False, description="Whether or not to generate an image that can tile without seams", )
+    #seamless_axes: str = Field(default="", description="The axes to tile the image on, 'x' and/or 'y'")
+    # fmt: on
+
+    @validator("cfg_scale")
+    def ge_one(cls, v):
+        """validate that all cfg_scale values are >= 1"""
+        if isinstance(v, list):
+            for i in v:
+                if i < 1:
+                    raise ValueError('cfg_scale must be greater than 1')
+        else:
+            if v < 1:
+                raise ValueError('cfg_scale must be greater than 1')
+        return v
+
+    # Schema customisation
+    class Config(InvocationConfig):
+        schema_extra = {
+            "ui": {
+                "tags": ["latents"],
+                "type_hints": {
+                  "model": "model",
+                  # "cfg_scale": "float",
+                  "cfg_scale": "number"
+                }
+            },
+        }
+
+    # based on
+    # https://github.com/huggingface/diffusers/blob/3ebbaf7c96801271f9e6c21400033b6aa5ffcf29/src/diffusers/pipelines/stable_diffusion/pipeline_onnx_stable_diffusion.py#L375
+    def invoke(self, context: InvocationContext) -> LatentsOutput:
+        c, _ = context.services.latents.get(self.positive_conditioning.conditioning_name)
+        uc, _ = context.services.latents.get(self.negative_conditioning.conditioning_name)
+        graph_execution_state = context.services.graph_execution_manager.get(
+            context.graph_execution_state_id)
+        source_node_id = graph_execution_state.prepared_source_mapping[self.id]
+        if isinstance(c, torch.Tensor):
+            c = c.cpu().numpy()
+        if isinstance(uc, torch.Tensor):
+            uc = uc.cpu().numpy()
+        device = torch.device(choose_torch_device())
+        prompt_embeds = np.concatenate([uc, c])
+
+        latents = context.services.latents.get(self.noise.latents_name)
+        if isinstance(latents, torch.Tensor):
+            latents = latents.cpu().numpy()
+
+        # TODO: better execution device handling
+        latents = latents.astype(np.float16)
+
+        # get the initial random noise unless the user supplied it
+        do_classifier_free_guidance = True
+        #latents_dtype = prompt_embeds.dtype
+        #latents_shape = (batch_size * num_images_per_prompt, 4, height // 8, width // 8)
+        #if latents.shape != latents_shape:
+        #    raise ValueError(f"Unexpected latents shape, got {latents.shape}, expected {latents_shape}")
+
+        scheduler = get_scheduler(
+            context=context,
+            scheduler_info=self.unet.scheduler,
+            scheduler_name=self.scheduler,
+        )
+
+        def torch2numpy(latent: torch.Tensor):
+            return latent.cpu().numpy()
+
+        def numpy2torch(latent, device):
+            return torch.from_numpy(latent).to(device)
+
+        def dispatch_progress(
+                self, context: InvocationContext, source_node_id: str,
+                intermediate_state: PipelineIntermediateState) -> None:
+            stable_diffusion_step_callback(
+                context=context,
+                intermediate_state=intermediate_state,
+                node=self.dict(),
+                source_node_id=source_node_id,
+            )
+
+        scheduler.set_timesteps(self.steps)
+        latents = latents * np.float64(scheduler.init_noise_sigma)
+
+        extra_step_kwargs = dict()
+        if "eta" in set(inspect.signature(scheduler.step).parameters.keys()):
+            extra_step_kwargs.update(
+                eta=0.0,
+            )
+
+        unet_info = context.services.model_manager.get_model(**self.unet.unet.dict())
+
+        with unet_info as unet,\
+             ExitStack() as stack:
+
+            #loras = [(stack.enter_context(context.services.model_manager.get_model(**lora.dict(exclude={"weight"}))), lora.weight) for lora in self.unet.loras]
+            loras = [(context.services.model_manager.get_model(**lora.dict(exclude={"weight"})).context.model, lora.weight) for lora in self.unet.loras]
+
+            with ONNXModelPatcher.apply_lora_unet(unet, loras):
+                # TODO: 
+                unet.create_session()
+
+                timestep_dtype = next(
+                    (input.type for input in unet.session.get_inputs() if input.name == "timestep"), "tensor(float16)"
+                )
+                timestep_dtype = ORT_TO_NP_TYPE[timestep_dtype]
+                import time
+                times = []
+                for i in tqdm(range(len(scheduler.timesteps))):
+                    t = scheduler.timesteps[i]
+                    # expand the latents if we are doing classifier free guidance
+                    latent_model_input = np.concatenate([latents] * 2) if do_classifier_free_guidance else latents
+                    latent_model_input = scheduler.scale_model_input(numpy2torch(latent_model_input, device), t)
+                    latent_model_input = latent_model_input.cpu().numpy()
+
+                    # predict the noise residual
+                    timestep = np.array([t], dtype=timestep_dtype)
+                    start_time = time.time()
+                    noise_pred = unet(sample=latent_model_input, timestep=timestep, encoder_hidden_states=prompt_embeds)
+                    times.append(time.time() - start_time)
+                    noise_pred = noise_pred[0]
+
+                    # perform guidance
+                    if do_classifier_free_guidance:
+                        noise_pred_uncond, noise_pred_text = np.split(noise_pred, 2)
+                        noise_pred = noise_pred_uncond + self.cfg_scale * (noise_pred_text - noise_pred_uncond)
+
+                    # compute the previous noisy sample x_t -> x_t-1
+                    scheduler_output = scheduler.step(
+                        numpy2torch(noise_pred, device), t, numpy2torch(latents, device), **extra_step_kwargs
+                    )
+                    latents = torch2numpy(scheduler_output.prev_sample)
+
+                    state = PipelineIntermediateState(
+                        run_id= "test",
+                        step=i,
+                        timestep=timestep,
+                        latents=scheduler_output.prev_sample
+                    )
+                    dispatch_progress(
+                        self,
+                        context=context,
+                        source_node_id=source_node_id,
+                        intermediate_state=state
+                    )
+
+                    # call the callback, if provided
+                    #if callback is not None and i % callback_steps == 0:
+                    #    callback(i, t, latents)
+                print(times)
+                unet.release_session()
+
+        torch.cuda.empty_cache()
+
+        name = f'{context.graph_execution_state_id}__{self.id}'
+        context.services.latents.save(name, latents)
+        return build_latents_output(latents_name=name, latents=torch.from_numpy(latents))
+
+# Latent to image
+class ONNXLatentsToImageInvocation(BaseInvocation):
+    """Generates an image from latents."""
+
+    type: Literal["l2i_onnx"] = "l2i_onnx"
+
+    # Inputs
+    latents: Optional[LatentsField] = Field(description="The latents to generate an image from")
+    vae: VaeField = Field(default=None, description="Vae submodel")
+    metadata: Optional[CoreMetadata] = Field(default=None, description="Optional core metadata to be written to the image")
+    #tiled: bool = Field(default=False, description="Decode latents by overlaping tiles(less memory consumption)")
+
+    # Schema customisation
+    class Config(InvocationConfig):
+        schema_extra = {
+            "ui": {
+                "tags": ["latents", "image"],
+            },
+        }
+
+    def invoke(self, context: InvocationContext) -> ImageOutput:
+        latents = context.services.latents.get(self.latents.latents_name)
+
+        if self.vae.vae.submodel != SubModelType.VaeDecoder:
+            raise Exception(f"Expected vae_decoder, found: {self.vae.vae.model_type}")
+
+        vae_info = context.services.model_manager.get_model(
+            **self.vae.vae.dict(),
+        )
+
+        # clear memory as vae decode can request a lot
+        torch.cuda.empty_cache()
+
+        with vae_info as vae:
+            vae.create_session()
+
+            # copied from
+            # https://github.com/huggingface/diffusers/blob/3ebbaf7c96801271f9e6c21400033b6aa5ffcf29/src/diffusers/pipelines/stable_diffusion/pipeline_onnx_stable_diffusion.py#L427
+            latents = 1 / 0.18215 * latents
+            # image = self.vae_decoder(latent_sample=latents)[0]
+            # it seems likes there is a strange result for using half-precision vae decoder if batchsize>1
+            image = np.concatenate(
+                [vae(latent_sample=latents[i : i + 1])[0] for i in range(latents.shape[0])]
+            )
+
+            image = np.clip(image / 2 + 0.5, 0, 1)
+            image = image.transpose((0, 2, 3, 1))
+            image = VaeImageProcessor.numpy_to_pil(image)[0]
+
+            vae.release_session()
+
+        torch.cuda.empty_cache()
+
+        image_dto = context.services.images.create(
+            image=image,
+            image_origin=ResourceOrigin.INTERNAL,
+            image_category=ImageCategory.GENERAL,
+            node_id=self.id,
+            session_id=context.graph_execution_state_id,
+            is_intermediate=self.is_intermediate,
+            metadata=self.metadata.dict() if self.metadata else None,
+        )
+
+        return ImageOutput(
+            image=ImageField(image_name=image_dto.image_name),
+            width=image_dto.width,
+            height=image_dto.height,
+        )
+
+class ONNXModelLoaderOutput(BaseInvocationOutput):
+    """Model loader output"""
+
+    #fmt: off
+    type: Literal["model_loader_output_onnx"] = "model_loader_output_onnx"
+
+    unet: UNetField = Field(default=None, description="UNet submodel")
+    clip: ClipField = Field(default=None, description="Tokenizer and text_encoder submodels")
+    vae_decoder: VaeField = Field(default=None, description="Vae submodel")
+    vae_encoder: VaeField = Field(default=None, description="Vae submodel")
+    #fmt: on
+
+class ONNXSD1ModelLoaderInvocation(BaseInvocation):
+    """Loading submodels of selected model."""
+
+    type: Literal["sd1_model_loader_onnx"] = "sd1_model_loader_onnx"
+
+    model_name: str = Field(default="", description="Model to load")
+    # TODO: precision?
+
+    # Schema customisation
+    class Config(InvocationConfig):
+        schema_extra = {
+            "ui": {
+                "tags": ["model", "loader"],
+                "type_hints": {
+                  "model_name": "model" # TODO: rename to model_name?
+                }
+            },
+        }
+
+    def invoke(self, context: InvocationContext) -> ONNXModelLoaderOutput:
+
+        model_name = "stable-diffusion-v1-5"
+        base_model = BaseModelType.StableDiffusion1
+
+        # TODO: not found exceptions
+        if not context.services.model_manager.model_exists(
+            model_name=model_name,
+            base_model=BaseModelType.StableDiffusion1,
+            model_type=ModelType.ONNX,
+        ):
+            raise Exception(f"Unkown model name: {model_name}!")
+
+
+        return ONNXModelLoaderOutput(
+            unet=UNetField(
+                unet=ModelInfo(
+                    model_name=model_name,
+                    base_model=base_model,
+                    model_type=ModelType.ONNX,
+                    submodel=SubModelType.UNet,
+                ),
+                scheduler=ModelInfo(
+                    model_name=model_name,
+                    base_model=base_model,
+                    model_type=ModelType.ONNX,
+                    submodel=SubModelType.Scheduler,
+                ),
+                loras=[],
+            ),
+            clip=ClipField(
+                tokenizer=ModelInfo(
+                    model_name=model_name,
+                    base_model=base_model,
+                    model_type=ModelType.ONNX,
+                    submodel=SubModelType.Tokenizer,
+                ),
+                text_encoder=ModelInfo(
+                    model_name=model_name,
+                    base_model=base_model,
+                    model_type=ModelType.ONNX,
+                    submodel=SubModelType.TextEncoder,
+                ),
+                loras=[],
+            ),
+            vae_decoder=VaeField(
+                vae=ModelInfo(
+                    model_name=model_name,
+                    base_model=base_model,
+                    model_type=ModelType.ONNX,
+                    submodel=SubModelType.VaeDecoder,
+                ),
+            ),
+            vae_encoder=VaeField(
+                vae=ModelInfo(
+                    model_name=model_name,
+                    base_model=base_model,
+                    model_type=ModelType.ONNX,
+                    submodel=SubModelType.VaeEncoder,
+                ),
+            )
+        )
+
+class OnnxModelField(BaseModel):
+    """Onnx model field"""
+
+    model_name: str = Field(description="Name of the model")
+    base_model: BaseModelType = Field(description="Base model")
+    model_type: ModelType = Field(description="Model Type")
+
+class OnnxModelLoaderInvocation(BaseInvocation):
+    """Loads a main model, outputting its submodels."""
+
+    type: Literal["onnx_model_loader"] = "onnx_model_loader"
+
+    model: OnnxModelField = Field(description="The model to load")
+    # TODO: precision?
+
+    # Schema customisation
+    class Config(InvocationConfig):
+        schema_extra = {
+            "ui": {
+                "title": "Onnx Model Loader",
+                "tags": ["model", "loader"],
+                "type_hints": {"model": "model"},
+            },
+        }
+
+    def invoke(self, context: InvocationContext) -> ONNXModelLoaderOutput:
+        base_model = self.model.base_model
+        model_name = self.model.model_name
+        model_type = ModelType.ONNX
+
+        # TODO: not found exceptions
+        if not context.services.model_manager.model_exists(
+            model_name=model_name,
+            base_model=base_model,
+            model_type=model_type,
+        ):
+            raise Exception(f"Unknown {base_model} {model_type} model: {model_name}")
+
+        """
+        if not context.services.model_manager.model_exists(
+            model_name=self.model_name,
+            model_type=SDModelType.Diffusers,
+            submodel=SDModelType.Tokenizer,
+        ):
+            raise Exception(
+                f"Failed to find tokenizer submodel in {self.model_name}! Check if model corrupted"
+            )
+
+        if not context.services.model_manager.model_exists(
+            model_name=self.model_name,
+            model_type=SDModelType.Diffusers,
+            submodel=SDModelType.TextEncoder,
+        ):
+            raise Exception(
+                f"Failed to find text_encoder submodel in {self.model_name}! Check if model corrupted"
+            )
+
+        if not context.services.model_manager.model_exists(
+            model_name=self.model_name,
+            model_type=SDModelType.Diffusers,
+            submodel=SDModelType.UNet,
+        ):
+            raise Exception(
+                f"Failed to find unet submodel from {self.model_name}! Check if model corrupted"
+            )
+        """
+
+        return ONNXModelLoaderOutput(
+            unet=UNetField(
+                unet=ModelInfo(
+                    model_name=model_name,
+                    base_model=base_model,
+                    model_type=model_type,
+                    submodel=SubModelType.UNet,
+                ),
+                scheduler=ModelInfo(
+                    model_name=model_name,
+                    base_model=base_model,
+                    model_type=model_type,
+                    submodel=SubModelType.Scheduler,
+                ),
+                loras=[],
+            ),
+            clip=ClipField(
+                tokenizer=ModelInfo(
+                    model_name=model_name,
+                    base_model=base_model,
+                    model_type=model_type,
+                    submodel=SubModelType.Tokenizer,
+                ),
+                text_encoder=ModelInfo(
+                    model_name=model_name,
+                    base_model=base_model,
+                    model_type=model_type,
+                    submodel=SubModelType.TextEncoder,
+                ),
+                loras=[],
+                skipped_layers=0,
+            ),
+            vae_decoder=VaeField(
+                vae=ModelInfo(
+                    model_name=model_name,
+                    base_model=base_model,
+                    model_type=model_type,
+                    submodel=SubModelType.VaeDecoder,
+                ),
+            ),
+            vae_encoder=VaeField(
+                vae=ModelInfo(
+                    model_name=model_name,
+                    base_model=base_model,
+                    model_type=model_type,
+                    submodel=SubModelType.VaeEncoder,
+                ),
+            )
+        )

invokeai/app/invocations/param_easing.py

@@ -43,6 +43,14 @@ class FloatLinearRangeInvocation(BaseInvocation):
     stop: float = Field(default=10, description="The last value of the range")
     steps: int = Field(default=30, description="number of values to interpolate over (including start and stop)")
 
+    class Config(InvocationConfig):
+        schema_extra = {
+            "ui": {
+                "title": "Linear Range (Float)",
+                "tags": ["math", "float", "linear", "range"]
+            },
+        }
+
     def invoke(self, context: InvocationContext) -> FloatCollectionOutput:
         param_list = list(np.linspace(self.start, self.stop, self.steps))
         return FloatCollectionOutput(
@@ -113,6 +121,14 @@ class StepParamEasingInvocation(BaseInvocation):
     show_easing_plot: bool = Field(default=False, description="show easing plot")
     # fmt: on
 
+    class Config(InvocationConfig):
+        schema_extra = {
+            "ui": {
+                "title": "Param Easing By Step",
+                "tags": ["param", "step", "easing"]
+            },
+        }
+
 
     def invoke(self, context: InvocationContext) -> FloatCollectionOutput:
         log_diagnostics = False

invokeai/app/invocations/params.py

@@ -1,9 +1,12 @@
 # Copyright (c) 2023 Kyle Schouviller (https://github.com/kyle0654)
 
 from typing import Literal
+
 from pydantic import Field
-from .baseinvocation import BaseInvocation, BaseInvocationOutput, InvocationContext
-from .math import IntOutput, FloatOutput
+
+from .baseinvocation import (BaseInvocation, BaseInvocationOutput,
+                             InvocationConfig, InvocationContext)
+from .math import FloatOutput, IntOutput
 
 # Pass-through parameter nodes - used by subgraphs
 
@@ -14,6 +17,14 @@ class ParamIntInvocation(BaseInvocation):
     a: int = Field(default=0, description="The integer value")
     #fmt: on
 
+    class Config(InvocationConfig):
+      schema_extra = {
+          "ui": {
+              "tags": ["param", "integer"],
+              "title": "Integer Parameter"
+          },
+      }
+
     def invoke(self, context: InvocationContext) -> IntOutput:
         return IntOutput(a=self.a)
 
@@ -24,5 +35,36 @@ class ParamFloatInvocation(BaseInvocation):
     param: float = Field(default=0.0, description="The float value")
     #fmt: on
 
+    class Config(InvocationConfig):
+      schema_extra = {
+          "ui": {
+              "tags": ["param", "float"],
+              "title": "Float Parameter"
+          },
+      }
+
     def invoke(self, context: InvocationContext) -> FloatOutput:
         return FloatOutput(param=self.param)
+
+class StringOutput(BaseInvocationOutput):
+    """A string output"""
+    type: Literal["string_output"] = "string_output"
+    text: str = Field(default=None, description="The output string")
+
+
+class ParamStringInvocation(BaseInvocation):
+    """A string parameter"""
+    type: Literal['param_string'] = 'param_string'
+    text: str = Field(default='', description='The string value')
+
+    class Config(InvocationConfig):
+      schema_extra = {
+          "ui": {
+              "tags": ["param", "string"],
+              "title": "String Parameter"
+          },
+      }
+
+    def invoke(self, context: InvocationContext) -> StringOutput:
+        return StringOutput(text=self.text)
+    

invokeai/app/invocations/prompt.py

@@ -1,8 +1,10 @@
-from typing import Literal
+from os.path import exists
+from typing import Literal, Optional
 
-from pydantic.fields import Field
+import numpy as np
+from pydantic import Field, validator
 
-from .baseinvocation import BaseInvocation, BaseInvocationOutput, InvocationContext
+from .baseinvocation import BaseInvocation, BaseInvocationOutput, InvocationConfig, InvocationContext
 from dynamicprompts.generators import RandomPromptGenerator, CombinatorialPromptGenerator
 
 class PromptOutput(BaseInvocationOutput):
@@ -46,6 +48,14 @@ class DynamicPromptInvocation(BaseInvocation):
         default=False, description="Whether to use the combinatorial generator"
     )
 
+    class Config(InvocationConfig):
+        schema_extra = {
+            "ui": {
+                "title": "Dynamic Prompt",
+                "tags": ["prompt", "dynamic"]
+            },
+        }
+
     def invoke(self, context: InvocationContext) -> PromptCollectionOutput:
         if self.combinatorial:
             generator = CombinatorialPromptGenerator()
@@ -55,3 +65,49 @@ class DynamicPromptInvocation(BaseInvocation):
             prompts = generator.generate(self.prompt, num_images=self.max_prompts)
 
         return PromptCollectionOutput(prompt_collection=prompts, count=len(prompts))
+    
+
+class PromptsFromFileInvocation(BaseInvocation):
+    '''Loads prompts from a text file'''
+    # fmt: off
+    type: Literal['prompt_from_file'] = 'prompt_from_file'
+
+    # Inputs
+    file_path: str = Field(description="Path to prompt text file")
+    pre_prompt: Optional[str] = Field(description="String to prepend to each prompt")
+    post_prompt: Optional[str] = Field(description="String to append to each prompt")
+    start_line: int = Field(default=1, ge=1, description="Line in the file to start start from")
+    max_prompts: int = Field(default=1, ge=0, description="Max lines to read from file (0=all)")
+    #fmt: on
+
+    class Config(InvocationConfig):
+        schema_extra = {
+            "ui": {
+                "title": "Prompts From File",
+                "tags": ["prompt", "file"]
+            },
+        }
+
+    @validator("file_path")
+    def file_path_exists(cls, v):
+        if not exists(v):
+            raise ValueError(FileNotFoundError)
+        return v
+
+    def promptsFromFile(self, file_path: str, pre_prompt: str, post_prompt: str, start_line: int, max_prompts: int):
+        prompts = []
+        start_line -= 1
+        end_line = start_line + max_prompts
+        if max_prompts <= 0:
+            end_line = np.iinfo(np.int32).max
+        with open(file_path) as f:
+            for i, line in enumerate(f):
+                if i >= start_line and i < end_line:
+                    prompts.append((pre_prompt or '') + line.strip() + (post_prompt or ''))
+                if i >= end_line:
+                    break
+        return prompts
+
+    def invoke(self, context: InvocationContext) -> PromptCollectionOutput:
+        prompts = self.promptsFromFile(self.file_path, self.pre_prompt, self.post_prompt, self.start_line, self.max_prompts)
+        return PromptCollectionOutput(prompt_collection=prompts, count=len(prompts))

invokeai/app/invocations/reconstruct.py

@@ -1,55 +0,0 @@
-from typing import Literal, Optional
-
-from pydantic import Field
-
-from invokeai.app.models.image import ImageCategory, ImageField, ResourceOrigin
-
-from .baseinvocation import BaseInvocation, InvocationContext, InvocationConfig
-from .image import ImageOutput
-
-
-class RestoreFaceInvocation(BaseInvocation):
-    """Restores faces in an image."""
-
-    # fmt: off
-    type:  Literal["restore_face"] = "restore_face"
-
-    # Inputs
-    image: Optional[ImageField] = Field(description="The input image")
-    strength:                float = Field(default=0.75, gt=0, le=1, description="The strength of the restoration"  )
-    # fmt: on
-
-    # Schema customisation
-    class Config(InvocationConfig):
-        schema_extra = {
-            "ui": {
-                "tags": ["restoration", "image"],
-            },
-        }
-
-    def invoke(self, context: InvocationContext) -> ImageOutput:
-        image = context.services.images.get_pil_image(self.image.image_name)
-        results = context.services.restoration.upscale_and_reconstruct(
-            image_list=[[image, 0]],
-            upscale=None,
-            strength=self.strength,  # GFPGAN strength
-            save_original=False,
-            image_callback=None,
-        )
-
-        # Results are image and seed, unwrap for now
-        # TODO: can this return multiple results?
-        image_dto = context.services.images.create(
-            image=results[0][0],
-            image_origin=ResourceOrigin.INTERNAL,
-            image_category=ImageCategory.GENERAL,
-            node_id=self.id,
-            session_id=context.graph_execution_state_id,
-            is_intermediate=self.is_intermediate,
-        )
-
-        return ImageOutput(
-            image=ImageField(image_name=image_dto.image_name),
-            width=image_dto.width,
-            height=image_dto.height,
-        )

invokeai/app/invocations/sdxl.py

@@ -0,0 +1,662 @@
+import torch
+import inspect
+from tqdm import tqdm
+from typing import List, Literal, Optional, Union
+
+from pydantic import Field, validator
+
+from ...backend.model_management import ModelType, SubModelType
+from .baseinvocation import (BaseInvocation, BaseInvocationOutput,
+                             InvocationConfig, InvocationContext)
+
+from .model import UNetField, ClipField, VaeField, MainModelField, ModelInfo
+from .compel import ConditioningField
+from .latent import LatentsField, SAMPLER_NAME_VALUES, LatentsOutput, get_scheduler, build_latents_output
+
+class SDXLModelLoaderOutput(BaseInvocationOutput):
+    """SDXL base model loader output"""
+
+    # fmt: off
+    type: Literal["sdxl_model_loader_output"] = "sdxl_model_loader_output"
+
+    unet: UNetField = Field(default=None, description="UNet submodel")
+    clip: ClipField = Field(default=None, description="Tokenizer and text_encoder submodels")
+    clip2: ClipField = Field(default=None, description="Tokenizer and text_encoder submodels")
+    vae: VaeField = Field(default=None, description="Vae submodel")
+    # fmt: on
+
+class SDXLRefinerModelLoaderOutput(BaseInvocationOutput):
+    """SDXL refiner model loader output"""
+    # fmt: off
+    type: Literal["sdxl_refiner_model_loader_output"] = "sdxl_refiner_model_loader_output"
+    unet: UNetField = Field(default=None, description="UNet submodel")
+    clip2: ClipField = Field(default=None, description="Tokenizer and text_encoder submodels")
+    vae: VaeField = Field(default=None, description="Vae submodel")
+    # fmt: on
+    #fmt: on
+    
+class SDXLModelLoaderInvocation(BaseInvocation):
+    """Loads an sdxl base model, outputting its submodels."""
+
+    type: Literal["sdxl_model_loader"] = "sdxl_model_loader"
+
+    model: MainModelField = Field(description="The model to load")
+    # TODO: precision?
+
+    # Schema customisation
+    class Config(InvocationConfig):
+        schema_extra = {
+            "ui": {
+                "title": "SDXL Model Loader",
+                "tags": ["model", "loader", "sdxl"],
+                "type_hints": {"model": "model"},
+            },
+        }
+
+    def invoke(self, context: InvocationContext) -> SDXLModelLoaderOutput:
+        base_model = self.model.base_model
+        model_name = self.model.model_name
+        model_type = ModelType.Main
+
+        # TODO: not found exceptions
+        if not context.services.model_manager.model_exists(
+            model_name=model_name,
+            base_model=base_model,
+            model_type=model_type,
+        ):
+            raise Exception(f"Unknown {base_model} {model_type} model: {model_name}")
+
+        return SDXLModelLoaderOutput(
+            unet=UNetField(
+                unet=ModelInfo(
+                    model_name=model_name,
+                    base_model=base_model,
+                    model_type=model_type,
+                    submodel=SubModelType.UNet,
+                ),
+                scheduler=ModelInfo(
+                    model_name=model_name,
+                    base_model=base_model,
+                    model_type=model_type,
+                    submodel=SubModelType.Scheduler,
+                ),
+                loras=[],
+            ),
+            clip=ClipField(
+                tokenizer=ModelInfo(
+                    model_name=model_name,
+                    base_model=base_model,
+                    model_type=model_type,
+                    submodel=SubModelType.Tokenizer,
+                ),
+                text_encoder=ModelInfo(
+                    model_name=model_name,
+                    base_model=base_model,
+                    model_type=model_type,
+                    submodel=SubModelType.TextEncoder,
+                ),
+                loras=[],
+                skipped_layers=0,
+            ),
+            clip2=ClipField(
+                tokenizer=ModelInfo(
+                    model_name=model_name,
+                    base_model=base_model,
+                    model_type=model_type,
+                    submodel=SubModelType.Tokenizer2,
+                ),
+                text_encoder=ModelInfo(
+                    model_name=model_name,
+                    base_model=base_model,
+                    model_type=model_type,
+                    submodel=SubModelType.TextEncoder2,
+                ),
+                loras=[],
+                skipped_layers=0,
+            ),
+            vae=VaeField(
+                vae=ModelInfo(
+                    model_name=model_name,
+                    base_model=base_model,
+                    model_type=model_type,
+                    submodel=SubModelType.Vae,
+                ),
+            ),
+        )
+
+class SDXLRefinerModelLoaderInvocation(BaseInvocation):
+    """Loads an sdxl refiner model, outputting its submodels."""
+    type: Literal["sdxl_refiner_model_loader"] = "sdxl_refiner_model_loader"
+
+    model: MainModelField = Field(description="The model to load")
+    # TODO: precision?
+
+    # Schema customisation
+    class Config(InvocationConfig):
+        schema_extra = {
+            "ui": {
+                "title": "SDXL Refiner Model Loader",
+                "tags": ["model", "loader", "sdxl_refiner"],
+                "type_hints": {"model": "model"},
+            },
+        }
+
+    def invoke(self, context: InvocationContext) -> SDXLRefinerModelLoaderOutput:
+        base_model = self.model.base_model
+        model_name = self.model.model_name
+        model_type = ModelType.Main
+
+        # TODO: not found exceptions
+        if not context.services.model_manager.model_exists(
+            model_name=model_name,
+            base_model=base_model,
+            model_type=model_type,
+        ):
+            raise Exception(f"Unknown {base_model} {model_type} model: {model_name}")
+
+        return SDXLRefinerModelLoaderOutput(
+            unet=UNetField(
+                unet=ModelInfo(
+                    model_name=model_name,
+                    base_model=base_model,
+                    model_type=model_type,
+                    submodel=SubModelType.UNet,
+                ),
+                scheduler=ModelInfo(
+                    model_name=model_name,
+                    base_model=base_model,
+                    model_type=model_type,
+                    submodel=SubModelType.Scheduler,
+                ),
+                loras=[],
+            ),
+            clip2=ClipField(
+                tokenizer=ModelInfo(
+                    model_name=model_name,
+                    base_model=base_model,
+                    model_type=model_type,
+                    submodel=SubModelType.Tokenizer2,
+                ),
+                text_encoder=ModelInfo(
+                    model_name=model_name,
+                    base_model=base_model,
+                    model_type=model_type,
+                    submodel=SubModelType.TextEncoder2,
+                ),
+                loras=[],
+                skipped_layers=0,
+            ),
+            vae=VaeField(
+                vae=ModelInfo(
+                    model_name=model_name,
+                    base_model=base_model,
+                    model_type=model_type,
+                    submodel=SubModelType.Vae,
+                ),
+            ),
+        )
+    
+# Text to image
+class SDXLTextToLatentsInvocation(BaseInvocation):
+    """Generates latents from conditionings."""
+
+    type: Literal["t2l_sdxl"] = "t2l_sdxl"
+
+    # Inputs
+    # fmt: off
+    positive_conditioning: Optional[ConditioningField] = Field(description="Positive conditioning for generation")
+    negative_conditioning: Optional[ConditioningField] = Field(description="Negative conditioning for generation")
+    noise: Optional[LatentsField] = Field(description="The noise to use")
+    steps:       int = Field(default=10, gt=0, description="The number of steps to use to generate the image")
+    cfg_scale: Union[float, List[float]] = Field(default=7.5, ge=1, description="The Classifier-Free Guidance, higher values may result in a result closer to the prompt", )
+    scheduler: SAMPLER_NAME_VALUES = Field(default="euler", description="The scheduler to use" )
+    unet: UNetField = Field(default=None, description="UNet submodel")
+    denoising_end: float = Field(default=1.0, gt=0, le=1, description="")
+    #control: Union[ControlField, list[ControlField]] = Field(default=None, description="The control to use")
+    #seamless:   bool = Field(default=False, description="Whether or not to generate an image that can tile without seams", )
+    #seamless_axes: str = Field(default="", description="The axes to tile the image on, 'x' and/or 'y'")
+    # fmt: on
+
+    @validator("cfg_scale")
+    def ge_one(cls, v):
+        """validate that all cfg_scale values are >= 1"""
+        if isinstance(v, list):
+            for i in v:
+                if i < 1:
+                    raise ValueError('cfg_scale must be greater than 1')
+        else:
+            if v < 1:
+                raise ValueError('cfg_scale must be greater than 1')
+        return v
+
+    # Schema customisation
+    class Config(InvocationConfig):
+        schema_extra = {
+            "ui": {
+                "title": "SDXL Text To Latents",
+                "tags": ["latents"],
+                "type_hints": {
+                  "model": "model",
+                  # "cfg_scale": "float",
+                  "cfg_scale": "number"
+                }
+            },
+        }
+
+    # based on
+    # https://github.com/huggingface/diffusers/blob/3ebbaf7c96801271f9e6c21400033b6aa5ffcf29/src/diffusers/pipelines/stable_diffusion/pipeline_onnx_stable_diffusion.py#L375
+    @torch.no_grad()
+    def invoke(self, context: InvocationContext) -> LatentsOutput:
+        latents = context.services.latents.get(self.noise.latents_name)
+
+        positive_cond_data = context.services.latents.get(self.positive_conditioning.conditioning_name)
+        prompt_embeds = positive_cond_data.conditionings[0].embeds
+        pooled_prompt_embeds = positive_cond_data.conditionings[0].pooled_embeds
+        add_time_ids = positive_cond_data.conditionings[0].add_time_ids
+
+        negative_cond_data = context.services.latents.get(self.negative_conditioning.conditioning_name)
+        negative_prompt_embeds = negative_cond_data.conditionings[0].embeds
+        negative_pooled_prompt_embeds = negative_cond_data.conditionings[0].pooled_embeds
+        add_neg_time_ids = negative_cond_data.conditionings[0].add_time_ids
+
+        scheduler = get_scheduler(
+            context=context,
+            scheduler_info=self.unet.scheduler,
+            scheduler_name=self.scheduler,
+        )
+
+        num_inference_steps = self.steps
+        scheduler.set_timesteps(num_inference_steps)
+        timesteps = scheduler.timesteps
+
+        latents = latents * scheduler.init_noise_sigma
+
+
+        unet_info = context.services.model_manager.get_model(
+            **self.unet.unet.dict()
+        )
+        do_classifier_free_guidance = True
+        cross_attention_kwargs = None
+        with unet_info as unet:
+
+            extra_step_kwargs = dict()
+            if "eta" in set(inspect.signature(scheduler.step).parameters.keys()):
+                extra_step_kwargs.update(
+                    eta=0.0,
+                )
+            if "generator" in set(inspect.signature(scheduler.step).parameters.keys()):
+                extra_step_kwargs.update(
+                    generator=torch.Generator(device=unet.device).manual_seed(0),
+                )
+
+            num_warmup_steps = len(timesteps) - self.steps * scheduler.order
+
+            # apply denoising_end
+            skipped_final_steps = int(round((1 - self.denoising_end) * self.steps))
+            num_inference_steps = num_inference_steps - skipped_final_steps
+            timesteps = timesteps[: num_warmup_steps + scheduler.order * num_inference_steps]
+
+            if not context.services.configuration.sequential_guidance:
+                prompt_embeds = torch.cat([negative_prompt_embeds, prompt_embeds], dim=0)
+                add_text_embeds = torch.cat([negative_pooled_prompt_embeds, pooled_prompt_embeds], dim=0)
+                add_time_ids = torch.cat([add_neg_time_ids, add_time_ids], dim=0)
+
+                prompt_embeds = prompt_embeds.to(device=unet.device, dtype=unet.dtype)
+                add_text_embeds = add_text_embeds.to(device=unet.device, dtype=unet.dtype)
+                add_time_ids = add_time_ids.to(device=unet.device, dtype=unet.dtype)
+                latents = latents.to(device=unet.device, dtype=unet.dtype)
+
+                with tqdm(total=num_inference_steps) as progress_bar:
+                    for i, t in enumerate(timesteps):
+                        # expand the latents if we are doing classifier free guidance
+                        latent_model_input = torch.cat([latents] * 2) if do_classifier_free_guidance else latents
+
+                        latent_model_input = scheduler.scale_model_input(latent_model_input, t)
+
+                        # predict the noise residual
+                        added_cond_kwargs = {"text_embeds": add_text_embeds, "time_ids": add_time_ids}
+                        noise_pred = unet(
+                            latent_model_input,
+                            t,
+                            encoder_hidden_states=prompt_embeds,
+                            cross_attention_kwargs=cross_attention_kwargs,
+                            added_cond_kwargs=added_cond_kwargs,
+                            return_dict=False,
+                        )[0]
+
+                        # perform guidance
+                        if do_classifier_free_guidance:
+                            noise_pred_uncond, noise_pred_text = noise_pred.chunk(2)
+                            noise_pred = noise_pred_uncond + self.cfg_scale * (noise_pred_text - noise_pred_uncond)
+                            #del noise_pred_uncond
+                            #del noise_pred_text
+
+                        #if do_classifier_free_guidance and guidance_rescale > 0.0:
+                        #    # Based on 3.4. in https://arxiv.org/pdf/2305.08891.pdf
+                        #    noise_pred = rescale_noise_cfg(noise_pred, noise_pred_text, guidance_rescale=guidance_rescale)
+
+                        # compute the previous noisy sample x_t -> x_t-1
+                        latents = scheduler.step(noise_pred, t, latents, **extra_step_kwargs, return_dict=False)[0]
+
+                        # call the callback, if provided
+                        if i == len(timesteps) - 1 or ((i + 1) > num_warmup_steps and (i + 1) % scheduler.order == 0):
+                            progress_bar.update()
+                            #if callback is not None and i % callback_steps == 0:
+                            #    callback(i, t, latents)
+            else:
+                negative_pooled_prompt_embeds = negative_pooled_prompt_embeds.to(device=unet.device, dtype=unet.dtype)
+                negative_prompt_embeds = negative_prompt_embeds.to(device=unet.device, dtype=unet.dtype)
+                add_neg_time_ids = add_neg_time_ids.to(device=unet.device, dtype=unet.dtype)
+                pooled_prompt_embeds = pooled_prompt_embeds.to(device=unet.device, dtype=unet.dtype)
+                prompt_embeds = prompt_embeds.to(device=unet.device, dtype=unet.dtype)
+                add_time_ids = add_time_ids.to(device=unet.device, dtype=unet.dtype)
+                latents = latents.to(device=unet.device, dtype=unet.dtype)
+
+                with tqdm(total=num_inference_steps) as progress_bar:
+                    for i, t in enumerate(timesteps):
+                        # expand the latents if we are doing classifier free guidance
+                        #latent_model_input = torch.cat([latents] * 2) if do_classifier_free_guidance else latents
+
+                        latent_model_input = scheduler.scale_model_input(latents, t)
+
+                        #import gc
+                        #gc.collect()
+                        #torch.cuda.empty_cache()
+
+                        # predict the noise residual
+
+                        added_cond_kwargs = {"text_embeds": negative_pooled_prompt_embeds, "time_ids": add_neg_time_ids}
+                        noise_pred_uncond = unet(
+                            latent_model_input,
+                            t,
+                            encoder_hidden_states=negative_prompt_embeds,
+                            cross_attention_kwargs=cross_attention_kwargs,
+                            added_cond_kwargs=added_cond_kwargs,
+                            return_dict=False,
+                        )[0]
+
+                        added_cond_kwargs = {"text_embeds": pooled_prompt_embeds, "time_ids": add_time_ids}
+                        noise_pred_text = unet(
+                            latent_model_input,
+                            t,
+                            encoder_hidden_states=prompt_embeds,
+                            cross_attention_kwargs=cross_attention_kwargs,
+                            added_cond_kwargs=added_cond_kwargs,
+                            return_dict=False,
+                        )[0]
+
+                        # perform guidance
+                        noise_pred = noise_pred_uncond + self.cfg_scale * (noise_pred_text - noise_pred_uncond)
+
+                        #del noise_pred_text
+                        #del noise_pred_uncond
+                        #import gc
+                        #gc.collect()
+                        #torch.cuda.empty_cache()
+
+                        #if do_classifier_free_guidance and guidance_rescale > 0.0:
+                        #    # Based on 3.4. in https://arxiv.org/pdf/2305.08891.pdf
+                        #    noise_pred = rescale_noise_cfg(noise_pred, noise_pred_text, guidance_rescale=guidance_rescale)
+
+                        # compute the previous noisy sample x_t -> x_t-1
+                        latents = scheduler.step(noise_pred, t, latents, **extra_step_kwargs, return_dict=False)[0]
+
+                        #del noise_pred
+                        #import gc
+                        #gc.collect()
+                        #torch.cuda.empty_cache()
+
+                        # call the callback, if provided
+                        if i == len(timesteps) - 1 or ((i + 1) > num_warmup_steps and (i + 1) % scheduler.order == 0):
+                            progress_bar.update()
+                            #if callback is not None and i % callback_steps == 0:
+                            #    callback(i, t, latents)
+
+
+
+        #################
+
+        latents = latents.to("cpu")
+        torch.cuda.empty_cache()
+
+        name = f'{context.graph_execution_state_id}__{self.id}'
+        context.services.latents.save(name, latents)
+        return build_latents_output(latents_name=name, latents=latents)
+
+class SDXLLatentsToLatentsInvocation(BaseInvocation):
+    """Generates latents from conditionings."""
+
+    type: Literal["l2l_sdxl"] = "l2l_sdxl"
+
+    # Inputs
+    # fmt: off
+    positive_conditioning: Optional[ConditioningField] = Field(description="Positive conditioning for generation")
+    negative_conditioning: Optional[ConditioningField] = Field(description="Negative conditioning for generation")
+    noise: Optional[LatentsField] = Field(description="The noise to use")
+    steps:       int = Field(default=10, gt=0, description="The number of steps to use to generate the image")
+    cfg_scale: Union[float, List[float]] = Field(default=7.5, ge=1, description="The Classifier-Free Guidance, higher values may result in a result closer to the prompt", )
+    scheduler: SAMPLER_NAME_VALUES = Field(default="euler", description="The scheduler to use" )
+    unet: UNetField = Field(default=None, description="UNet submodel")
+    latents: Optional[LatentsField] = Field(description="Initial latents")
+
+    denoising_start: float = Field(default=0.0, ge=0, lt=1, description="")
+    denoising_end: float = Field(default=1.0, gt=0, le=1, description="")
+
+    #control: Union[ControlField, list[ControlField]] = Field(default=None, description="The control to use")
+    #seamless:   bool = Field(default=False, description="Whether or not to generate an image that can tile without seams", )
+    #seamless_axes: str = Field(default="", description="The axes to tile the image on, 'x' and/or 'y'")
+    # fmt: on
+
+    @validator("cfg_scale")
+    def ge_one(cls, v):
+        """validate that all cfg_scale values are >= 1"""
+        if isinstance(v, list):
+            for i in v:
+                if i < 1:
+                    raise ValueError('cfg_scale must be greater than 1')
+        else:
+            if v < 1:
+                raise ValueError('cfg_scale must be greater than 1')
+        return v
+
+    # Schema customisation
+    class Config(InvocationConfig):
+        schema_extra = {
+            "ui": {
+                "title": "SDXL Latents to Latents",
+                "tags": ["latents"],
+                "type_hints": {
+                  "model": "model",
+                  # "cfg_scale": "float",
+                  "cfg_scale": "number"
+                }
+            },
+        }
+
+    # based on
+    # https://github.com/huggingface/diffusers/blob/3ebbaf7c96801271f9e6c21400033b6aa5ffcf29/src/diffusers/pipelines/stable_diffusion/pipeline_onnx_stable_diffusion.py#L375
+    @torch.no_grad()
+    def invoke(self, context: InvocationContext) -> LatentsOutput:
+        latents = context.services.latents.get(self.latents.latents_name)
+
+        positive_cond_data = context.services.latents.get(self.positive_conditioning.conditioning_name)
+        prompt_embeds = positive_cond_data.conditionings[0].embeds
+        pooled_prompt_embeds = positive_cond_data.conditionings[0].pooled_embeds
+        add_time_ids = positive_cond_data.conditionings[0].add_time_ids
+
+        negative_cond_data = context.services.latents.get(self.negative_conditioning.conditioning_name)
+        negative_prompt_embeds = negative_cond_data.conditionings[0].embeds
+        negative_pooled_prompt_embeds = negative_cond_data.conditionings[0].pooled_embeds
+        add_neg_time_ids = negative_cond_data.conditionings[0].add_time_ids
+
+        scheduler = get_scheduler(
+            context=context,
+            scheduler_info=self.unet.scheduler,
+            scheduler_name=self.scheduler,
+        )
+
+        # apply denoising_start
+        num_inference_steps = self.steps
+        scheduler.set_timesteps(num_inference_steps)
+
+        t_start = int(round(self.denoising_start * num_inference_steps))
+        timesteps = scheduler.timesteps[t_start * scheduler.order:]
+        num_inference_steps = num_inference_steps - t_start
+
+        # apply noise(if provided)
+        if self.noise is not None:
+            noise = context.services.latents.get(self.noise.latents_name)
+            latents = scheduler.add_noise(latents, noise, timesteps[:1])
+            del noise
+
+        unet_info = context.services.model_manager.get_model(
+            **self.unet.unet.dict()
+        )
+        do_classifier_free_guidance = True
+        cross_attention_kwargs = None
+        with unet_info as unet:
+
+            # apply scheduler extra args
+            extra_step_kwargs = dict()
+            if "eta" in set(inspect.signature(scheduler.step).parameters.keys()):
+                extra_step_kwargs.update(
+                    eta=0.0,
+                )
+            if "generator" in set(inspect.signature(scheduler.step).parameters.keys()):
+                extra_step_kwargs.update(
+                    generator=torch.Generator(device=unet.device).manual_seed(0),
+                )
+
+            num_warmup_steps = max(len(timesteps) - num_inference_steps * scheduler.order, 0)
+
+            # apply denoising_end
+            skipped_final_steps = int(round((1 - self.denoising_end) * self.steps))
+            num_inference_steps = num_inference_steps - skipped_final_steps
+            timesteps = timesteps[: num_warmup_steps + scheduler.order * num_inference_steps]
+
+            if not context.services.configuration.sequential_guidance:
+                prompt_embeds = torch.cat([negative_prompt_embeds, prompt_embeds], dim=0)
+                add_text_embeds = torch.cat([negative_pooled_prompt_embeds, pooled_prompt_embeds], dim=0)
+                add_time_ids = torch.cat([add_neg_time_ids, add_time_ids], dim=0)
+
+                prompt_embeds = prompt_embeds.to(device=unet.device, dtype=unet.dtype)
+                add_text_embeds = add_text_embeds.to(device=unet.device, dtype=unet.dtype)
+                add_time_ids = add_time_ids.to(device=unet.device, dtype=unet.dtype)
+                latents = latents.to(device=unet.device, dtype=unet.dtype)
+
+                with tqdm(total=num_inference_steps) as progress_bar:
+                    for i, t in enumerate(timesteps):
+                        # expand the latents if we are doing classifier free guidance
+                        latent_model_input = torch.cat([latents] * 2) if do_classifier_free_guidance else latents
+
+                        latent_model_input = scheduler.scale_model_input(latent_model_input, t)
+
+                        # predict the noise residual
+                        added_cond_kwargs = {"text_embeds": add_text_embeds, "time_ids": add_time_ids}
+                        noise_pred = unet(
+                            latent_model_input,
+                            t,
+                            encoder_hidden_states=prompt_embeds,
+                            cross_attention_kwargs=cross_attention_kwargs,
+                            added_cond_kwargs=added_cond_kwargs,
+                            return_dict=False,
+                        )[0]
+
+                        # perform guidance
+                        if do_classifier_free_guidance:
+                            noise_pred_uncond, noise_pred_text = noise_pred.chunk(2)
+                            noise_pred = noise_pred_uncond + self.cfg_scale * (noise_pred_text - noise_pred_uncond)
+                            #del noise_pred_uncond
+                            #del noise_pred_text
+
+                        #if do_classifier_free_guidance and guidance_rescale > 0.0:
+                        #    # Based on 3.4. in https://arxiv.org/pdf/2305.08891.pdf
+                        #    noise_pred = rescale_noise_cfg(noise_pred, noise_pred_text, guidance_rescale=guidance_rescale)
+
+                        # compute the previous noisy sample x_t -> x_t-1
+                        latents = scheduler.step(noise_pred, t, latents, **extra_step_kwargs, return_dict=False)[0]
+
+                        # call the callback, if provided
+                        if i == len(timesteps) - 1 or ((i + 1) > num_warmup_steps and (i + 1) % scheduler.order == 0):
+                            progress_bar.update()
+                            #if callback is not None and i % callback_steps == 0:
+                            #    callback(i, t, latents)
+            else:
+                negative_pooled_prompt_embeds = negative_pooled_prompt_embeds.to(device=unet.device, dtype=unet.dtype)
+                negative_prompt_embeds = negative_prompt_embeds.to(device=unet.device, dtype=unet.dtype)
+                add_neg_time_ids = add_neg_time_ids.to(device=unet.device, dtype=unet.dtype)
+                pooled_prompt_embeds = pooled_prompt_embeds.to(device=unet.device, dtype=unet.dtype)
+                prompt_embeds = prompt_embeds.to(device=unet.device, dtype=unet.dtype)
+                add_time_ids = add_time_ids.to(device=unet.device, dtype=unet.dtype)
+                latents = latents.to(device=unet.device, dtype=unet.dtype)
+
+                with tqdm(total=num_inference_steps) as progress_bar:
+                    for i, t in enumerate(timesteps):
+                        # expand the latents if we are doing classifier free guidance
+                        #latent_model_input = torch.cat([latents] * 2) if do_classifier_free_guidance else latents
+
+                        latent_model_input = scheduler.scale_model_input(latents, t)
+
+                        #import gc
+                        #gc.collect()
+                        #torch.cuda.empty_cache()
+
+                        # predict the noise residual
+
+                        added_cond_kwargs = {"text_embeds": negative_pooled_prompt_embeds, "time_ids": add_time_ids}
+                        noise_pred_uncond = unet(
+                            latent_model_input,
+                            t,
+                            encoder_hidden_states=negative_prompt_embeds,
+                            cross_attention_kwargs=cross_attention_kwargs,
+                            added_cond_kwargs=added_cond_kwargs,
+                            return_dict=False,
+                        )[0]
+
+                        added_cond_kwargs = {"text_embeds": pooled_prompt_embeds, "time_ids": add_time_ids}
+                        noise_pred_text = unet(
+                            latent_model_input,
+                            t,
+                            encoder_hidden_states=prompt_embeds,
+                            cross_attention_kwargs=cross_attention_kwargs,
+                            added_cond_kwargs=added_cond_kwargs,
+                            return_dict=False,
+                        )[0]
+
+                        # perform guidance
+                        noise_pred = noise_pred_uncond + self.cfg_scale * (noise_pred_text - noise_pred_uncond)
+
+                        #del noise_pred_text
+                        #del noise_pred_uncond
+                        #import gc
+                        #gc.collect()
+                        #torch.cuda.empty_cache()
+
+                        #if do_classifier_free_guidance and guidance_rescale > 0.0:
+                        #    # Based on 3.4. in https://arxiv.org/pdf/2305.08891.pdf
+                        #    noise_pred = rescale_noise_cfg(noise_pred, noise_pred_text, guidance_rescale=guidance_rescale)
+
+                        # compute the previous noisy sample x_t -> x_t-1
+                        latents = scheduler.step(noise_pred, t, latents, **extra_step_kwargs, return_dict=False)[0]
+
+                        #del noise_pred
+                        #import gc
+                        #gc.collect()
+                        #torch.cuda.empty_cache()
+
+                        # call the callback, if provided
+                        if i == len(timesteps) - 1 or ((i + 1) > num_warmup_steps and (i + 1) % scheduler.order == 0):
+                            progress_bar.update()
+                            #if callback is not None and i % callback_steps == 0:
+                            #    callback(i, t, latents)
+
+
+
+        #################
+
+        latents = latents.to("cpu")
+        torch.cuda.empty_cache()
+
+        name = f'{context.graph_execution_state_id}__{self.id}'
+        context.services.latents.save(name, latents)
+        return build_latents_output(latents_name=name, latents=latents)

invokeai/app/invocations/upscale.py

@@ -1,48 +1,119 @@
-# Copyright (c) 2022 Kyle Schouviller (https://github.com/kyle0654)
-
-from typing import Literal, Optional
+# Copyright (c) 2022 Kyle Schouviller (https://github.com/kyle0654) & the InvokeAI Team
+from pathlib import Path
+from typing import Literal, Union
 
+import cv2 as cv
+import numpy as np
+from basicsr.archs.rrdbnet_arch import RRDBNet
+from PIL import Image
 from pydantic import Field
+from realesrgan import RealESRGANer
 
 from invokeai.app.models.image import ImageCategory, ImageField, ResourceOrigin
-from .baseinvocation import BaseInvocation, InvocationContext, InvocationConfig
+
+from .baseinvocation import BaseInvocation, InvocationConfig, InvocationContext
 from .image import ImageOutput
 
+# TODO: Populate this from disk?
+# TODO: Use model manager to load?
+ESRGAN_MODELS = Literal[
+    "RealESRGAN_x4plus.pth",
+    "RealESRGAN_x4plus_anime_6B.pth",
+    "ESRGAN_SRx4_DF2KOST_official-ff704c30.pth",
+    "RealESRGAN_x2plus.pth",
+]
 
-class UpscaleInvocation(BaseInvocation):
-    """Upscales an image."""
 
-    # fmt: off
-    type: Literal["upscale"] = "upscale"
+class ESRGANInvocation(BaseInvocation):
+    """Upscales an image using RealESRGAN."""
 
-    # Inputs
-    image: Optional[ImageField] = Field(description="The input image", default=None)
-    strength: float = Field(default=0.75, gt=0, le=1, description="The strength")
-    level: Literal[2, 4] = Field(default=2, description="The upscale level")
-    # fmt: on
+    type: Literal["esrgan"] = "esrgan"
+    image: Union[ImageField, None] = Field(default=None, description="The input image")
+    model_name: ESRGAN_MODELS = Field(
+        default="RealESRGAN_x4plus.pth", description="The Real-ESRGAN model to use"
+    )
 
-    # Schema customisation
     class Config(InvocationConfig):
         schema_extra = {
             "ui": {
-                "tags": ["upscaling", "image"],
+                "title": "Upscale (RealESRGAN)",
+                "tags": ["image", "upscale", "realesrgan"]
             },
         }
 
     def invoke(self, context: InvocationContext) -> ImageOutput:
         image = context.services.images.get_pil_image(self.image.image_name)
-        results = context.services.restoration.upscale_and_reconstruct(
-            image_list=[[image, 0]],
-            upscale=(self.level, self.strength),
-            strength=0.0,  # GFPGAN strength
-            save_original=False,
-            image_callback=None,
+        models_path = context.services.configuration.models_path
+
+        rrdbnet_model = None
+        netscale = None
+        esrgan_model_path = None
+
+        if self.model_name in [
+            "RealESRGAN_x4plus.pth",
+            "ESRGAN_SRx4_DF2KOST_official-ff704c30.pth",
+        ]:
+            # x4 RRDBNet model
+            rrdbnet_model = RRDBNet(
+                num_in_ch=3,
+                num_out_ch=3,
+                num_feat=64,
+                num_block=23,
+                num_grow_ch=32,
+                scale=4,
+            )
+            netscale = 4
+        elif self.model_name in ["RealESRGAN_x4plus_anime_6B.pth"]:
+            # x4 RRDBNet model, 6 blocks
+            rrdbnet_model = RRDBNet(
+                num_in_ch=3,
+                num_out_ch=3,
+                num_feat=64,
+                num_block=6,  # 6 blocks
+                num_grow_ch=32,
+                scale=4,
+            )
+            netscale = 4
+        elif self.model_name in ["RealESRGAN_x2plus.pth"]:
+            # x2 RRDBNet model
+            rrdbnet_model = RRDBNet(
+                num_in_ch=3,
+                num_out_ch=3,
+                num_feat=64,
+                num_block=23,
+                num_grow_ch=32,
+                scale=2,
+            )
+            netscale = 2
+        else:
+            msg = f"Invalid RealESRGAN model: {self.model_name}"
+            context.services.logger.error(msg)
+            raise ValueError(msg)
+
+        esrgan_model_path = Path(f"core/upscaling/realesrgan/{self.model_name}")
+
+        upsampler = RealESRGANer(
+            scale=netscale,
+            model_path=str(models_path / esrgan_model_path),
+            model=rrdbnet_model,
+            half=False,
         )
 
-        # Results are image and seed, unwrap for now
-        # TODO: can this return multiple results?
+        # prepare image - Real-ESRGAN uses cv2 internally, and cv2 uses BGR vs RGB for PIL
+        cv_image = cv.cvtColor(np.array(image.convert("RGB")), cv.COLOR_RGB2BGR)
+
+        # We can pass an `outscale` value here, but it just resizes the image by that factor after
+        # upscaling, so it's kinda pointless for our purposes. If you want something other than 4x
+        # upscaling, you'll need to add a resize node after this one.
+        upscaled_image, img_mode = upsampler.enhance(cv_image)
+
+        # back to PIL
+        pil_image = Image.fromarray(
+            cv.cvtColor(upscaled_image, cv.COLOR_BGR2RGB)
+        ).convert("RGBA")
+
         image_dto = context.services.images.create(
-            image=results[0][0],
+            image=pil_image,
             image_origin=ResourceOrigin.INTERNAL,
             image_category=ImageCategory.GENERAL,
             node_id=self.id,

invokeai/app/models/metadata.py

@@ -1,93 +0,0 @@
-from typing import Optional, Union, List
-from pydantic import BaseModel, Extra, Field, StrictFloat, StrictInt, StrictStr
-
-
-class ImageMetadata(BaseModel):
-    """
-    Core generation metadata for an image/tensor generated in InvokeAI.
-
-    Also includes any metadata from the image's PNG tEXt chunks.
-
-    Generated by traversing the execution graph, collecting the parameters of the nearest ancestors
-    of a given node.
-
-    Full metadata may be accessed by querying for the session in the `graph_executions` table.
-    """
-
-    class Config:
-        extra = Extra.allow
-        """
-        This lets the ImageMetadata class accept arbitrary additional fields. The CoreMetadataService
-        won't add any fields that are not already defined, but other a different metadata service
-        implementation might.
-        """
-
-    type: Optional[StrictStr] = Field(
-        default=None,
-        description="The type of the ancestor node of the image output node.",
-    )
-    """The type of the ancestor node of the image output node."""
-    positive_conditioning: Optional[StrictStr] = Field(
-        default=None, description="The positive conditioning."
-    )
-    """The positive conditioning"""
-    negative_conditioning: Optional[StrictStr] = Field(
-        default=None, description="The negative conditioning."
-    )
-    """The negative conditioning"""
-    width: Optional[StrictInt] = Field(
-        default=None, description="Width of the image/latents in pixels."
-    )
-    """Width of the image/latents in pixels"""
-    height: Optional[StrictInt] = Field(
-        default=None, description="Height of the image/latents in pixels."
-    )
-    """Height of the image/latents in pixels"""
-    seed: Optional[StrictInt] = Field(
-        default=None, description="The seed used for noise generation."
-    )
-    """The seed used for noise generation"""
-    # cfg_scale: Optional[StrictFloat] = Field(
-    # cfg_scale: Union[float, list[float]] = Field(
-    cfg_scale: Union[StrictFloat, List[StrictFloat]] = Field(
-        default=None, description="The classifier-free guidance scale."
-    )
-    """The classifier-free guidance scale"""
-    steps: Optional[StrictInt] = Field(
-        default=None, description="The number of steps used for inference."
-    )
-    """The number of steps used for inference"""
-    scheduler: Optional[StrictStr] = Field(
-        default=None, description="The scheduler used for inference."
-    )
-    """The scheduler used for inference"""
-    model: Optional[StrictStr] = Field(
-        default=None, description="The model used for inference."
-    )
-    """The model used for inference"""
-    strength: Optional[StrictFloat] = Field(
-        default=None,
-        description="The strength used for image-to-image/latents-to-latents.",
-    )
-    """The strength used for image-to-image/latents-to-latents."""
-    latents: Optional[StrictStr] = Field(
-        default=None, description="The ID of the initial latents."
-    )
-    """The ID of the initial latents"""
-    vae: Optional[StrictStr] = Field(
-        default=None, description="The VAE used for decoding."
-    )
-    """The VAE used for decoding"""
-    unet: Optional[StrictStr] = Field(
-        default=None, description="The UNet used dor inference."
-    )
-    """The UNet used dor inference"""
-    clip: Optional[StrictStr] = Field(
-        default=None, description="The CLIP Encoder used for conditioning."
-    )
-    """The CLIP Encoder used for conditioning"""
-    extra: Optional[StrictStr] = Field(
-        default=None,
-        description="Uploaded image metadata, extracted from the PNG tEXt chunk.",
-    )
-    """Uploaded image metadata, extracted from the PNG tEXt chunk."""

invokeai/app/services/config.py

@@ -23,7 +23,8 @@ InvokeAI:
     xformers_enabled: false
     sequential_guidance: false
     precision: float16
-    max_loaded_models: 4
+    max_cache_size: 6
+    max_vram_cache_size: 2.7
     always_use_cpu: false
     free_gpu_mem: false
   Features:
@@ -168,7 +169,7 @@ from argparse import ArgumentParser
 from omegaconf import OmegaConf, DictConfig
 from pathlib import Path
 from pydantic import BaseSettings, Field, parse_obj_as
-from typing import ClassVar, Dict, List, Literal, Union, get_origin, get_type_hints, get_args
+from typing import ClassVar, Dict, List, Set, Literal, Union, get_origin, get_type_hints, get_args
 
 INIT_FILE = Path('invokeai.yaml')
 MODEL_CORE = Path('models/core')
@@ -199,7 +200,7 @@ class InvokeAISettings(BaseSettings):
         type = get_args(get_type_hints(cls)['type'])[0]
         field_dict = dict({type:dict()})
         for name,field in self.__fields__.items():
-            if name in cls._excluded():
+            if name in cls._excluded_from_yaml():
                 continue
             category = field.field_info.extra.get("category") or "Uncategorized"
             value = getattr(self,name)
@@ -270,7 +271,13 @@ class InvokeAISettings(BaseSettings):
 
     @classmethod
     def _excluded(self)->List[str]:
+        # internal fields that shouldn't be exposed as command line options
         return ['type','initconf']
+    
+    @classmethod
+    def _excluded_from_yaml(self)->List[str]:
+        # combination of deprecated parameters and internal ones that shouldn't be exposed as invokeai.yaml options
+        return ['type','initconf', 'gpu_mem_reserved', 'max_loaded_models', 'version', 'from_file', 'model', 'restore']
 
     class Config:
         env_file_encoding = 'utf-8'
@@ -359,12 +366,14 @@ setting environment variables INVOKEAI_<setting>.
     log_tokenization    : bool = Field(default=False, description="Enable logging of parsed prompt tokens.", category='Features')
     nsfw_checker        : bool = Field(default=True, description="Enable/disable the NSFW checker", category='Features')
     patchmatch          : bool = Field(default=True, description="Enable/disable patchmatch inpaint code", category='Features')
-    restore             : bool = Field(default=True, description="Enable/disable face restoration code", category='Features')
+    restore             : bool = Field(default=True, description="Enable/disable face restoration code (DEPRECATED)", category='DEPRECATED')
 
     always_use_cpu      : bool = Field(default=False, description="If true, use the CPU for rendering even if a GPU is available.", category='Memory/Performance')
     free_gpu_mem        : bool = Field(default=False, description="If true, purge model from GPU after each generation.", category='Memory/Performance')
-    max_loaded_models   : int = Field(default=3, gt=0, description="(DEPRECATED: use max_cache_size) Maximum number of models to keep in memory for rapid switching", category='Memory/Performance')
+    max_loaded_models   : int = Field(default=3, gt=0, description="(DEPRECATED: use max_cache_size) Maximum number of models to keep in memory for rapid switching", category='DEPRECATED')
     max_cache_size      : float = Field(default=6.0, gt=0, description="Maximum memory amount used by model cache for rapid switching", category='Memory/Performance')
+    max_vram_cache_size : float = Field(default=2.75, ge=0, description="Amount of VRAM reserved for model storage", category='Memory/Performance')
+    gpu_mem_reserved    : float = Field(default=2.75, ge=0, description="DEPRECATED: use max_vram_cache_size. Amount of VRAM reserved for model storage", category='DEPRECATED')
     precision           : Literal[tuple(['auto','float16','float32','autocast'])] = Field(default='float16',description='Floating point precision', category='Memory/Performance')
     sequential_guidance : bool = Field(default=False, description="Whether to calculate guidance in serial instead of in parallel, lowering memory requirements", category='Memory/Performance')
     xformers_enabled    : bool = Field(default=True, description="Enable/disable memory-efficient attention", category='Memory/Performance')
@@ -389,6 +398,8 @@ setting environment variables INVOKEAI_<setting>.
     # note - would be better to read the log_format values from logging.py, but this creates circular dependencies issues
     log_format          : Literal[tuple(['plain','color','syslog','legacy'])] = Field(default="color", description='Log format. Use "plain" for text-only, "color" for colorized output, "legacy" for 2.3-style logging and "syslog" for syslog-style', category="Logging")
     log_level           : Literal[tuple(["debug","info","warning","error","critical"])] = Field(default="debug", description="Emit logging messages at this level or  higher", category="Logging")
+
+    version             : bool = Field(default=False, description="Show InvokeAI version and exit", category="Other")
     #fmt: on
 
     def parse_args(self, argv: List[str]=None, conf: DictConfig = None, clobber=False):

invokeai/app/services/events.py

@@ -105,8 +105,6 @@ class EventServiceBase:
     def emit_model_load_started (
             self,
             graph_execution_state_id: str,
-            node: dict,
-            source_node_id: str,
             model_name: str,
             base_model: BaseModelType,
             model_type: ModelType,
@@ -117,8 +115,6 @@ class EventServiceBase:
             event_name="model_load_started",
             payload=dict(
                 graph_execution_state_id=graph_execution_state_id,
-                node=node,
-                source_node_id=source_node_id,
                 model_name=model_name,
                 base_model=base_model,
                 model_type=model_type,
@@ -129,8 +125,6 @@ class EventServiceBase:
     def emit_model_load_completed(
             self,
             graph_execution_state_id: str,
-            node: dict,
-            source_node_id: str,
             model_name: str,
             base_model: BaseModelType,
             model_type: ModelType,
@@ -142,12 +136,12 @@ class EventServiceBase:
             event_name="model_load_completed",
             payload=dict(
                 graph_execution_state_id=graph_execution_state_id,
-                node=node,
-                source_node_id=source_node_id,
                 model_name=model_name,
                 base_model=base_model,
                 model_type=model_type,
                 submodel=submodel,
-                model_info=model_info,
+                hash=model_info.hash,
+                location=model_info.location,
+                precision=str(model_info.precision),
             ),
         )

invokeai/app/services/image_file_storage.py

@@ -1,14 +1,14 @@
 # Copyright (c) 2022 Kyle Schouviller (https://github.com/kyle0654) and the InvokeAI Team
+import json
 from abc import ABC, abstractmethod
 from pathlib import Path
 from queue import Queue
 from typing import Dict, Optional, Union
 
-from PIL.Image import Image as PILImageType
 from PIL import Image, PngImagePlugin
+from PIL.Image import Image as PILImageType
 from send2trash import send2trash
 
-from invokeai.app.models.metadata import ImageMetadata
 from invokeai.app.util.thumbnails import get_thumbnail_name, make_thumbnail
 
 
@@ -59,7 +59,8 @@ class ImageFileStorageBase(ABC):
         self,
         image: PILImageType,
         image_name: str,
-        metadata: Optional[ImageMetadata] = None,
+        metadata: Optional[dict] = None,
+        graph: Optional[dict] = None,
         thumbnail_size: int = 256,
     ) -> None:
         """Saves an image and a 256x256 WEBP thumbnail. Returns a tuple of the image name, thumbnail name, and created timestamp."""
@@ -110,20 +111,22 @@ class DiskImageFileStorage(ImageFileStorageBase):
         self,
         image: PILImageType,
         image_name: str,
-        metadata: Optional[ImageMetadata] = None,
+        metadata: Optional[dict] = None,
+        graph: Optional[dict] = None,
         thumbnail_size: int = 256,
     ) -> None:
         try:
             self.__validate_storage_folders()
             image_path = self.get_path(image_name)
 
+            pnginfo = PngImagePlugin.PngInfo()
+            
             if metadata is not None:
-                pnginfo = PngImagePlugin.PngInfo()
-                pnginfo.add_text("invokeai", metadata.json())
-                image.save(image_path, "PNG", pnginfo=pnginfo)
-            else:
-                image.save(image_path, "PNG")
+                pnginfo.add_text("invokeai_metadata", json.dumps(metadata))
+            if graph is not None:
+                pnginfo.add_text("invokeai_graph", json.dumps(graph))
 
+            image.save(image_path, "PNG", pnginfo=pnginfo)
             thumbnail_name = get_thumbnail_name(image_name)
             thumbnail_path = self.get_path(thumbnail_name, thumbnail=True)
             thumbnail_image = make_thumbnail(image, thumbnail_size)

invokeai/app/services/image_record_storage.py

@@ -1,22 +1,16 @@
+import json
+import sqlite3
+import threading
 from abc import ABC, abstractmethod
 from datetime import datetime
 from typing import Generic, Optional, TypeVar, cast
-import sqlite3
-import threading
 
 from pydantic import BaseModel, Field
 from pydantic.generics import GenericModel
 
-from invokeai.app.models.metadata import ImageMetadata
-from invokeai.app.models.image import (
-    ImageCategory,
-    ResourceOrigin,
-)
+from invokeai.app.models.image import ImageCategory, ResourceOrigin
 from invokeai.app.services.models.image_record import (
-    ImageRecord,
-    ImageRecordChanges,
-    deserialize_image_record,
-)
+    ImageRecord, ImageRecordChanges, deserialize_image_record)
 
 T = TypeVar("T", bound=BaseModel)
 
@@ -54,6 +48,28 @@ class ImageRecordDeleteException(Exception):
         super().__init__(message)
 
 
+IMAGE_DTO_COLS = ", ".join(
+    list(
+        map(
+            lambda c: "images." + c,
+            [
+                "image_name",
+                "image_origin",
+                "image_category",
+                "width",
+                "height",
+                "session_id",
+                "node_id",
+                "is_intermediate",
+                "created_at",
+                "updated_at",
+                "deleted_at",
+            ],
+        )
+    )
+)
+
+
 class ImageRecordStorageBase(ABC):
     """Low-level service responsible for interfacing with the image record store."""
 
@@ -64,6 +80,11 @@ class ImageRecordStorageBase(ABC):
         """Gets an image record."""
         pass
 
+    @abstractmethod
+    def get_metadata(self, image_name: str) -> Optional[dict]:
+        """Gets an image's metadata'."""
+        pass
+
     @abstractmethod
     def update(
         self,
@@ -108,7 +129,7 @@ class ImageRecordStorageBase(ABC):
         height: int,
         session_id: Optional[str],
         node_id: Optional[str],
-        metadata: Optional[ImageMetadata],
+        metadata: Optional[dict],
         is_intermediate: bool = False,
     ) -> datetime:
         """Saves an image record."""
@@ -162,7 +183,6 @@ class SqliteImageRecordStorage(ImageRecordStorageBase):
                 node_id TEXT,
                 metadata TEXT,
                 is_intermediate BOOLEAN DEFAULT FALSE,
-                board_id TEXT,
                 created_at DATETIME NOT NULL DEFAULT(STRFTIME('%Y-%m-%d %H:%M:%f', 'NOW')),
                 -- Updated via trigger
                 updated_at DATETIME NOT NULL DEFAULT(STRFTIME('%Y-%m-%d %H:%M:%f', 'NOW')),
@@ -213,7 +233,7 @@ class SqliteImageRecordStorage(ImageRecordStorageBase):
 
             self._cursor.execute(
                 f"""--sql
-                SELECT * FROM images
+                SELECT {IMAGE_DTO_COLS} FROM images
                 WHERE image_name = ?;
                 """,
                 (image_name,),
@@ -231,6 +251,28 @@ class SqliteImageRecordStorage(ImageRecordStorageBase):
 
         return deserialize_image_record(dict(result))
 
+    def get_metadata(self, image_name: str) -> Optional[dict]:
+        try:
+            self._lock.acquire()
+
+            self._cursor.execute(
+                f"""--sql
+                SELECT images.metadata FROM images
+                WHERE image_name = ?;
+                """,
+                (image_name,),
+            )
+
+            result = cast(Optional[sqlite3.Row], self._cursor.fetchone())
+            if not result or not result[0]:
+                return None
+            return json.loads(result[0])
+        except sqlite3.Error as e:
+            self._conn.rollback()
+            raise ImageRecordNotFoundException from e
+        finally:
+            self._lock.release()
+
     def update(
         self,
         image_name: str,
@@ -298,8 +340,8 @@ class SqliteImageRecordStorage(ImageRecordStorageBase):
             WHERE 1=1
             """
 
-            images_query = """--sql
-            SELECT images.*
+            images_query = f"""--sql
+            SELECT {IMAGE_DTO_COLS}
             FROM images
             LEFT JOIN board_images ON board_images.image_name = images.image_name
             WHERE 1=1
@@ -417,12 +459,12 @@ class SqliteImageRecordStorage(ImageRecordStorageBase):
         width: int,
         height: int,
         node_id: Optional[str],
-        metadata: Optional[ImageMetadata],
+        metadata: Optional[dict],
         is_intermediate: bool = False,
     ) -> datetime:
         try:
             metadata_json = (
-                None if metadata is None else metadata.json(exclude_none=True)
+                None if metadata is None else json.dumps(metadata)
             )
             self._lock.acquire()
             self._cursor.execute(
@@ -472,9 +514,7 @@ class SqliteImageRecordStorage(ImageRecordStorageBase):
         finally:
             self._lock.release()
 
-    def get_most_recent_image_for_board(
-        self, board_id: str
-    ) -> Optional[ImageRecord]:
+    def get_most_recent_image_for_board(self, board_id: str) -> Optional[ImageRecord]:
         try:
             self._lock.acquire()
             self._cursor.execute(

invokeai/app/services/images.py

@@ -1,39 +1,30 @@
+import json
 from abc import ABC, abstractmethod
 from logging import Logger
-from typing import Optional, TYPE_CHECKING, Union
+from typing import TYPE_CHECKING, Optional
+
 from PIL.Image import Image as PILImageType
 
-from invokeai.app.models.image import (
-    ImageCategory,
-    ResourceOrigin,
-    InvalidImageCategoryException,
-    InvalidOriginException,
-)
-from invokeai.app.models.metadata import ImageMetadata
-from invokeai.app.services.board_image_record_storage import BoardImageRecordStorageBase
-from invokeai.app.services.image_record_storage import (
-    ImageRecordDeleteException,
-    ImageRecordNotFoundException,
-    ImageRecordSaveException,
-    ImageRecordStorageBase,
-    OffsetPaginatedResults,
-)
-from invokeai.app.services.models.image_record import (
-    ImageRecord,
-    ImageDTO,
-    ImageRecordChanges,
-    image_record_to_dto,
-)
+from invokeai.app.invocations.metadata import ImageMetadata
+from invokeai.app.models.image import (ImageCategory,
+                                       InvalidImageCategoryException,
+                                       InvalidOriginException, ResourceOrigin)
+from invokeai.app.services.board_image_record_storage import \
+    BoardImageRecordStorageBase
+from invokeai.app.services.graph import Graph
 from invokeai.app.services.image_file_storage import (
-    ImageFileDeleteException,
-    ImageFileNotFoundException,
-    ImageFileSaveException,
-    ImageFileStorageBase,
-)
-from invokeai.app.services.item_storage import ItemStorageABC, PaginatedResults
-from invokeai.app.services.metadata import MetadataServiceBase
+    ImageFileDeleteException, ImageFileNotFoundException,
+    ImageFileSaveException, ImageFileStorageBase)
+from invokeai.app.services.image_record_storage import (
+    ImageRecordDeleteException, ImageRecordNotFoundException,
+    ImageRecordSaveException, ImageRecordStorageBase, OffsetPaginatedResults)
+from invokeai.app.services.item_storage import ItemStorageABC
+from invokeai.app.services.models.image_record import (ImageDTO, ImageRecord,
+                                                       ImageRecordChanges,
+                                                       image_record_to_dto)
 from invokeai.app.services.resource_name import NameServiceBase
 from invokeai.app.services.urls import UrlServiceBase
+from invokeai.app.util.metadata import get_metadata_graph_from_raw_session
 
 if TYPE_CHECKING:
     from invokeai.app.services.graph import GraphExecutionState
@@ -51,6 +42,7 @@ class ImageServiceABC(ABC):
         node_id: Optional[str] = None,
         session_id: Optional[str] = None,
         is_intermediate: bool = False,
+        metadata: Optional[dict] = None,
     ) -> ImageDTO:
         """Creates an image, storing the file and its metadata."""
         pass
@@ -79,6 +71,11 @@ class ImageServiceABC(ABC):
         """Gets an image DTO."""
         pass
 
+    @abstractmethod
+    def get_metadata(self, image_name: str) -> ImageMetadata:
+        """Gets an image's metadata."""
+        pass
+
     @abstractmethod
     def get_path(self, image_name: str, thumbnail: bool = False) -> str:
         """Gets an image's path."""
@@ -124,7 +121,6 @@ class ImageServiceDependencies:
     image_records: ImageRecordStorageBase
     image_files: ImageFileStorageBase
     board_image_records: BoardImageRecordStorageBase
-    metadata: MetadataServiceBase
     urls: UrlServiceBase
     logger: Logger
     names: NameServiceBase
@@ -135,7 +131,6 @@ class ImageServiceDependencies:
         image_record_storage: ImageRecordStorageBase,
         image_file_storage: ImageFileStorageBase,
         board_image_record_storage: BoardImageRecordStorageBase,
-        metadata: MetadataServiceBase,
         url: UrlServiceBase,
         logger: Logger,
         names: NameServiceBase,
@@ -144,7 +139,6 @@ class ImageServiceDependencies:
         self.image_records = image_record_storage
         self.image_files = image_file_storage
         self.board_image_records = board_image_record_storage
-        self.metadata = metadata
         self.urls = url
         self.logger = logger
         self.names = names
@@ -165,6 +159,7 @@ class ImageService(ImageServiceABC):
         node_id: Optional[str] = None,
         session_id: Optional[str] = None,
         is_intermediate: bool = False,
+        metadata: Optional[dict] = None,
     ) -> ImageDTO:
         if image_origin not in ResourceOrigin:
             raise InvalidOriginException
@@ -174,7 +169,16 @@ class ImageService(ImageServiceABC):
 
         image_name = self._services.names.create_image_name()
 
-        metadata = self._get_metadata(session_id, node_id)
+        graph = None
+
+        if session_id is not None:
+            session_raw = self._services.graph_execution_manager.get_raw(session_id)
+            if session_raw is not None:
+                try:
+                    graph = get_metadata_graph_from_raw_session(session_raw)
+                except Exception as e:
+                    self._services.logger.warn(f"Failed to parse session graph: {e}")
+                    graph = None
 
         (width, height) = image.size
 
@@ -191,14 +195,12 @@ class ImageService(ImageServiceABC):
                 is_intermediate=is_intermediate,
                 # Nullable fields
                 node_id=node_id,
-                session_id=session_id,
                 metadata=metadata,
+                session_id=session_id,
             )
 
             self._services.image_files.save(
-                image_name=image_name,
-                image=image,
-                metadata=metadata,
+                image_name=image_name, image=image, metadata=metadata, graph=graph
             )
 
             image_dto = self.get_dto(image_name)
@@ -268,6 +270,34 @@ class ImageService(ImageServiceABC):
             self._services.logger.error("Problem getting image DTO")
             raise e
 
+    def get_metadata(self, image_name: str) -> Optional[ImageMetadata]:
+        try:
+            image_record = self._services.image_records.get(image_name)
+
+            if not image_record.session_id:
+                return ImageMetadata()
+
+            session_raw = self._services.graph_execution_manager.get_raw(
+                image_record.session_id
+            )
+            graph = None
+
+            if session_raw:
+                try:
+                    graph = get_metadata_graph_from_raw_session(session_raw)
+                except Exception as e:
+                    self._services.logger.warn(f"Failed to parse session graph: {e}")
+                    graph = None
+
+            metadata = self._services.image_records.get_metadata(image_name)
+            return ImageMetadata(graph=graph, metadata=metadata)
+        except ImageRecordNotFoundException:
+            self._services.logger.error("Image record not found")
+            raise
+        except Exception as e:
+            self._services.logger.error("Problem getting image DTO")
+            raise e
+
     def get_path(self, image_name: str, thumbnail: bool = False) -> str:
         try:
             return self._services.image_files.get_path(image_name, thumbnail)
@@ -367,15 +397,3 @@ class ImageService(ImageServiceABC):
         except Exception as e:
             self._services.logger.error("Problem deleting image records and files")
             raise e
-
-    def _get_metadata(
-        self, session_id: Optional[str] = None, node_id: Optional[str] = None
-    ) -> Optional[ImageMetadata]:
-        """Get the metadata for a node."""
-        metadata = None
-
-        if node_id is not None and session_id is not None:
-            session = self._services.graph_execution_manager.get(session_id)
-            metadata = self._services.metadata.create_image_metadata(session, node_id)
-
-        return metadata

invokeai/app/services/invocation_services.py

@@ -10,10 +10,9 @@ if TYPE_CHECKING:
     from invokeai.app.services.model_manager_service import ModelManagerServiceBase
     from invokeai.app.services.events import EventServiceBase
     from invokeai.app.services.latent_storage import LatentsStorageBase
-    from invokeai.app.services.restoration_services import RestorationServices
     from invokeai.app.services.invocation_queue import InvocationQueueABC
     from invokeai.app.services.item_storage import ItemStorageABC
-    from invokeai.app.services.config import InvokeAISettings
+    from invokeai.app.services.config import InvokeAIAppConfig
     from invokeai.app.services.graph import GraphExecutionState, LibraryGraph
     from invokeai.app.services.invoker import InvocationProcessorABC
 
@@ -24,7 +23,7 @@ class InvocationServices:
     # TODO: Just forward-declared everything due to circular dependencies. Fix structure.
     board_images: "BoardImagesServiceABC"
     boards: "BoardServiceABC"
-    configuration: "InvokeAISettings"
+    configuration: "InvokeAIAppConfig"
     events: "EventServiceBase"
     graph_execution_manager: "ItemStorageABC"["GraphExecutionState"]
     graph_library: "ItemStorageABC"["LibraryGraph"]
@@ -34,13 +33,12 @@ class InvocationServices:
     model_manager: "ModelManagerServiceBase"
     processor: "InvocationProcessorABC"
     queue: "InvocationQueueABC"
-    restoration: "RestorationServices"
 
     def __init__(
         self,
         board_images: "BoardImagesServiceABC",
         boards: "BoardServiceABC",
-        configuration: "InvokeAISettings",
+        configuration: "InvokeAIAppConfig",
         events: "EventServiceBase",
         graph_execution_manager: "ItemStorageABC"["GraphExecutionState"],
         graph_library: "ItemStorageABC"["LibraryGraph"],
@@ -50,7 +48,6 @@ class InvocationServices:
         model_manager: "ModelManagerServiceBase",
         processor: "InvocationProcessorABC",
         queue: "InvocationQueueABC",
-        restoration: "RestorationServices",
     ):
         self.board_images = board_images
         self.boards = boards
@@ -65,4 +62,3 @@ class InvocationServices:
         self.model_manager = model_manager
         self.processor = processor
         self.queue = queue
-        self.restoration = restoration

invokeai/app/services/item_storage.py

@@ -1,5 +1,5 @@
 from abc import ABC, abstractmethod
-from typing import Callable, Generic, TypeVar
+from typing import Callable, Generic, Optional, TypeVar
 
 from pydantic import BaseModel, Field
 from pydantic.generics import GenericModel
@@ -29,14 +29,22 @@ class ItemStorageABC(ABC, Generic[T]):
 
     @abstractmethod
     def get(self, item_id: str) -> T:
+        """Gets the item, parsing it into a Pydantic model"""
+        pass
+
+    @abstractmethod
+    def get_raw(self, item_id: str) -> Optional[str]:
+        """Gets the raw item as a string, skipping Pydantic parsing"""
         pass
 
     @abstractmethod
     def set(self, item: T) -> None:
+        """Sets the item"""
         pass
 
     @abstractmethod
     def list(self, page: int = 0, per_page: int = 10) -> PaginatedResults[T]:
+        """Gets a paginated list of items"""
         pass
 
     @abstractmethod

invokeai/app/services/metadata.py

@@ -1,142 +0,0 @@
-from abc import ABC, abstractmethod
-from typing import Any, Optional
-import networkx as nx
-
-from invokeai.app.models.metadata import ImageMetadata
-from invokeai.app.services.graph import Graph, GraphExecutionState
-
-
-class MetadataServiceBase(ABC):
-    """Handles building metadata for nodes, images, and outputs."""
-
-    @abstractmethod
-    def create_image_metadata(
-        self, session: GraphExecutionState, node_id: str
-    ) -> ImageMetadata:
-        """Builds an ImageMetadata object for a node."""
-        pass
-
-
-class CoreMetadataService(MetadataServiceBase):
-    _ANCESTOR_TYPES = ["t2l", "l2l"]
-    """The ancestor types that contain the core metadata"""
-
-    _ANCESTOR_PARAMS = ["type", "steps", "model", "cfg_scale", "scheduler", "strength"]
-    """The core metadata parameters in the ancestor types"""
-
-    _NOISE_FIELDS = ["seed", "width", "height"]
-    """The core metadata parameters in the noise node"""
-
-    def create_image_metadata(
-        self, session: GraphExecutionState, node_id: str
-    ) -> ImageMetadata:
-        metadata = self._build_metadata_from_graph(session, node_id)
-
-        return metadata
-
-    def _find_nearest_ancestor(self, G: nx.DiGraph, node_id: str) -> Optional[str]:
-        """
-        Finds the id of the nearest ancestor (of a valid type) of a given node.
-
-        Parameters:
-        G (nx.DiGraph): The execution graph, converted in to a networkx DiGraph. Its nodes must
-        have the same data as the execution graph.
-        node_id (str): The ID of the node.
-
-        Returns:
-        str | None: The ID of the nearest ancestor, or None if there are no valid ancestors.
-        """
-
-        # Retrieve the node from the graph
-        node = G.nodes[node_id]
-
-        # If the node type is one of the core metadata node types, return its id
-        if node.get("type") in self._ANCESTOR_TYPES:
-            return node.get("id")
-
-        # Else, look for the ancestor in the predecessor nodes
-        for predecessor in G.predecessors(node_id):
-            result = self._find_nearest_ancestor(G, predecessor)
-            if result:
-                return result
-
-        # If there are no valid ancestors, return None
-        return None
-
-    def _get_additional_metadata(
-        self, graph: Graph, node_id: str
-    ) -> Optional[dict[str, Any]]:
-        """
-        Returns additional metadata for a given node.
-
-        Parameters:
-        graph (Graph): The execution graph.
-        node_id (str): The ID of the node.
-
-        Returns:
-        dict[str, Any] | None: A dictionary of additional metadata.
-        """
-
-        metadata = {}
-
-        # Iterate over all edges in the graph
-        for edge in graph.edges:
-            dest_node_id = edge.destination.node_id
-            dest_field = edge.destination.field
-            source_node_dict = graph.nodes[edge.source.node_id].dict()
-
-            # If the destination node ID matches the given node ID, gather necessary metadata
-            if dest_node_id == node_id:
-                # Prompt
-                if dest_field == "positive_conditioning":
-                    metadata["positive_conditioning"] = source_node_dict.get("prompt")
-                # Negative prompt
-                if dest_field == "negative_conditioning":
-                    metadata["negative_conditioning"] = source_node_dict.get("prompt")
-                # Seed, width and height
-                if dest_field == "noise":
-                    for field in self._NOISE_FIELDS:
-                        metadata[field] = source_node_dict.get(field)
-        return metadata
-
-    def _build_metadata_from_graph(
-        self, session: GraphExecutionState, node_id: str
-    ) -> ImageMetadata:
-        """
-        Builds an ImageMetadata object for a node.
-
-        Parameters:
-        session (GraphExecutionState): The session.
-        node_id (str): The ID of the node.
-
-        Returns:
-        ImageMetadata: The metadata for the node.
-        """
-
-        # We need to do all the traversal on the execution graph
-        graph = session.execution_graph
-
-        # Find the nearest `t2l`/`l2l` ancestor of the given node
-        ancestor_id = self._find_nearest_ancestor(graph.nx_graph_with_data(), node_id)
-
-        # If no ancestor was found, return an empty ImageMetadata object
-        if ancestor_id is None:
-            return ImageMetadata()
-
-        ancestor_node = graph.get_node(ancestor_id)
-
-        # Grab all the core metadata from the ancestor node
-        ancestor_metadata = {
-            param: val
-            for param, val in ancestor_node.dict().items()
-            if param in self._ANCESTOR_PARAMS
-        }
-
-        # Get this image's prompts and noise parameters
-        addl_metadata = self._get_additional_metadata(graph, ancestor_id)
-
-        # If additional metadata was found, add it to the main metadata
-        if addl_metadata is not None:
-            ancestor_metadata.update(addl_metadata)
-
-        return ImageMetadata(**ancestor_metadata)

invokeai/app/services/model_manager_service.py

@@ -18,8 +18,9 @@ from invokeai.backend.model_management import (
     SchedulerPredictionType,
     ModelMerger,
     MergeInterpolationMethod,
+    ModelNotFoundException,
 )
-
+from invokeai.backend.model_management.model_search import FindModels
 
 import torch
 from invokeai.app.models.exceptions import CanceledException
@@ -145,7 +146,7 @@ class ModelManagerServiceBase(ABC):
     ) -> AddModelResult:
         """
         Update the named model with a dictionary of attributes. Will fail with a
-        KeyErrorException if the name does not already exist.
+        ModelNotFoundException if the name does not already exist.
 
         On a successful update, the config will be changed in memory. Will fail 
         with an assertion error if provided attributes are incorrect or 
@@ -167,6 +168,27 @@ class ModelManagerServiceBase(ABC):
         """
         pass
 
+    @abstractmethod
+    def rename_model(self,
+                     model_name: str,
+                     base_model: BaseModelType,
+                     model_type: ModelType,
+                     new_name: str,
+                     ):
+        """
+        Rename the indicated model.
+        """
+        pass
+
+    @abstractmethod
+    def list_checkpoint_configs(
+        self
+    )->List[Path]:
+        """
+        List the checkpoint config paths from ROOT/configs/stable-diffusion.
+        """
+        pass
+
     @abstractmethod
     def convert_model(
         self,
@@ -220,6 +242,7 @@ class ModelManagerServiceBase(ABC):
             alpha: Optional[float] = 0.5,
             interp: Optional[MergeInterpolationMethod] = None,
             force: Optional[bool] = False,
+            merge_dest_directory: Optional[Path] = None
     ) -> AddModelResult:
         """
         Merge two to three diffusrs pipeline models and save as a new model.
@@ -228,9 +251,26 @@ class ModelManagerServiceBase(ABC):
         :param merged_model_name: Name of destination merged model
         :param alpha: Alpha strength to apply to 2d and 3d model
         :param interp: Interpolation method. None (default) 
+        :param merge_dest_directory: Save the merged model to the designated directory (with 'merged_model_name' appended)
         """
         pass
-    
+
+    @abstractmethod
+    def search_for_models(self, directory: Path)->List[Path]:
+        """
+        Return list of all models found in the designated directory.
+        """
+        pass
+        
+    @abstractmethod
+    def sync_to_config(self):
+        """
+        Re-read models.yaml, rescan the models directory, and reimport models 
+        in the autoimport directories. Call after making changes outside the
+        model manager API.
+        """
+        pass
+        
     @abstractmethod
     def commit(self, conf_file: Optional[Path] = None) -> None:
         """
@@ -258,9 +298,7 @@ class ModelManagerService(ModelManagerServiceBase):
             config_file = config.model_conf_path
         else:
             config_file = config.root_dir / "configs/models.yaml"
-        if not config_file.exists():
-            raise IOError(f"The file {config_file} could not be found.")
-
+            
         logger.debug(f'config file={config_file}')
 
         device = torch.device(choose_torch_device())
@@ -301,7 +339,6 @@ class ModelManagerService(ModelManagerServiceBase):
         base_model: BaseModelType,
         model_type: ModelType,
         submodel: Optional[SubModelType] = None,
-        node: Optional[BaseInvocation] = None,
         context: Optional[InvocationContext] = None,
     ) -> ModelInfo:
         """
@@ -309,11 +346,9 @@ class ModelManagerService(ModelManagerServiceBase):
         part (such as the vae) of a diffusers mode.
         """
 
-        # if we are called from within a node, then we get to emit
-        # load start and complete events
-        if node and context:
+        # we can emit model loading events if we are executing with access to the invocation context
+        if context:
             self._emit_load_event(
-                node=node,
                 context=context,
                 model_name=model_name,
                 base_model=base_model,
@@ -328,9 +363,8 @@ class ModelManagerService(ModelManagerServiceBase):
             submodel,
         )
 
-        if node and context:
+        if context:
             self._emit_load_event(
-                node=node,
                 context=context,
                 model_name=model_name,
                 base_model=base_model,
@@ -414,14 +448,14 @@ class ModelManagerService(ModelManagerServiceBase):
     ) -> AddModelResult:
         """
         Update the named model with a dictionary of attributes. Will fail with a
-        KeyError exception if the name does not already exist.
+        ModelNotFoundException exception if the name does not already exist.
         On a successful update, the config will be changed in memory. Will fail 
         with an assertion error if provided attributes are incorrect or 
         the model name is missing. Call commit() to write changes to disk.
         """
         self.logger.debug(f'update model {model_name}')
         if not self.model_exists(model_name, base_model, model_type):
-            raise KeyError(f"Unknown model {model_name}")
+            raise ModelNotFoundException(f"Unknown model {model_name}")
         return self.add_model(model_name, base_model, model_type, model_attributes, clobber=True)
     
     def del_model(
@@ -433,16 +467,18 @@ class ModelManagerService(ModelManagerServiceBase):
         """
         Delete the named model from configuration. If delete_files is true,
         then the underlying weight file or diffusers directory will be deleted
-        as well. Call commit() to write to disk.
+        as well.
         """
         self.logger.debug(f'delete model {model_name}')
         self.mgr.del_model(model_name, base_model, model_type)
+        self.mgr.commit()
 
     def convert_model(
         self,
         model_name: str,
         base_model: BaseModelType,
         model_type: Union[ModelType.Main,ModelType.Vae],
+        convert_dest_directory: Optional[Path] = Field(default=None, description="Optional directory location for merged model"),        
     ) -> AddModelResult:
         """
         Convert a checkpoint file into a diffusers folder, deleting the cached
@@ -451,13 +487,14 @@ class ModelManagerService(ModelManagerServiceBase):
         :param model_name: Name of the model to convert
         :param base_model: Base model type
         :param model_type: Type of model ['vae' or 'main']
+        :param convert_dest_directory: Save the converted model to the designated directory (`models/etc/etc` by default)
 
         This will raise a ValueError unless the model is not a checkpoint. It will
         also raise a ValueError in the event that there is a similarly-named diffusers
         directory already in place.
         """
         self.logger.debug(f'convert model {model_name}')        
-        return self.mgr.convert_model(model_name, base_model, model_type)
+        return self.mgr.convert_model(model_name, base_model, model_type, convert_dest_directory)
 
     def commit(self, conf_file: Optional[Path]=None):
         """
@@ -469,23 +506,19 @@ class ModelManagerService(ModelManagerServiceBase):
 
     def _emit_load_event(
         self,
-        node,
         context,
         model_name: str,
         base_model: BaseModelType,
         model_type: ModelType,
-        submodel: SubModelType,
+        submodel: Optional[SubModelType] = None,
         model_info: Optional[ModelInfo] = None,
     ):
         if context.services.queue.is_canceled(context.graph_execution_state_id):
             raise CanceledException()
-        graph_execution_state = context.services.graph_execution_manager.get(context.graph_execution_state_id)
-        source_node_id = graph_execution_state.prepared_source_mapping[node.id]
+
         if model_info:
             context.services.events.emit_model_load_completed(
                 graph_execution_state_id=context.graph_execution_state_id,
-                node=node.dict(),
-                source_node_id=source_node_id,
                 model_name=model_name,
                 base_model=base_model,
                 model_type=model_type,
@@ -495,8 +528,6 @@ class ModelManagerService(ModelManagerServiceBase):
         else:
             context.services.events.emit_model_load_started(
                 graph_execution_state_id=context.graph_execution_state_id,
-                node=node.dict(),
-                source_node_id=source_node_id,
                 model_name=model_name,
                 base_model=base_model,
                 model_type=model_type,
@@ -538,6 +569,7 @@ class ModelManagerService(ModelManagerServiceBase):
             alpha: Optional[float] = 0.5,
             interp: Optional[MergeInterpolationMethod] = None,
             force: Optional[bool] = False,
+            merge_dest_directory: Optional[Path] = Field(default=None, description="Optional directory location for merged model"),
     ) -> AddModelResult:
         """
         Merge two to three diffusrs pipeline models and save as a new model.
@@ -546,6 +578,7 @@ class ModelManagerService(ModelManagerServiceBase):
         :param merged_model_name: Name of destination merged model
         :param alpha: Alpha strength to apply to 2d and 3d model
         :param interp: Interpolation method. None (default) 
+        :param merge_dest_directory: Save the merged model to the designated directory (with 'merged_model_name' appended)
         """
         merger = ModelMerger(self.mgr)
         try:
@@ -556,7 +589,55 @@ class ModelManagerService(ModelManagerServiceBase):
                 alpha = alpha,
                 interp = interp,
                 force = force,
+                merge_dest_directory=merge_dest_directory,
             )
         except AssertionError as e:
             raise ValueError(e)
         return result
+
+    def search_for_models(self, directory: Path)->List[Path]:
+        """
+        Return list of all models found in the designated directory.
+        """
+        search = FindModels(directory,self.logger)
+        return search.list_models()
+
+    def sync_to_config(self):
+        """
+        Re-read models.yaml, rescan the models directory, and reimport models 
+        in the autoimport directories. Call after making changes outside the
+        model manager API.
+        """
+        return self.mgr.sync_to_config()
+
+    def list_checkpoint_configs(self)->List[Path]:
+        """
+        List the checkpoint config paths from ROOT/configs/stable-diffusion.
+        """
+        config = self.mgr.app_config
+        conf_path = config.legacy_conf_path
+        root_path = config.root_path
+        return [(conf_path / x).relative_to(root_path) for x in conf_path.glob('**/*.yaml')]
+
+    def rename_model(self,
+                     model_name: str,
+                     base_model: BaseModelType,
+                     model_type: ModelType,
+                     new_name: str = None,
+                     new_base: BaseModelType = None,
+                     ):
+        """
+        Rename the indicated model. Can provide a new name and/or a new base.
+        :param model_name: Current name of the model
+        :param base_model: Current base of the model
+        :param model_type: Model type (can't be changed)
+        :param new_name: New name for the model
+        :param new_base: New base for the model
+        """
+        self.mgr.rename_model(base_model = base_model,
+                              model_type = model_type,
+                              model_name = model_name,
+                              new_name = new_name,
+                              new_base = new_base,
+                              )
+        

invokeai/app/services/models/image_record.py

@@ -1,13 +1,14 @@
 import datetime
 from typing import Optional, Union
+
 from pydantic import BaseModel, Extra, Field, StrictBool, StrictStr
+
 from invokeai.app.models.image import ImageCategory, ResourceOrigin
-from invokeai.app.models.metadata import ImageMetadata
 from invokeai.app.util.misc import get_iso_timestamp
 
 
 class ImageRecord(BaseModel):
-    """Deserialized image record."""
+    """Deserialized image record without metadata."""
 
     image_name: str = Field(description="The unique name of the image.")
     """The unique name of the image."""
@@ -43,11 +44,6 @@ class ImageRecord(BaseModel):
         description="The node ID that generated this image, if it is a generated image.",
     )
     """The node ID that generated this image, if it is a generated image."""
-    metadata: Optional[ImageMetadata] = Field(
-        default=None,
-        description="A limited subset of the image's generation metadata. Retrieve the image's session for full metadata.",
-    )
-    """A limited subset of the image's generation metadata. Retrieve the image's session for full metadata."""
 
 
 class ImageRecordChanges(BaseModel, extra=Extra.forbid):
@@ -112,6 +108,7 @@ def deserialize_image_record(image_dict: dict) -> ImageRecord:
 
     # Retrieve all the values, setting "reasonable" defaults if they are not present.
 
+    # TODO: do we really need to handle default values here? ideally the data is the correct shape...
     image_name = image_dict.get("image_name", "unknown")
     image_origin = ResourceOrigin(
         image_dict.get("image_origin", ResourceOrigin.INTERNAL.value)
@@ -128,13 +125,6 @@ def deserialize_image_record(image_dict: dict) -> ImageRecord:
     deleted_at = image_dict.get("deleted_at", get_iso_timestamp())
     is_intermediate = image_dict.get("is_intermediate", False)
 
-    raw_metadata = image_dict.get("metadata")
-
-    if raw_metadata is not None:
-        metadata = ImageMetadata.parse_raw(raw_metadata)
-    else:
-        metadata = None
-
     return ImageRecord(
         image_name=image_name,
         image_origin=image_origin,
@@ -143,7 +133,6 @@ def deserialize_image_record(image_dict: dict) -> ImageRecord:
         height=height,
         session_id=session_id,
         node_id=node_id,
-        metadata=metadata,
         created_at=created_at,
         updated_at=updated_at,
         deleted_at=deleted_at,

invokeai/app/services/processor.py

@@ -104,6 +104,7 @@ class DefaultInvocationProcessor(InvocationProcessorABC):
 
                 except Exception as e:
                     error = traceback.format_exc()
+                    logger.error(error)
 
                     # Save error
                     graph_execution_state.set_node_error(invocation.id, error)

invokeai/app/services/restoration_services.py

@@ -1,113 +0,0 @@
-import sys
-import traceback
-import torch
-from typing import types
-from ...backend.restoration import Restoration
-from ...backend.util import choose_torch_device, CPU_DEVICE, MPS_DEVICE
-
-# This should be a real base class for postprocessing functions,
-# but right now we just instantiate the existing gfpgan, esrgan
-# and codeformer functions.
-class RestorationServices:
-    '''Face restoration and upscaling'''
-    
-    def __init__(self,args,logger:types.ModuleType):
-        try:
-            gfpgan, codeformer, esrgan = None, None, None
-            if args.restore or args.esrgan:
-                restoration = Restoration()
-                # TODO: redo for new model structure
-                if False and args.restore:
-                    gfpgan, codeformer = restoration.load_face_restore_models(
-                        args.gfpgan_model_path
-                    )
-                else:
-                    logger.info("Face restoration disabled")
-                    if False and args.esrgan:
-                        esrgan = restoration.load_esrgan(args.esrgan_bg_tile)
-                    else:
-                        logger.info("Upscaling disabled")
-            else:
-                logger.info("Face restoration and upscaling disabled")
-        except (ModuleNotFoundError, ImportError):
-            print(traceback.format_exc(), file=sys.stderr)
-            logger.info("You may need to install the ESRGAN and/or GFPGAN modules")
-        self.device = torch.device(choose_torch_device())
-        self.gfpgan = gfpgan
-        self.codeformer = codeformer
-        self.esrgan = esrgan
-        self.logger = logger
-        self.logger.info('Face restoration initialized')
-
-    # note that this one method does gfpgan and codepath reconstruction, as well as
-    # esrgan upscaling
-    # TO DO: refactor into separate methods
-    def upscale_and_reconstruct(
-        self,
-        image_list,
-        facetool="gfpgan",
-        upscale=None,
-        upscale_denoise_str=0.75,
-        strength=0.0,
-        codeformer_fidelity=0.75,
-        save_original=False,
-        image_callback=None,
-        prefix=None,
-    ):
-        results = []
-        for r in image_list:
-            image, seed = r
-            try:
-                if strength > 0:
-                    if self.gfpgan is not None or self.codeformer is not None:
-                        if facetool == "gfpgan":
-                            if self.gfpgan is None:
-                                self.logger.info(
-                                    "GFPGAN not found. Face restoration is disabled."
-                                )
-                            else:
-                                image = self.gfpgan.process(image, strength, seed)
-                        if facetool == "codeformer":
-                            if self.codeformer is None:
-                                self.logger.info(
-                                    "CodeFormer not found. Face restoration is disabled."
-                                )
-                            else:
-                                cf_device = (
-                                    CPU_DEVICE if self.device == MPS_DEVICE else self.device
-                                )
-                                image = self.codeformer.process(
-                                    image=image,
-                                    strength=strength,
-                                    device=cf_device,
-                                    seed=seed,
-                                    fidelity=codeformer_fidelity,
-                                )
-                    else:
-                        self.logger.info("Face Restoration is disabled.")
-                if upscale is not None:
-                    if self.esrgan is not None:
-                        if len(upscale) < 2:
-                            upscale.append(0.75)
-                        image = self.esrgan.process(
-                            image,
-                            upscale[1],
-                            seed,
-                            int(upscale[0]),
-                            denoise_str=upscale_denoise_str,
-                        )
-                    else:
-                        self.logger.info("ESRGAN is disabled. Image not upscaled.")
-            except Exception as e:
-                self.logger.info(
-                    f"Error running RealESRGAN or GFPGAN. Your image was not upscaled.\n{e}"
-                )
-
-            if image_callback is not None:
-                image_callback(image, seed, upscaled=True, use_prefix=prefix)
-            else:
-                r[0] = image
-
-            results.append([image, seed])
-
-        return results

invokeai/app/services/sqlite.py

@@ -1,6 +1,6 @@
 import sqlite3
 from threading import Lock
-from typing import Generic, TypeVar, Optional, Union, get_args
+from typing import Generic, Optional, TypeVar, get_args
 
 from pydantic import BaseModel, parse_raw_as
 
@@ -78,6 +78,21 @@ class SqliteItemStorage(ItemStorageABC, Generic[T]):
 
         return self._parse_item(result[0])
 
+    def get_raw(self, id: str) -> Optional[str]:
+        try:
+            self._lock.acquire()
+            self._cursor.execute(
+                f"""SELECT item FROM {self._table_name} WHERE id = ?;""", (str(id),)
+            )
+            result = self._cursor.fetchone()
+        finally:
+            self._lock.release()
+
+        if not result:
+            return None
+
+        return result[0]
+
     def delete(self, id: str):
         try:
             self._lock.acquire()

invokeai/app/services/urls.py

@@ -22,4 +22,4 @@ class LocalUrlService(UrlServiceBase):
         if thumbnail:
             return f"{self._base_url}/images/{image_basename}/thumbnail"
 
-        return f"{self._base_url}/images/{image_basename}"
+        return f"{self._base_url}/images/{image_basename}/full"

invokeai/app/util/metadata.py

@@ -0,0 +1,55 @@
+import json
+from typing import Optional
+
+from pydantic import ValidationError
+
+from invokeai.app.services.graph import Edge
+
+
+def get_metadata_graph_from_raw_session(session_raw: str) -> Optional[dict]:
+    """
+    Parses raw session string, returning a dict of the graph.
+
+    Only the general graph shape is validated; none of the fields are validated.
+
+    Any `metadata_accumulator` nodes and edges are removed.
+
+    Any validation failure will return None.
+    """
+
+    graph = json.loads(session_raw).get("graph", None)
+
+    # sanity check make sure the graph is at least reasonably shaped
+    if (
+        type(graph) is not dict
+        or "nodes" not in graph
+        or type(graph["nodes"]) is not dict
+        or "edges" not in graph
+        or type(graph["edges"]) is not list
+    ):
+        # something has gone terribly awry, return an empty dict
+        return None
+
+    try:
+        # delete the `metadata_accumulator` node
+        del graph["nodes"]["metadata_accumulator"]
+    except KeyError:
+        # no accumulator node, all good
+        pass
+
+    # delete any edges to or from it
+    for i, edge in enumerate(graph["edges"]):
+        try:
+            # try to parse the edge
+            Edge(**edge)
+        except ValidationError:
+            # something has gone terribly awry, return an empty dict
+            return None
+
+        if (
+            edge["source"]["node_id"] == "metadata_accumulator"
+            or edge["destination"]["node_id"] == "metadata_accumulator"
+        ):
+            del graph["edges"][i]
+
+    return graph

invokeai/backend/generator/base.py

@@ -466,7 +466,6 @@ class Generator:
             dtype=samples.dtype,
             device=samples.device,
         )
-
         latent_image = samples[0].permute(1, 2, 0) @ v1_5_latent_rgb_factors
         latents_ubyte = (
             ((latent_image + 1) / 2)

invokeai/backend/install/invokeai_configure.py

@@ -30,8 +30,6 @@ from huggingface_hub import login as hf_hub_login
 from omegaconf import OmegaConf
 from tqdm import tqdm
 from transformers import (
-    AutoProcessor,
-    CLIPSegForImageSegmentation,
     CLIPTextModel,
     CLIPTokenizer,
     AutoFeatureExtractor,
@@ -45,7 +43,6 @@ from invokeai.app.services.config import (
 from invokeai.backend.util.logging import InvokeAILogger
 from invokeai.frontend.install.model_install import addModelsForm, process_and_execute
 from invokeai.frontend.install.widgets import (
-    SingleSelectColumns,
     CenteredButtonPress,
     IntTitleSlider,
     set_min_terminal_size,
@@ -72,7 +69,6 @@ transformers.logging.set_verbosity_error()
 config = InvokeAIAppConfig.get_config()
 
 Model_dir = "models"
-Weights_dir = "ldm/stable-diffusion-v1/"
 
 Default_config_file = config.model_conf_path
 SD_Configs = config.legacy_conf_path
@@ -226,64 +222,35 @@ def download_conversion_models():
 
 # ---------------------------------------------
 def download_realesrgan():
-    logger.info("Installing models from RealESRGAN...")
-    model_url = "https://github.com/xinntao/Real-ESRGAN/releases/download/v0.2.5.0/realesr-general-x4v3.pth"
-    wdn_model_url = "https://github.com/xinntao/Real-ESRGAN/releases/download/v0.2.5.0/realesr-general-wdn-x4v3.pth"
-
-    model_dest = config.root_path / "models/core/upscaling/realesrgan/realesr-general-x4v3.pth"
-    wdn_model_dest = config.root_path / "models/core/upscaling/realesrgan/realesr-general-wdn-x4v3.pth"
-
-    download_with_progress_bar(model_url, str(model_dest), "RealESRGAN")
-    download_with_progress_bar(wdn_model_url, str(wdn_model_dest), "RealESRGANwdn")
-
-
-def download_gfpgan():
-    logger.info("Installing GFPGAN models...")
-    for model in (
-        [
-            "https://github.com/TencentARC/GFPGAN/releases/download/v1.3.0/GFPGANv1.4.pth",
-            "./models/core/face_restoration/gfpgan/GFPGANv1.4.pth",
-        ],
-        [
-            "https://github.com/xinntao/facexlib/releases/download/v0.1.0/detection_Resnet50_Final.pth",
-            "./models/core/face_restoration/gfpgan/weights/detection_Resnet50_Final.pth",
-        ],
-        [
-            "https://github.com/xinntao/facexlib/releases/download/v0.2.2/parsing_parsenet.pth",
-            "./models/core/face_restoration/gfpgan/weights/parsing_parsenet.pth",
-        ],
-    ):
-        model_url, model_dest = model[0], config.root_path / model[1]
-        download_with_progress_bar(model_url, str(model_dest), "GFPGAN weights")
-
+    logger.info("Installing ESRGAN Upscaling models...")
+    URLs = [
+        dict(
+            url = "https://github.com/xinntao/Real-ESRGAN/releases/download/v0.1.0/RealESRGAN_x4plus.pth",
+            dest = "core/upscaling/realesrgan/RealESRGAN_x4plus.pth",
+            description = "RealESRGAN_x4plus.pth",
+        ),
+        dict(
+            url = "https://github.com/xinntao/Real-ESRGAN/releases/download/v0.2.2.4/RealESRGAN_x4plus_anime_6B.pth",
+            dest = "core/upscaling/realesrgan/RealESRGAN_x4plus_anime_6B.pth",
+            description = "RealESRGAN_x4plus_anime_6B.pth",
+        ),
+        dict(
+            url= "https://github.com/xinntao/Real-ESRGAN/releases/download/v0.1.1/ESRGAN_SRx4_DF2KOST_official-ff704c30.pth",
+            dest= "core/upscaling/realesrgan/ESRGAN_SRx4_DF2KOST_official-ff704c30.pth",
+            description = "ESRGAN_SRx4_DF2KOST_official.pth",
+        ),
+        dict(
+            url= "https://github.com/xinntao/Real-ESRGAN/releases/download/v0.2.1/RealESRGAN_x2plus.pth",
+            dest= "core/upscaling/realesrgan/RealESRGAN_x2plus.pth",
+            description = "RealESRGAN_x2plus.pth",
+        ),
+    ]
+    for model in URLs:
+        download_with_progress_bar(model['url'], config.models_path / model['dest'], model['description'])
 
 # ---------------------------------------------
-def download_codeformer():
-    logger.info("Installing CodeFormer model file...")
-    model_url = (
-        "https://github.com/sczhou/CodeFormer/releases/download/v0.1.0/codeformer.pth"
-    )
-    model_dest = config.root_path / "models/core/face_restoration/codeformer/codeformer.pth"
-    download_with_progress_bar(model_url, str(model_dest), "CodeFormer")
-
-
-# ---------------------------------------------
-def download_clipseg():
-    logger.info("Installing clipseg model for text-based masking...")
-    CLIPSEG_MODEL = "CIDAS/clipseg-rd64-refined"
-    try:
-        hf_download_from_pretrained(AutoProcessor, CLIPSEG_MODEL, config.root_path / 'models/core/misc/clipseg')
-        hf_download_from_pretrained(CLIPSegForImageSegmentation, CLIPSEG_MODEL, config.root_path / 'models/core/misc/clipseg')
-    except Exception:
-        logger.info("Error installing clipseg model:")
-        logger.info(traceback.format_exc())
-
-
 def download_support_models():
     download_realesrgan()
-    download_gfpgan()
-    download_codeformer()
-    download_clipseg()
     download_conversion_models()
 
 # -------------------------------------
@@ -666,7 +633,7 @@ def run_console_ui(
 
     # The third argument is needed in the Windows 11 environment to
     # launch a console window running this program.
-    set_min_terminal_size(MIN_COLS, MIN_LINES,'invokeai-configure')
+    set_min_terminal_size(MIN_COLS, MIN_LINES)
 
     # the install-models application spawns a subprocess to install
     # models, and will crash unless this is set before running.
@@ -743,7 +710,7 @@ def migrate_if_needed(opt: Namespace, root: Path)->bool:
     old_init_file = root / 'invokeai.init'
     new_init_file = root / 'invokeai.yaml'
     old_hub = root / 'models/hub'
-    migration_needed =  old_init_file.exists() and not new_init_file.exists() or old_hub.exists()
+    migration_needed =  (old_init_file.exists() and not new_init_file.exists()) and old_hub.exists()
     
     if migration_needed:
         if opt.yes_to_all or \
@@ -858,9 +825,9 @@ def main():
             download_support_models()
 
         if opt.skip_sd_weights:
-            logger.info("\n** SKIPPING DIFFUSION WEIGHTS DOWNLOAD PER USER REQUEST **")
+            logger.warning("SKIPPING DIFFUSION WEIGHTS DOWNLOAD PER USER REQUEST")
         elif models_to_download:
-            logger.info("\n** DOWNLOADING DIFFUSION WEIGHTS **")
+            logger.info("DOWNLOADING DIFFUSION WEIGHTS")
             process_and_execute(opt, models_to_download)
 
         postscript(errors=errors)

invokeai/backend/install/migrate_to_3.py

@@ -593,9 +593,12 @@ script, which will perform a full upgrade in place."""
     config = InvokeAIAppConfig.get_config()
     config.parse_args(['--root',str(dest_root)])
 
-    # TODO: revisit
-    # assert (dest_root / 'models').is_dir(), f"{dest_root} does not contain a 'models' subdirectory"
-    # assert (dest_root / 'invokeai.yaml').exists(), f"{dest_root} does not contain an InvokeAI init file."
+    # TODO: revisit - don't rely on invokeai.yaml to exist yet!
+    dest_is_setup = (dest_root / 'models/core').exists() and (dest_root / 'databases').exists()
+    if not dest_is_setup:
+        import invokeai.frontend.install.invokeai_configure
+        from invokeai.backend.install.invokeai_configure import initialize_rootdir
+        initialize_rootdir(dest_root, True)
 
     do_migrate(src_root,dest_root)
 

invokeai/backend/install/model_install_backend.py

@@ -10,7 +10,7 @@ from tempfile import TemporaryDirectory
 from typing import List, Dict, Callable, Union, Set
 
 import requests
-from diffusers import StableDiffusionPipeline
+from diffusers import DiffusionPipeline
 from diffusers import logging as dlogging
 from huggingface_hub import hf_hub_url, HfFolder, HfApi
 from omegaconf import OmegaConf
@@ -71,8 +71,6 @@ class ModelInstallList:
 class InstallSelections():
     install_models: List[str]= field(default_factory=list)
     remove_models: List[str]=field(default_factory=list)
-#    scan_directory: Path = None
-#    autoscan_on_startup: bool=False
 
 @dataclass
 class ModelLoadInfo():
@@ -119,10 +117,11 @@ class ModelInstall(object):
 
         # supplement with entries in models.yaml
         installed_models = self.mgr.list_models()
+        
         for md in installed_models:
             base = md['base_model']
-            model_type = md['type']
-            name = md['name']
+            model_type = md['model_type']
+            name = md['model_name']
             key = ModelManager.create_key(name, base, model_type)
             if key in model_dict:
                 model_dict[key].installed = True
@@ -136,6 +135,12 @@ class ModelInstall(object):
                 )
         return {x : model_dict[x] for x in sorted(model_dict.keys(),key=lambda y: model_dict[y].name.lower())}
 
+    def list_models(self, model_type):
+        installed = self.mgr.list_models(model_type=model_type)
+        print(f'Installed models of type `{model_type}`:')
+        for i in installed:
+            print(f"{i['model_name']}\t{i['base_model']}\t{i['path']}")
+
     def starter_models(self)->Set[str]:
         models = set()
         for key, value in self.datasets.items():
@@ -207,7 +212,7 @@ class ModelInstall(object):
                                     {'config.json','model_index.json','learned_embeds.bin','pytorch_lora_weights.bin'}
                                     ]
                                    ):
-            models_installed.update(self._install_path(path))
+            models_installed.update({str(model_path_id_or_url): self._install_path(path)})
 
         # recursive scan
         elif path.is_dir():
@@ -305,6 +310,8 @@ class ModelInstall(object):
         if key := self.reverse_paths.get(path_name):
             (name, base, mtype) = ModelManager.parse_key(key)
             return name
+        elif location.is_dir():
+            return location.name
         else:
             return location.stem
 
@@ -360,7 +367,7 @@ class ModelInstall(object):
         model = None
         for revision in revisions:
             try:
-                model = StableDiffusionPipeline.from_pretrained(repo_id,revision=revision,safety_checker=None)
+                model = DiffusionPipeline.from_pretrained(repo_id,revision=revision,safety_checker=None)
             except:  # most errors are due to fp16 not being present. Fix this to catch other errors
                 pass
             if model:

invokeai/backend/model_management/__init__.py

@@ -3,6 +3,7 @@ Initialization file for invokeai.backend.model_management
 """
 from .model_manager import ModelManager, ModelInfo, AddModelResult, SchedulerPredictionType
 from .model_cache import ModelCache
-from .models import BaseModelType, ModelType, SubModelType, ModelVariantType
+from .lora import ModelPatcher, ONNXModelPatcher
+from .models import BaseModelType, ModelType, SubModelType, ModelVariantType, ModelNotFoundException
 from .model_merge import ModelMerger, MergeInterpolationMethod
 

invokeai/backend/model_management/lora.py

@@ -6,11 +6,22 @@ from typing import Optional, Dict, Tuple, Any, Union, List
 from pathlib import Path
 
 import torch
+from safetensors.torch import load_file
+from torch.utils.hooks import RemovableHandle
+
+from diffusers.models import UNet2DConditionModel
+from transformers import CLIPTextModel
+from onnx import numpy_helper
+from onnxruntime import OrtValue
+import numpy as np
+
 from compel.embeddings_provider import BaseTextualInversionManager
 from diffusers.models import UNet2DConditionModel
 from safetensors.torch import load_file
 from transformers import CLIPTextModel, CLIPTokenizer
 
+# TODO: rename and split this file
+
 class LoRALayerBase:
     #rank: Optional[int]
     #alpha: Optional[float]
@@ -708,3 +719,185 @@ class TextualInversionManager(BaseTextualInversionManager):
 
         return new_token_ids
 
+
+class ONNXModelPatcher:
+
+    @classmethod
+    @contextmanager
+    def apply_lora_unet(
+        cls,
+        unet: OnnxRuntimeModel,
+        loras: List[Tuple[LoRAModel, float]],
+    ):
+        with cls.apply_lora(unet, loras, "lora_unet_"):
+            yield
+
+
+    @classmethod
+    @contextmanager
+    def apply_lora_text_encoder(
+        cls,
+        text_encoder: OnnxRuntimeModel,
+        loras: List[Tuple[LoRAModel, float]],
+    ):
+        with cls.apply_lora(text_encoder, loras, "lora_te_"):
+            yield
+
+    # based on
+    # https://github.com/ssube/onnx-web/blob/ca2e436f0623e18b4cfe8a0363fcfcf10508acf7/api/onnx_web/convert/diffusion/lora.py#L323
+    @classmethod
+    @contextmanager
+    def apply_lora(
+        cls,
+        model: IAIOnnxRuntimeModel,
+        loras: List[Tuple[LoraModel, float]],
+        prefix: str,
+    ):
+        from .models.base import IAIOnnxRuntimeModel
+        if not isinstance(model, IAIOnnxRuntimeModel):
+            raise Exception("Only IAIOnnxRuntimeModel models supported")
+
+        orig_weights = dict()
+
+        try:
+            blended_loras = dict()
+
+            for lora, lora_weight in loras:
+                for layer_key, layer in lora.layers.items():
+                    if not layer_key.startswith(prefix):
+                        continue
+
+                    layer_key = layer_key.replace(prefix, "")
+                    layer_weight = layer.get_weight().detach().cpu().numpy() * lora_weight
+                    if layer_key is blended_loras:
+                        blended_loras[layer_key] += layer_weight
+                    else:
+                        blended_loras[layer_key] = layer_weight
+
+            node_names = dict()
+            for node in model.nodes.values():
+                node_names[node.name.replace("/", "_").replace(".", "_").lstrip("_")] = node.name
+
+            for layer_key, lora_weight in blended_loras.items():
+                conv_key = layer_key + "_Conv"
+                gemm_key = layer_key + "_Gemm"
+                matmul_key = layer_key + "_MatMul"
+
+                if conv_key in node_names or gemm_key in node_names:
+                    if conv_key in node_names:
+                        conv_node = model.nodes[node_names[conv_key]]
+                    else:
+                        conv_node = model.nodes[node_names[gemm_key]]
+
+                    weight_name = [n for n in conv_node.input if ".weight" in n][0]
+                    orig_weight = model.tensors[weight_name]
+
+                    if orig_weight.shape[-2:] == (1, 1):
+                        if lora_weight.shape[-2:] == (1, 1):
+                            new_weight = orig_weight.squeeze((3, 2)) + lora_weight.squeeze((3, 2))
+                        else:
+                            new_weight = orig_weight.squeeze((3, 2)) + lora_weight
+
+                        new_weight = np.expand_dims(new_weight, (2, 3))
+                    else:
+                        if orig_weight.shape != lora_weight.shape:
+                            new_weight = orig_weight + lora_weight.reshape(orig_weight.shape)
+                        else:
+                            new_weight = orig_weight + lora_weight
+
+                    orig_weights[weight_name] = orig_weight
+                    model.tensors[weight_name] = new_weight.astype(orig_weight.dtype)
+
+                elif matmul_key in node_names:
+                    weight_node = model.nodes[node_names[matmul_key]]
+                    matmul_name = [n for n in weight_node.input if "MatMul" in n][0]
+
+                    orig_weight = model.tensors[matmul_name]
+                    new_weight = orig_weight + lora_weight.transpose()
+
+                    orig_weights[matmul_name] = orig_weight
+                    model.tensors[matmul_name] = new_weight.astype(orig_weight.dtype)
+
+                else:
+                    # warn? err?
+                    pass
+
+            yield
+
+        finally:
+            # restore original weights
+            for name, orig_weight in orig_weights.items():
+                model.tensors[name] = orig_weight
+
+
+
+    @classmethod
+    @contextmanager
+    def apply_ti(
+        cls,
+        tokenizer: CLIPTokenizer,
+        text_encoder: IAIOnnxRuntimeModel,
+        ti_list: List[Any],
+    ) -> Tuple[CLIPTokenizer, TextualInversionManager]:
+        from .models.base import IAIOnnxRuntimeModel
+        if not isinstance(text_encoder, IAIOnnxRuntimeModel):
+            raise Exception("Only IAIOnnxRuntimeModel models supported")
+
+        orig_embeddings = None
+
+        try:
+            ti_tokenizer = copy.deepcopy(tokenizer)
+            ti_manager = TextualInversionManager(ti_tokenizer)
+
+            def _get_trigger(ti, index):
+                trigger = ti.name
+                if index > 0:
+                    trigger += f"-!pad-{i}"
+                return f"<{trigger}>"
+
+            # modify tokenizer
+            new_tokens_added = 0
+            for ti in ti_list:
+                for i in range(ti.embedding.shape[0]):
+                    new_tokens_added += ti_tokenizer.add_tokens(_get_trigger(ti, i))
+
+            # modify text_encoder
+            orig_embeddings = text_encoder.tensors["text_model.embeddings.token_embedding.weight"]
+
+            embeddings = np.concatenate(
+                (
+                    np.copy(orig_embeddings),
+                    np.zeros((new_tokens_added, orig_embeddings.shape[1]))
+                ),
+                axis=0,
+            )
+
+            for ti in ti_list:
+                ti_tokens = []
+                for i in range(ti.embedding.shape[0]):
+                    embedding = ti.embedding[i].detach().numpy()
+                    trigger = _get_trigger(ti, i)
+
+                    token_id = ti_tokenizer.convert_tokens_to_ids(trigger)
+                    if token_id == ti_tokenizer.unk_token_id:
+                        raise RuntimeError(f"Unable to find token id for token '{trigger}'")
+
+                    if embeddings[token_id].shape != embedding.shape:
+                        raise ValueError(
+                            f"Cannot load embedding for {trigger}. It was trained on a model with token dimension {embedding.shape[0]}, but the current model has token dimension {embeddings[token_id].shape[0]}."
+                        )
+
+                    embeddings[token_id] = embedding
+                    ti_tokens.append(token_id)
+
+                if len(ti_tokens) > 1:
+                    ti_manager.pad_tokens[ti_tokens[0]] = ti_tokens[1:]
+
+            text_encoder.tensors["text_model.embeddings.token_embedding.weight"] = embeddings.astype(orig_embeddings.dtype)
+
+            yield ti_tokenizer, ti_manager
+
+        finally:
+            # restore
+            if orig_embeddings is not None:
+                text_encoder.tensors["text_model.embeddings.token_embedding.weight"] = orig_embeddings

invokeai/backend/model_management/model_cache.py

@@ -36,6 +36,9 @@ from .models import BaseModelType, ModelType, SubModelType, ModelBase
 # Default is roughly enough to hold three fp16 diffusers models in RAM simultaneously
 DEFAULT_MAX_CACHE_SIZE = 6.0
 
+# amount of GPU memory to hold in reserve for use by generations (GB)
+DEFAULT_MAX_VRAM_CACHE_SIZE= 2.75
+
 # actual size of a gig
 GIG = 1073741824
 
@@ -82,6 +85,7 @@ class ModelCache(object):
     def __init__(
         self,
         max_cache_size: float=DEFAULT_MAX_CACHE_SIZE,
+        max_vram_cache_size: float=DEFAULT_MAX_VRAM_CACHE_SIZE,
         execution_device: torch.device=torch.device('cuda'),
         storage_device: torch.device=torch.device('cpu'),
         precision: torch.dtype=torch.float16,
@@ -99,12 +103,12 @@ class ModelCache(object):
         :param sequential_offload: Conserve VRAM by loading and unloading each stage of the pipeline sequentially
         :param sha_chunksize: Chunksize to use when calculating sha256 model hash
         '''
-        #max_cache_size = 9999
         self.model_infos: Dict[str, ModelBase] = dict()
-        self.lazy_offloading = lazy_offloading
-        #self.sequential_offload: bool=sequential_offload
+        # allow lazy offloading only when vram cache enabled
+        self.lazy_offloading = lazy_offloading and max_vram_cache_size > 0
         self.precision: torch.dtype=precision
-        self.max_cache_size: int=max_cache_size
+        self.max_cache_size: float=max_cache_size
+        self.max_vram_cache_size: float=max_vram_cache_size
         self.execution_device: torch.device=execution_device
         self.storage_device: torch.device=storage_device
         self.sha_chunksize=sha_chunksize
@@ -201,14 +205,22 @@ class ModelCache(object):
             self._cache_stack.remove(key)
         self._cache_stack.append(key)
 
-        return self.ModelLocker(self, key, cache_entry.model, gpu_load)
+        return self.ModelLocker(self, key, cache_entry.model, gpu_load, cache_entry.size)
 
     class ModelLocker(object):
-        def __init__(self, cache, key, model, gpu_load):
+        def __init__(self, cache, key, model, gpu_load, size_needed):
+            '''
+            :param cache: The model_cache object
+            :param key: The key of the model to lock in GPU
+            :param model: The model to lock
+            :param gpu_load: True if load into gpu
+            :param size_needed: Size of the model to load
+            '''
             self.gpu_load = gpu_load
             self.cache = cache
             self.key = key
             self.model = model
+            self.size_needed = size_needed
             self.cache_entry = self.cache._cached_models[self.key]
 
         def __enter__(self) -> Any:
@@ -222,7 +234,7 @@ class ModelCache(object):
 
                 try:
                     if self.cache.lazy_offloading:
-                       self.cache._offload_unlocked_models()
+                       self.cache._offload_unlocked_models(self.size_needed)
                        
                     if self.model.device != self.cache.execution_device:
                         self.cache.logger.debug(f'Moving {self.key} into {self.cache.execution_device}')
@@ -316,6 +328,25 @@ class ModelCache(object):
 
             refs = sys.getrefcount(cache_entry.model)
 
+            # manualy clear local variable references of just finished function calls
+            # for some reason python don't want to collect it even by gc.collect() immidiately
+            if refs > 2:
+                while True:
+                    cleared = False
+                    for referrer in gc.get_referrers(cache_entry.model):
+                        if type(referrer).__name__ == "frame":
+                            # RuntimeError: cannot clear an executing frame
+                            with suppress(RuntimeError):
+                                referrer.clear()
+                                cleared = True
+                                #break
+
+                    # repeat if referrers changes(due to frame clear), else exit loop
+                    if cleared:
+                        gc.collect()
+                    else:
+                        break
+
             device = cache_entry.model.device if hasattr(cache_entry.model, "device") else None
             self.logger.debug(f"Model: {model_key}, locks: {cache_entry._locks}, device: {device}, loaded: {cache_entry.loaded}, refs: {refs}")
 
@@ -337,14 +368,23 @@ class ModelCache(object):
 
         self.logger.debug(f"After unloading: cached_models={len(self._cached_models)}")
 
-
-    def _offload_unlocked_models(self):
-        for model_key, cache_entry in self._cached_models.items():
+    def _offload_unlocked_models(self, size_needed: int=0):
+        reserved = self.max_vram_cache_size * GIG
+        vram_in_use = torch.cuda.memory_allocated()
+        self.logger.debug(f'{(vram_in_use/GIG):.2f}GB VRAM used for models; max allowed={(reserved/GIG):.2f}GB')
+        for model_key, cache_entry in sorted(self._cached_models.items(), key=lambda x:x[1].size):
+            if vram_in_use <= reserved:
+                break
             if not cache_entry.locked and cache_entry.loaded:
                 self.logger.debug(f'Offloading {model_key} from {self.execution_device} into {self.storage_device}')
                 with VRAMUsage() as mem:
                     cache_entry.model.to(self.storage_device)
                 self.logger.debug(f'GPU VRAM freed: {(mem.vram_used/GIG):.2f} GB')
+                vram_in_use += mem.vram_used  # note vram_used is negative
+                self.logger.debug(f'{(vram_in_use/GIG):.2f}GB VRAM used for models; max allowed={(reserved/GIG):.2f}GB')
+
+        gc.collect()
+        torch.cuda.empty_cache()
         
     def _local_model_hash(self, model_path: Union[str, Path]) -> str:
         sha = hashlib.sha256()

invokeai/backend/model_management/model_manager.py

@@ -106,16 +106,16 @@ providing information about a model defined in models.yaml. For example:
 
    >>> models = mgr.list_models()
    >>> json.dumps(models[0])
-  {"path": "/home/lstein/invokeai-main/models/sd-1/controlnet/canny", 
-    "model_format": "diffusers", 
-    "name": "canny", 
-    "base_model": "sd-1", 
+  {"path": "/home/lstein/invokeai-main/models/sd-1/controlnet/canny",
+    "model_format": "diffusers",
+    "name": "canny",
+    "base_model": "sd-1",
     "type": "controlnet"
    }
 
 You can filter by model type and base model as shown here:
 
-  
+
   controlnets = mgr.list_models(model_type=ModelType.ControlNet,
                                 base_model=BaseModelType.StableDiffusion1)
   for c in controlnets:
@@ -140,14 +140,14 @@ Layout of the `models` directory:
 
  models
  ├── sd-1
- │   ├── controlnet
- │   ├── lora
- │   ├── main
- │   └── embedding
+ │   ├── controlnet
+ │   ├── lora
+ │   ├── main
+ │   └── embedding
  ├── sd-2
- │   ├── controlnet
- │   ├── lora
- │   ├── main
+ │   ├── controlnet
+ │   ├── lora
+ │   ├── main
  │   └── embedding
  └── core
      ├── face_reconstruction
@@ -195,7 +195,7 @@ name, base model, type and a dict of model attributes. See
 `invokeai/backend/model_management/models` for the attributes required
 by each model type.
 
-A model can be deleted using `del_model()`, providing the same 
+A model can be deleted using `del_model()`, providing the same
 identifying information as `get_model()`
 
 The `heuristic_import()` method will take a set of strings
@@ -231,6 +231,7 @@ from __future__ import annotations
 import os
 import hashlib
 import textwrap
+import yaml
 from dataclasses import dataclass
 from pathlib import Path
 from typing import Optional, List, Tuple, Union, Dict, Set, Callable, types
@@ -246,11 +247,12 @@ import invokeai.backend.util.logging as logger
 from invokeai.app.services.config import InvokeAIAppConfig
 from invokeai.backend.util import CUDA_DEVICE, Chdir
 from .model_cache import ModelCache, ModelLocker
+from .model_search import ModelSearch
 from .models import (
     BaseModelType, ModelType, SubModelType,
     ModelError, SchedulerPredictionType, MODEL_CLASSES,
-    ModelConfigBase, ModelNotFoundException,
-    )
+    ModelConfigBase, ModelNotFoundException, InvalidModelException,
+)
 
 # We are only starting to number the config file with release 3.
 # The config file version doesn't have to start at release version, but it will help
@@ -274,10 +276,6 @@ class ModelInfo():
     def __exit__(self,*args, **kwargs):
         self.context.__exit__(*args, **kwargs)
 
-class InvalidModelError(Exception):
-    "Raised when an invalid model is requested"
-    pass
-
 class AddModelResult(BaseModel):
     name: str = Field(description="The name of the model after installation")
     model_type: ModelType = Field(description="The type of model")
@@ -306,7 +304,7 @@ class ModelManager(object):
         logger: types.ModuleType = logger,
     ):
         """
-        Initialize with the path to the models.yaml config file. 
+        Initialize with the path to the models.yaml config file.
         Optional parameters are the torch device type, precision, max_models,
         and sequential_offload boolean. Note that the default device
         type and precision are set up for a CUDA system running at half precision.
@@ -314,6 +312,9 @@ class ModelManager(object):
         self.config_path = None
         if isinstance(config, (str, Path)):
             self.config_path = Path(config)
+            if not self.config_path.exists():
+                logger.warning(f'The file {self.config_path} was not found. Initializing a new file')
+                self.initialize_model_config(self.config_path)
             config = OmegaConf.load(self.config_path)
 
         elif not isinstance(config, DictConfig):
@@ -323,8 +324,30 @@ class ModelManager(object):
         # TODO: metadata not found
         # TODO: version check
 
+        self.app_config = InvokeAIAppConfig.get_config()
+        self.logger = logger
+        self.cache = ModelCache(
+            max_cache_size=max_cache_size,
+            max_vram_cache_size = self.app_config.max_vram_cache_size,
+            execution_device = device_type,
+            precision = precision,
+            sequential_offload = sequential_offload,
+            logger = logger,
+        )
+
+        self._read_models(config)
+
+    def _read_models(self, config: Optional[DictConfig] = None):
+        if not config:
+            if self.config_path:
+                config = OmegaConf.load(self.config_path)
+            else:
+                return
+
         self.models = dict()
         for model_key, model_config in config.items():
+            if model_key.startswith('_'):
+                continue
             model_name, base_model, model_type = self.parse_key(model_key)
             model_class = MODEL_CLASSES[base_model][model_type]
             # alias for config file
@@ -332,20 +355,20 @@ class ModelManager(object):
             self.models[model_key] = model_class.create_config(**model_config)
 
         # check config version number and update on disk/RAM if necessary
-        self.app_config = InvokeAIAppConfig.get_config()
-        self.logger = logger
-        self.cache = ModelCache(
-            max_cache_size=max_cache_size,
-            execution_device = device_type,
-            precision = precision,
-            sequential_offload = sequential_offload,
-            logger = logger,
-        )
         self.cache_keys = dict()
 
         # add controlnet, lora and textual_inversion models from disk
         self.scan_models_directory()
 
+    def sync_to_config(self):
+        """
+        Call this when `models.yaml` has been changed externally.
+        This will reinitialize internal data structures
+        """
+        # Reread models directory; note that this will reinitialize the cache,
+        # causing otherwise unreferenced models to be removed from memory
+        self._read_models()
+
     def model_exists(
         self,
         model_name: str,
@@ -386,6 +409,16 @@ class ModelManager(object):
     def _get_model_cache_path(self, model_path):
         return self.app_config.models_path / ".cache" / hashlib.md5(str(model_path).encode()).hexdigest()
 
+    @classmethod
+    def initialize_model_config(cls, config_path: Path):
+        """Create empty config file"""
+        with open(config_path,'w') as yaml_file:
+            yaml_file.write(yaml.dump({'__metadata__':
+                                       {'version':'3.0.0'}
+                                       }
+                                      )
+                            )
+
     def get_model(
         self,
         model_name: str,
@@ -398,7 +431,7 @@ class ModelManager(object):
         :param model_name: symbolic name of the model in models.yaml
         :param model_type: ModelType enum indicating the type of model to return
         :param base_model: BaseModelType enum indicating the base model used by this model
-        :param submode_typel: an ModelType enum indicating the portion of 
+        :param submode_typel: an ModelType enum indicating the portion of
                the model to retrieve (e.g. ModelType.Vae)
         """
         model_class = MODEL_CLASSES[base_model][model_type]
@@ -423,7 +456,7 @@ class ModelManager(object):
                 raise ModelNotFoundException(f"Model not found - {model_key}")
 
         # vae/movq override
-        # TODO: 
+        # TODO:
         if submodel_type is not None and hasattr(model_config, submodel_type):
             override_path = getattr(model_config, submodel_type)
             if override_path:
@@ -456,7 +489,7 @@ class ModelManager(object):
         self.cache_keys[model_key].add(model_context.key)
 
         model_hash = "<NO_HASH>" # TODO:
-            
+
         return ModelInfo(
             context = model_context,
             name = model_name,
@@ -485,7 +518,7 @@ class ModelManager(object):
 
     def model_names(self) -> List[Tuple[str, BaseModelType, ModelType]]:
         """
-        Return a list of (str, BaseModelType, ModelType) corresponding to all models 
+        Return a list of (str, BaseModelType, ModelType) corresponding to all models
         known to the configuration.
         """
         return [(self.parse_key(x)) for x in self.models.keys()]
@@ -516,7 +549,10 @@ class ModelManager(object):
         model_keys = [self.create_key(model_name, base_model, model_type)] if model_name else sorted(self.models, key=str.casefold)
         models = []
         for model_key in model_keys:
-            model_config = self.models[model_key]
+            model_config = self.models.get(model_key)
+            if not model_config:
+                self.logger.error(f'Unknown model {model_name}')
+                raise ModelNotFoundException(f'Unknown model {model_name}')
 
             cur_model_name, cur_base_model, cur_model_type = self.parse_key(model_key)
             if base_model is not None and cur_base_model != base_model:
@@ -527,11 +563,14 @@ class ModelManager(object):
             model_dict = dict(
                 **model_config.dict(exclude_defaults=True),
                 # OpenAPIModelInfoBase
-                name=cur_model_name,
+                model_name=cur_model_name,
                 base_model=cur_base_model,
-                type=cur_model_type,
+                model_type=cur_model_type,
             )
 
+            # expose paths as absolute to help web UI
+            if path := model_dict.get('path'):
+                model_dict['path'] = str(self.app_config.root_path / path)
             models.append(model_dict)
 
         return models
@@ -560,7 +599,7 @@ class ModelManager(object):
         model_cfg = self.models.pop(model_key, None)
 
         if model_cfg is None:
-            raise KeyError(f"Unknown model {model_key}")
+            raise ModelNotFoundException(f"Unknown model {model_key}")
 
         # note: it not garantie to release memory(model can has other references)
         cache_ids = self.cache_keys.pop(model_key, [])
@@ -578,6 +617,7 @@ class ModelManager(object):
                 rmtree(str(model_path))
             else:
                 model_path.unlink()
+        self.commit()
 
     # LS: tested
     def add_model(
@@ -598,6 +638,10 @@ class ModelManager(object):
         The returned dict has the same format as the dict returned by
         model_info().
         """
+        # relativize paths as they go in - this makes it easier to move the root directory around
+        if path := model_attributes.get('path'):
+            if Path(path).is_relative_to(self.app_config.root_path):
+                model_attributes['path'] = str(Path(path).relative_to(self.app_config.root_path))
 
         model_class = MODEL_CLASSES[base_model][model_type]
         model_config = model_class.create_config(**model_attributes)
@@ -634,11 +678,61 @@ class ModelManager(object):
             config = model_config,
         )
 
+    def rename_model(
+            self,
+            model_name: str,
+            base_model: BaseModelType,
+            model_type: ModelType,
+            new_name: str = None,
+            new_base: BaseModelType = None,
+    ):
+        '''
+        Rename or rebase a model.
+        '''
+        if new_name is None and new_base is None:
+            self.logger.error("rename_model() called with neither a new_name nor a new_base. {model_name} unchanged.")
+            return
+
+        model_key = self.create_key(model_name, base_model, model_type)
+        model_cfg = self.models.get(model_key, None)
+        if not model_cfg:
+            raise ModelNotFoundException(f"Unknown model: {model_key}")
+
+        old_path = self.app_config.root_path / model_cfg.path
+        new_name = new_name or model_name
+        new_base = new_base or base_model
+        new_key = self.create_key(new_name, new_base, model_type)
+        if new_key in self.models:
+            raise ValueError(f'Attempt to overwrite existing model definition "{new_key}"')
+
+        # if this is a model file/directory that we manage ourselves, we need to move it
+        if old_path.is_relative_to(self.app_config.models_path):
+            new_path = self.app_config.root_path / 'models' / BaseModelType(new_base).value / ModelType(model_type).value / new_name
+            move(old_path, new_path)
+            model_cfg.path = str(new_path.relative_to(self.app_config.root_path))
+
+        # clean up caches
+        old_model_cache = self._get_model_cache_path(old_path)
+        if old_model_cache.exists():
+            if old_model_cache.is_dir():
+                rmtree(str(old_model_cache))
+            else:
+                old_model_cache.unlink()
+
+        cache_ids = self.cache_keys.pop(model_key, [])
+        for cache_id in cache_ids:
+            self.cache.uncache_model(cache_id)
+
+        self.models.pop(model_key, None) # delete
+        self.models[new_key] = model_cfg
+        self.commit()
+
     def convert_model (
             self,
             model_name: str,
             base_model: BaseModelType,
             model_type: Union[ModelType.Main,ModelType.Vae],
+            dest_directory: Optional[Path]=None,
     ) -> AddModelResult:
         '''
         Convert a checkpoint file into a diffusers folder, deleting the cached
@@ -665,14 +759,14 @@ class ModelManager(object):
                                )
         checkpoint_path = self.app_config.root_path / info["path"]
         old_diffusers_path = self.app_config.models_path / model.location
-        new_diffusers_path = self.app_config.models_path / base_model.value / model_type.value / model_name
+        new_diffusers_path = (dest_directory or self.app_config.models_path / base_model.value / model_type.value) / model_name
         if new_diffusers_path.exists():
             raise ValueError(f"A diffusers model already exists at {new_diffusers_path}")
 
         try:
             move(old_diffusers_path,new_diffusers_path)
             info["model_format"] = "diffusers"
-            info["path"] = str(new_diffusers_path.relative_to(self.app_config.root_path))
+            info["path"] = str(new_diffusers_path) if dest_directory else str(new_diffusers_path.relative_to(self.app_config.root_path))
             info.pop('config')
 
             result = self.add_model(model_name, base_model, model_type,
@@ -682,12 +776,12 @@ class ModelManager(object):
             # something went wrong, so don't leave dangling diffusers model in directory or it will cause a duplicate model error!
             rmtree(new_diffusers_path)
             raise
-        
+
         if checkpoint_path.exists() and checkpoint_path.is_relative_to(self.app_config.models_path):
             checkpoint_path.unlink()
-        
+
         return result
-    
+
     def search_models(self, search_folder):
         self.logger.info(f"Finding Models In: {search_folder}")
         models_folder_ckpt = Path(search_folder).glob("**/*.ckpt")
@@ -730,10 +824,14 @@ class ModelManager(object):
         assert config_file_path is not None,'no config file path to write to'
         config_file_path = self.app_config.root_path / config_file_path
         tmpfile = os.path.join(os.path.dirname(config_file_path), "new_config.tmp")
-        with open(tmpfile, "w", encoding="utf-8") as outfile:
-            outfile.write(self.preamble())
-            outfile.write(yaml_str)
-        os.replace(tmpfile, config_file_path)
+        try:
+            with open(tmpfile, "w", encoding="utf-8") as outfile:
+                outfile.write(self.preamble())
+                outfile.write(yaml_str)
+            os.replace(tmpfile, config_file_path)
+        except OSError as err:
+            self.logger.warning(f"Could not modify the config file at {config_file_path}")
+            self.logger.warning(err)
 
     def preamble(self) -> str:
         """
@@ -802,6 +900,8 @@ class ModelManager(object):
                                 model_config: ModelConfigBase = model_class.probe_config(str(model_path))
                                 self.models[model_key] = model_config
                                 new_models_found = True
+                            except InvalidModelException:
+                                self.logger.warning(f"Not a valid model: {model_path}")
                             except NotImplementedError as e:
                                 self.logger.warning(e)
 
@@ -810,6 +910,7 @@ class ModelManager(object):
         if (new_models_found or imported_models) and self.config_path:
             self.commit()
 
+
     def autoimport(self)->Dict[str, AddModelResult]:
         '''
         Scan the autoimport directory (if defined) and import new models, delete defunct models.
@@ -817,57 +918,41 @@ class ModelManager(object):
         # avoid circular import
         from invokeai.backend.install.model_install_backend import ModelInstall
         from invokeai.frontend.install.model_install import ask_user_for_prediction_type
-        
+
+        class ScanAndImport(ModelSearch):
+            def __init__(self, directories, logger, ignore: Set[Path], installer: ModelInstall):
+                super().__init__(directories, logger)
+                self.installer = installer
+                self.ignore = ignore
+
+            def on_search_started(self):
+                self.new_models_found = dict()
+
+            def on_model_found(self, model: Path):
+                if model not in self.ignore:
+                    self.new_models_found.update(self.installer.heuristic_import(model))
+
+            def on_search_completed(self):
+                self.logger.info(f'Scanned {self._items_scanned} files and directories, imported {len(self.new_models_found)} models')
+
+            def models_found(self):
+                return self.new_models_found
+
+
         installer = ModelInstall(config = self.app_config,
                                  model_manager = self,
                                  prediction_type_helper = ask_user_for_prediction_type,
                                  )
-        
-        scanned_dirs = set()
-        
         config = self.app_config
-        known_paths = {(self.app_config.root_path / x['path']) for x in self.list_models()}
-
-        for autodir in [config.autoimport_dir,
-                        config.lora_dir,
-                        config.embedding_dir,
-                        config.controlnet_dir]:
-            if autodir is None:
-                continue
-
-            self.logger.info(f'Scanning {autodir} for models to import')
-            installed = dict()
-        
-            autodir = self.app_config.root_path / autodir
-            if not autodir.exists():
-                continue
-
-            items_scanned = 0
-            new_models_found = dict()
-            
-            for root, dirs, files in os.walk(autodir):
-                items_scanned += len(dirs) + len(files)
-                for d in dirs:
-                    path = Path(root) / d
-                    if path in known_paths or path.parent in scanned_dirs:
-                        scanned_dirs.add(path)
-                        continue
-                    if any([(path/x).exists() for x in {'config.json','model_index.json','learned_embeds.bin','pytorch_lora_weights.bin'}]):
-                        new_models_found.update(installer.heuristic_import(path))
-                        scanned_dirs.add(path)
-
-                for f in files:
-                    path = Path(root) / f
-                    if path in known_paths or path.parent in scanned_dirs:
-                        continue
-                    if path.suffix in {'.ckpt','.bin','.pth','.safetensors','.pt'}:
-                        import_result = installer.heuristic_import(path)
-                        new_models_found.update(import_result)
-
-            self.logger.info(f'Scanned {items_scanned} files and directories, imported {len(new_models_found)} models')
-            installed.update(new_models_found)
-
-        return installed
+        known_paths = {config.root_path / x['path'] for x in self.list_models()}
+        directories = {config.root_path / x for x in [config.autoimport_dir,
+                                                      config.lora_dir,
+                                                      config.embedding_dir,
+                                                      config.controlnet_dir]
+                       }
+        scanner = ScanAndImport(directories, self.logger, ignore=known_paths, installer=installer)
+        scanner.search()
+        return scanner.models_found()
 
     def heuristic_import(self,
                          items_to_import: Set[str],
@@ -890,18 +975,18 @@ class ModelManager(object):
         that model.
 
         May return the following exceptions:
-        - KeyError   - one or more of the items to import is not a valid path, repo_id or URL
+        - ModelNotFoundException   - one or more of the items to import is not a valid path, repo_id or URL
         - ValueError - a corresponding model already exists
         '''
         # avoid circular import here
         from invokeai.backend.install.model_install_backend import ModelInstall
         successfully_installed = dict()
-        
+
         installer = ModelInstall(config = self.app_config,
                                  prediction_type_helper = prediction_type_helper,
                                  model_manager = self)
         for thing in items_to_import:
             installed = installer.heuristic_import(thing)
             successfully_installed.update(installed)
-        self.commit()                
+        self.commit()
         return successfully_installed

invokeai/backend/model_management/model_merge.py

@@ -11,7 +11,7 @@ from enum import Enum
 from pathlib import Path
 from diffusers import DiffusionPipeline
 from diffusers import logging as dlogging
-from typing import List, Union
+from typing import List, Union, Optional
 
 import invokeai.backend.util.logging as logger
 
@@ -74,6 +74,7 @@ class ModelMerger(object):
             alpha: float = 0.5,
             interp: MergeInterpolationMethod = None,
             force: bool = False,
+            merge_dest_directory: Optional[Path] = None,
             **kwargs,
     ) -> AddModelResult:
         """
@@ -85,7 +86,7 @@ class ModelMerger(object):
         :param interp: The interpolation method to use for the merging. Supports "weighted_average", "sigmoid", "inv_sigmoid", "add_difference" and None.
                    Passing None uses the default interpolation which is weighted sum interpolation. For merging three checkpoints, only "add_difference" is supported. Add_difference is A+(B-C).
         :param force:  Whether to ignore mismatch in model_config.json for the current models. Defaults to False.
-
+        :param merge_dest_directory: Save the merged model to the designated directory (with 'merged_model_name' appended)
         **kwargs - the default DiffusionPipeline.get_config_dict kwargs:
              cache_dir, resume_download, force_download, proxies, local_files_only, use_auth_token, revision, torch_dtype, device_map
         """
@@ -111,7 +112,7 @@ class ModelMerger(object):
         merged_pipe = self.merge_diffusion_models(
            model_paths, alpha, merge_method, force, **kwargs
         )
-        dump_path = config.models_path / base_model.value / ModelType.Main.value
+        dump_path = Path(merge_dest_directory) if merge_dest_directory else config.models_path / base_model.value / ModelType.Main.value
         dump_path.mkdir(parents=True, exist_ok=True)
         dump_path = dump_path / merged_model_name
 

invokeai/backend/model_management/model_probe.py

@@ -12,6 +12,7 @@ from picklescan.scanner import scan_file_path
 from .models import (
     BaseModelType, ModelType, ModelVariantType,
     SchedulerPredictionType, SilenceWarnings,
+    InvalidModelException
 )
 from .models.base import read_checkpoint_meta
 
@@ -22,7 +23,7 @@ class ModelProbeInfo(object):
     variant_type: ModelVariantType
     prediction_type: SchedulerPredictionType
     upcast_attention: bool
-    format: Literal['diffusers','checkpoint', 'lycoris']
+    format: Literal['diffusers','checkpoint', 'lycoris', 'olive']
     image_size: int
 
 class ProbeBase(object):
@@ -38,6 +39,8 @@ class ModelProbe(object):
 
     CLASS2TYPE = {
         'StableDiffusionPipeline' : ModelType.Main,
+        'StableDiffusionXLPipeline' : ModelType.Main,
+        'StableDiffusionXLImg2ImgPipeline' : ModelType.Main,
         'AutoencoderKL' : ModelType.Vae,
         'ControlNetModel' : ModelType.ControlNet,
     }
@@ -59,7 +62,7 @@ class ModelProbe(object):
         elif isinstance(model,(dict,ModelMixin,ConfigMixin)):
             return cls.probe(model_path=None, model=model, prediction_type_helper=prediction_type_helper)
         else:
-            raise Exception("model parameter {model} is neither a Path, nor a model")
+            raise InvalidModelException("model parameter {model} is neither a Path, nor a model")
 
     @classmethod
     def probe(cls,
@@ -99,9 +102,10 @@ class ModelProbe(object):
                 upcast_attention = (base_type==BaseModelType.StableDiffusion2 \
                                      and prediction_type==SchedulerPredictionType.VPrediction),
                 format = format,
-                image_size = 768 if (base_type==BaseModelType.StableDiffusion2 \
-                                     and prediction_type==SchedulerPredictionType.VPrediction \
-                                     ) else 512,
+                image_size = 1024 if (base_type in {BaseModelType.StableDiffusionXL,BaseModelType.StableDiffusionXLRefiner}) else \
+                              768 if (base_type==BaseModelType.StableDiffusion2 \
+                                     and prediction_type==SchedulerPredictionType.VPrediction ) else \
+                              512
             )
         except Exception:
             raise
@@ -138,7 +142,7 @@ class ModelProbe(object):
             if len(ckpt) < 10 and all(isinstance(v, torch.Tensor) for v in ckpt.values()):
                 return ModelType.TextualInversion
         
-        raise ValueError(f"Unable to determine model type for {model_path}")
+        raise InvalidModelException(f"Unable to determine model type for {model_path}")
 
     @classmethod
     def get_model_type_from_folder(cls, folder_path: Path, model: ModelMixin)->ModelType:
@@ -168,7 +172,7 @@ class ModelProbe(object):
             return type
 
         # give up
-        raise ValueError(f"Unable to determine model type for {folder_path}")
+        raise InvalidModelException(f"Unable to determine model type for {folder_path}")
 
     @classmethod
     def _scan_and_load_checkpoint(cls,model_path: Path)->dict:
@@ -237,7 +241,7 @@ class CheckpointProbeBase(ProbeBase):
         elif in_channels == 4:
             return ModelVariantType.Normal
         else:
-            raise Exception("Cannot determine variant type")
+            raise InvalidModelException(f"Cannot determine variant type (in_channels={in_channels}) at {self.checkpoint_path}")
 
 class PipelineCheckpointProbe(CheckpointProbeBase):
     def get_base_type(self)->BaseModelType:
@@ -248,7 +252,10 @@ class PipelineCheckpointProbe(CheckpointProbeBase):
             return BaseModelType.StableDiffusion1
         if key_name in state_dict and state_dict[key_name].shape[-1] == 1024:
             return BaseModelType.StableDiffusion2
-        raise Exception("Cannot determine base type")
+        # TODO: Verify that this is correct! Need an XL checkpoint file for this.
+        if key_name in state_dict and state_dict[key_name].shape[-1] == 2048:
+            return BaseModelType.StableDiffusionXL
+        raise InvalidModelException("Cannot determine base type")
 
     def get_scheduler_prediction_type(self)->SchedulerPredictionType:
         type = self.get_base_type()
@@ -329,7 +336,7 @@ class ControlNetCheckpointProbe(CheckpointProbeBase):
                 return BaseModelType.StableDiffusion2
             elif self.checkpoint_path and self.helper:
                 return self.helper(self.checkpoint_path)
-        raise Exception("Unable to determine base type for {self.checkpoint_path}")
+        raise InvalidModelException("Unable to determine base type for {self.checkpoint_path}")
 
 ########################################################
 # classes for probing folders
@@ -360,8 +367,12 @@ class PipelineFolderProbe(FolderProbeBase):
             return BaseModelType.StableDiffusion1  
         elif unet_conf['cross_attention_dim'] == 1024:
             return BaseModelType.StableDiffusion2
+        elif unet_conf['cross_attention_dim'] == 1280:
+            return BaseModelType.StableDiffusionXLRefiner
+        elif unet_conf['cross_attention_dim'] == 2048:
+            return BaseModelType.StableDiffusionXL
         else:
-            raise ValueError(f'Unknown base model for {self.folder_path}')
+            raise InvalidModelException(f'Unknown base model for {self.folder_path}')
 
     def get_scheduler_prediction_type(self)->SchedulerPredictionType:
         if self.model:
@@ -418,7 +429,7 @@ class ControlNetFolderProbe(FolderProbeBase):
     def get_base_type(self)->BaseModelType:
         config_file = self.folder_path / 'config.json'
         if not config_file.exists():
-            raise Exception(f"Cannot determine base type for {self.folder_path}")
+            raise InvalidModelException(f"Cannot determine base type for {self.folder_path}")
         with open(config_file,'r') as file:
             config = json.load(file)
         # no obvious way to distinguish between sd2-base and sd2-768
@@ -435,7 +446,7 @@ class LoRAFolderProbe(FolderProbeBase):
                 model_file = base_file
                 break
         if not model_file:
-            raise Exception('Unknown LoRA format encountered')
+            raise InvalidModelException('Unknown LoRA format encountered')
         return LoRACheckpointProbe(model_file,None).get_base_type()
 
 ############## register probe classes ######

invokeai/backend/model_management/model_search.py

@@ -0,0 +1,103 @@
+# Copyright 2023, Lincoln D. Stein and the InvokeAI Team
+"""
+Abstract base class for recursive directory search for models.
+"""
+
+import os
+from abc import ABC, abstractmethod
+from typing import List, Set, types
+from pathlib import Path
+
+import invokeai.backend.util.logging as logger
+
+class ModelSearch(ABC):
+    def __init__(self, directories: List[Path], logger: types.ModuleType=logger):
+        """
+        Initialize a recursive model directory search.
+        :param directories: List of directory Paths to recurse through
+        :param logger: Logger to use
+        """
+        self.directories = directories
+        self.logger = logger
+        self._items_scanned = 0
+        self._models_found = 0
+        self._scanned_dirs = set()
+        self._scanned_paths = set()
+        self._pruned_paths = set()
+
+    @abstractmethod
+    def on_search_started(self):
+        """
+        Called before the scan starts.
+        """
+        pass
+
+    @abstractmethod
+    def on_model_found(self, model: Path):
+        """
+        Process a found model. Raise an exception if something goes wrong.
+        :param model: Model to process - could be a directory or checkpoint.
+        """
+        pass
+
+    @abstractmethod
+    def on_search_completed(self):
+        """
+        Perform some activity when the scan is completed. May use instance
+        variables, items_scanned and models_found
+        """
+        pass
+
+    def search(self):
+        self.on_search_started()
+        for dir in self.directories:
+            self.walk_directory(dir)
+        self.on_search_completed()
+
+    def walk_directory(self, path: Path):
+        for root, dirs, files in os.walk(path):
+            if str(Path(root).name).startswith('.'):
+                self._pruned_paths.add(root)
+            if any([Path(root).is_relative_to(x) for x in self._pruned_paths]):
+                continue
+            
+            self._items_scanned += len(dirs) + len(files)
+            for d in dirs:
+                path = Path(root) / d
+                if path in self._scanned_paths or path.parent in self._scanned_dirs:
+                    self._scanned_dirs.add(path)
+                    continue
+                if any([(path/x).exists() for x in {'config.json','model_index.json','learned_embeds.bin','pytorch_lora_weights.bin'}]):
+                    try:
+                        self.on_model_found(path)
+                        self._models_found += 1
+                        self._scanned_dirs.add(path)
+                    except Exception as e:
+                        self.logger.warning(str(e))
+
+            for f in files:
+                path = Path(root) / f
+                if path.parent in self._scanned_dirs:
+                    continue
+                if path.suffix in {'.ckpt','.bin','.pth','.safetensors','.pt'}:
+                    try:
+                        self.on_model_found(path)
+                        self._models_found += 1
+                    except Exception as e:
+                        self.logger.warning(str(e))
+
+class FindModels(ModelSearch):
+    def on_search_started(self):
+        self.models_found: Set[Path] = set()
+
+    def on_model_found(self,model: Path):
+        self.models_found.add(model)
+
+    def on_search_completed(self):
+        pass
+
+    def list_models(self) -> List[Path]:
+        self.search()
+        return self.models_found
+
+    

invokeai/backend/model_management/models/__init__.py

@@ -2,15 +2,19 @@ import inspect
 from enum import Enum
 from pydantic import BaseModel
 from typing import Literal, get_origin
-from .base import BaseModelType, ModelType, SubModelType, ModelBase, ModelConfigBase, ModelVariantType, SchedulerPredictionType, ModelError, SilenceWarnings, ModelNotFoundException
+from .base import BaseModelType, ModelType, SubModelType, ModelBase, ModelConfigBase, ModelVariantType, SchedulerPredictionType, ModelError, SilenceWarnings, ModelNotFoundException, InvalidModelException
 from .stable_diffusion import StableDiffusion1Model, StableDiffusion2Model
+from .sdxl import StableDiffusionXLModel
 from .vae import VaeModel
 from .lora import LoRAModel
 from .controlnet import ControlNetModel # TODO:
 from .textual_inversion import TextualInversionModel
 
+from .stable_diffusion_onnx import ONNXStableDiffusion1Model, ONNXStableDiffusion2Model
+
 MODEL_CLASSES = {
     BaseModelType.StableDiffusion1: {
+        ModelType.ONNX: ONNXStableDiffusion1Model,
         ModelType.Main: StableDiffusion1Model,
         ModelType.Vae: VaeModel,
         ModelType.Lora: LoRAModel,
@@ -18,12 +22,31 @@ MODEL_CLASSES = {
         ModelType.TextualInversion: TextualInversionModel,
     },
     BaseModelType.StableDiffusion2: {
+        ModelType.ONNX: ONNXStableDiffusion2Model,
         ModelType.Main: StableDiffusion2Model,
         ModelType.Vae: VaeModel,
         ModelType.Lora: LoRAModel,
         ModelType.ControlNet: ControlNetModel,
         ModelType.TextualInversion: TextualInversionModel,
     },
+    BaseModelType.StableDiffusionXL: {
+        ModelType.Main: StableDiffusionXLModel,
+        ModelType.Vae: VaeModel,
+        # will not work until support written
+        ModelType.Lora: LoRAModel,
+        ModelType.ControlNet: ControlNetModel,
+        ModelType.TextualInversion: TextualInversionModel,
+        ModelType.ONNX: ONNXStableDiffusion2Model,
+    },
+    BaseModelType.StableDiffusionXLRefiner: {
+        ModelType.Main: StableDiffusionXLModel,
+        ModelType.Vae: VaeModel,
+        # will not work until support written
+        ModelType.Lora: LoRAModel,
+        ModelType.ControlNet: ControlNetModel,
+        ModelType.TextualInversion: TextualInversionModel,
+        ModelType.ONNX: ONNXStableDiffusion2Model,
+    },
     #BaseModelType.Kandinsky2_1: {
     #    ModelType.Main: Kandinsky2_1Model,
     #    ModelType.MoVQ: MoVQModel,
@@ -37,9 +60,9 @@ MODEL_CONFIGS = list()
 OPENAPI_MODEL_CONFIGS = list()
 
 class OpenAPIModelInfoBase(BaseModel):
-    name: str
+    model_name: str
     base_model: BaseModelType
-    type: ModelType
+    model_type: ModelType
 
 
 for base_model, models in MODEL_CLASSES.items():
@@ -48,7 +71,9 @@ for base_model, models in MODEL_CLASSES.items():
         model_configs.discard(None)
         MODEL_CONFIGS.extend(model_configs)
 
-        for cfg in model_configs:
+        # LS: sort to get the checkpoint configs first, which makes
+        # for a better template in the Swagger docs
+        for cfg in sorted(model_configs, key=lambda x: str(x)):
             model_name, cfg_name = cfg.__qualname__.split('.')[-2:]
             openapi_cfg_name = model_name + cfg_name
             if openapi_cfg_name in vars():
@@ -56,7 +81,7 @@ for base_model, models in MODEL_CLASSES.items():
 
             api_wrapper = type(openapi_cfg_name, (cfg, OpenAPIModelInfoBase), dict(
                 __annotations__ = dict(
-                    type=Literal[model_type.value],
+                    model_type=Literal[model_type.value],
                 ),
             ))
 

invokeai/backend/model_management/models/base.py

@@ -8,22 +8,34 @@ from abc import ABCMeta, abstractmethod
 from pathlib import Path
 from picklescan.scanner import scan_file_path
 import torch
+import numpy as np
 import safetensors.torch
-from diffusers import DiffusionPipeline, ConfigMixin
+from pathlib import Path
+from diffusers import DiffusionPipeline, ConfigMixin, OnnxRuntimeModel
 
 from contextlib import suppress
 from pydantic import BaseModel, Field
 from typing import List, Dict, Optional, Type, Literal, TypeVar, Generic, Callable, Any, Union
 
+import onnx
+from onnx import numpy_helper
+from onnx.external_data_helper import set_external_data
+from onnxruntime import InferenceSession, OrtValue, SessionOptions, ExecutionMode, GraphOptimizationLevel
+class InvalidModelException(Exception):
+    pass
+
 class ModelNotFoundException(Exception):
     pass
 
 class BaseModelType(str, Enum):
     StableDiffusion1 = "sd-1"
     StableDiffusion2 = "sd-2"
+    StableDiffusionXL = "sdxl"
+    StableDiffusionXLRefiner = "sdxl-refiner"
     #Kandinsky2_1 = "kandinsky-2.1"
 
 class ModelType(str, Enum):
+    ONNX = "onnx"
     Main = "main"
     Vae = "vae"
     Lora = "lora"
@@ -33,8 +45,12 @@ class ModelType(str, Enum):
 class SubModelType(str, Enum):
     UNet = "unet"
     TextEncoder = "text_encoder"
+    TextEncoder2 = "text_encoder_2"
     Tokenizer = "tokenizer"
+    Tokenizer2 = "tokenizer_2"
     Vae = "vae"
+    VaeDecoder = "vae_decoder"
+    VaeEncoder = "vae_encoder"
     Scheduler = "scheduler"
     SafetyChecker = "safety_checker"
     #MoVQ = "movq"
@@ -56,7 +72,6 @@ class ModelConfigBase(BaseModel):
     path: str # or Path
     description: Optional[str] = Field(None)
     model_format: Optional[str] = Field(None)
-    # do not save to config
     error: Optional[ModelError] = Field(None)
 
     class Config:
@@ -248,16 +263,18 @@ class DiffusersModel(ModelBase):
             try:
                 # TODO: set cache_dir to /dev/null to be sure that cache not used?
                 model = self.child_types[child_type].from_pretrained(
-                    self.model_path,
-                    subfolder=child_type.value,
+                    os.path.join(self.model_path, child_type.value),
+                    #subfolder=child_type.value,
                     torch_dtype=torch_dtype,
                     variant=variant,
                     local_files_only=True,
                 )
                 break
             except Exception as e:
-                #print("====ERR LOAD====")
-                #print(f"{variant}: {e}")
+                print("====ERR LOAD====")
+                print(f"{variant}: {e}")
+                import traceback
+                traceback.print_exc()
                 pass
         else:
             raise Exception(f"Failed to load {self.base_model}:{self.model_type}:{child_type} model")
@@ -424,3 +441,188 @@ class SilenceWarnings(object):
         transformers_logging.set_verbosity(self.transformers_verbosity)
         diffusers_logging.set_verbosity(self.diffusers_verbosity)
         warnings.simplefilter('default')
+
+ONNX_WEIGHTS_NAME = "model.onnx"
+class IAIOnnxRuntimeModel:
+    class _tensor_access:
+
+        def __init__(self, model):
+            self.model = model
+            self.indexes = dict()
+            for idx, obj in enumerate(self.model.proto.graph.initializer):
+                self.indexes[obj.name] = idx
+
+        def __getitem__(self, key: str):
+            return self.model.data[key].numpy()
+
+        def __setitem__(self, key: str, value: np.ndarray):
+            new_node = numpy_helper.from_array(value)
+            # set_external_data(new_node, location="in-memory-location")
+            new_node.name = key
+            # new_node.ClearField("raw_data")
+            del self.model.proto.graph.initializer[self.indexes[key]]
+            self.model.proto.graph.initializer.insert(self.indexes[key], new_node)
+            self.model.data[key] = OrtValue.ortvalue_from_numpy(value)
+
+        # __delitem__
+
+        def __contains__(self, key: str):
+            return key in self.model.data
+
+        def items(self):
+            raise NotImplementedError("tensor.items")
+            #return [(obj.name, obj) for obj in self.raw_proto]
+
+        def keys(self):
+            return self.model.data.keys()
+
+        def values(self):
+            raise NotImplementedError("tensor.values")
+            #return [obj for obj in self.raw_proto]
+
+
+
+    class _access_helper:
+        def __init__(self, raw_proto):
+            self.indexes = dict()
+            self.raw_proto = raw_proto
+            for idx, obj in enumerate(raw_proto):
+                self.indexes[obj.name] = idx
+
+        def __getitem__(self, key: str):
+            return self.raw_proto[self.indexes[key]]
+
+        def __setitem__(self, key: str, value):
+            index = self.indexes[key]
+            del self.raw_proto[index]
+            self.raw_proto.insert(index, value)
+
+        # __delitem__
+
+        def __contains__(self, key: str):
+            return key in self.indexes
+
+        def items(self):
+            return [(obj.name, obj) for obj in self.raw_proto]
+
+        def keys(self):
+            return self.indexes.keys()
+
+        def values(self):
+            return [obj for obj in self.raw_proto]
+    
+    def __init__(self, model_path: str, provider: Optional[str]):
+        self.path = model_path
+        self.session = None
+        self.provider = provider or "CPUExecutionProvider"
+        """
+        self.data_path = self.path + "_data"
+        if not os.path.exists(self.data_path):
+            print(f"Moving model tensors to separate file: {self.data_path}")
+            tmp_proto = onnx.load(model_path, load_external_data=True)
+            onnx.save_model(tmp_proto, self.path, save_as_external_data=True, all_tensors_to_one_file=True, location=os.path.basename(self.data_path), size_threshold=1024, convert_attribute=False)
+            del tmp_proto
+            gc.collect()
+
+        self.proto = onnx.load(model_path, load_external_data=False)
+        """
+
+        self.proto = onnx.load(model_path, load_external_data=True)
+        self.data = dict()
+        for tensor in self.proto.graph.initializer:
+            name = tensor.name
+
+            if tensor.HasField("raw_data"):
+                npt = numpy_helper.to_array(tensor)
+                orv = OrtValue.ortvalue_from_numpy(npt)
+                self.data[name] = orv
+                # set_external_data(tensor, location="in-memory-location")
+                tensor.name = name
+                # tensor.ClearField("raw_data")
+
+        self.nodes = self._access_helper(self.proto.graph.node)
+        self.initializers = self._access_helper(self.proto.graph.initializer)
+        # print(self.proto.graph.input)
+        # print(self.proto.graph.initializer)
+
+        self.tensors = self._tensor_access(self)
+
+    # TODO: integrate with model manager/cache
+    def create_session(self):
+        if self.session is None:
+            #onnx.save(self.proto, "tmp.onnx")
+            #onnx.save_model(self.proto, "tmp.onnx", save_as_external_data=True, all_tensors_to_one_file=True, location="tmp.onnx_data", size_threshold=1024, convert_attribute=False)
+            # TODO: something to be able to get weight when they already moved outside of model proto
+            #(trimmed_model, external_data) = buffer_external_data_tensors(self.proto)
+            sess = SessionOptions()
+            #self._external_data.update(**external_data)
+            # sess.add_external_initializers(list(self.data.keys()), list(self.data.values()))
+            # sess.enable_profiling = True
+
+            # sess.intra_op_num_threads = 1
+            # sess.inter_op_num_threads = 1
+            # sess.execution_mode = ExecutionMode.ORT_SEQUENTIAL
+            # sess.graph_optimization_level = GraphOptimizationLevel.ORT_ENABLE_ALL
+            # sess.enable_cpu_mem_arena = True
+            # sess.enable_mem_pattern = True
+            # sess.add_session_config_entry("session.intra_op.use_xnnpack_threadpool", "1") ########### It's the key code
+
+
+            sess.add_free_dimension_override_by_name("unet_sample_batch", 2)
+            sess.add_free_dimension_override_by_name("unet_sample_channels", 4)
+            sess.add_free_dimension_override_by_name("unet_hidden_batch", 2)
+            sess.add_free_dimension_override_by_name("unet_hidden_sequence", 77)
+            sess.add_free_dimension_override_by_name("unet_sample_height", 64)
+            sess.add_free_dimension_override_by_name("unet_sample_width", 64)
+            sess.add_free_dimension_override_by_name("unet_time_batch", 1)
+            self.session = InferenceSession(self.proto.SerializeToString(), providers=['CUDAExecutionProvider', 'CPUExecutionProvider'], sess_options=sess)
+            #self.session = InferenceSession("tmp.onnx", providers=[self.provider], sess_options=self.sess_options)
+            self.io_binding = self.session.io_binding()
+
+    def release_session(self):
+        self.session = None
+        import gc
+        gc.collect()
+
+    def __call__(self, **kwargs):
+        if self.session is None:
+            raise Exception("You should call create_session before running model")
+
+        inputs = {k: np.array(v) for k, v in kwargs.items()}
+        output_names = self.session.get_outputs()
+        for k in inputs:
+            self.io_binding.bind_cpu_input(k, inputs[k])
+        for name in output_names:
+            self.io_binding.bind_output(name.name)
+        self.session.run_with_iobinding(self.io_binding, None)
+        return self.io_binding.copy_outputs_to_cpu()
+
+    # compatability with diffusers load code
+    @classmethod
+    def from_pretrained(
+        cls,
+        model_id: Union[str, Path],
+        subfolder: Union[str, Path] = None,
+        file_name: Optional[str] = None,
+        provider: Optional[str] = None,
+        sess_options: Optional["SessionOptions"] = None,
+        **kwargs,
+    ):
+        file_name = file_name or ONNX_WEIGHTS_NAME
+
+        if os.path.isdir(model_id):
+            model_path = model_id
+            if subfolder is not None:
+                model_path = os.path.join(model_path, subfolder)
+            model_path = os.path.join(model_path, file_name)
+
+        else:
+            model_path = model_id
+
+        # load model from local directory
+        if not os.path.isfile(model_path):
+            raise Exception(f"Model not found: {model_path}")
+
+        # TODO: session options
+        return cls(model_path, provider=provider)
+

invokeai/backend/model_management/models/controlnet.py

@@ -1,8 +1,7 @@
 import os
 import torch
 from enum import Enum
-from pathlib import Path
-from typing import Optional, Union, Literal
+from typing import Optional
 from .base import (
     ModelBase,
     ModelConfigBase,
@@ -13,6 +12,8 @@ from .base import (
     calc_model_size_by_fs,
     calc_model_size_by_data,
     classproperty,
+    InvalidModelException,
+    ModelNotFoundException,
 )
 
 class ControlNetModelFormat(str, Enum):
@@ -59,10 +60,20 @@ class ControlNetModel(ModelBase):
         if child_type is not None:
             raise Exception("There is no child models in controlnet model")
 
-        model = self.model_class.from_pretrained(
-            self.model_path,
-            torch_dtype=torch_dtype,
-        )
+        model = None
+        for variant in ['fp16',None]:
+            try:
+                model = self.model_class.from_pretrained(
+                    self.model_path,
+                    torch_dtype=torch_dtype,
+                    variant=variant,
+                )
+                break
+            except:
+                pass
+        if not model:
+            raise ModelNotFoundException()
+        
         # calc more accurate size
         self.model_size = calc_model_size_by_data(model)
         return model
@@ -73,10 +84,18 @@ class ControlNetModel(ModelBase):
 
     @classmethod
     def detect_format(cls, path: str):
+        if not os.path.exists(path):
+            raise ModelNotFoundException()
+
         if os.path.isdir(path):
-            return ControlNetModelFormat.Diffusers
-        else:
-            return ControlNetModelFormat.Checkpoint
+            if os.path.exists(os.path.join(path, "config.json")):
+                return ControlNetModelFormat.Diffusers
+
+        if os.path.isfile(path):
+            if any([path.endswith(f".{ext}") for ext in ["safetensors", "ckpt", "pt", "pth"]]):
+                return ControlNetModelFormat.Checkpoint
+
+        raise InvalidModelException(f"Not a valid model: {path}")
 
     @classmethod
     def convert_if_required(

invokeai/backend/model_management/models/lora.py

@@ -9,6 +9,7 @@ from .base import (
     ModelType,
     SubModelType,
     classproperty,
+    InvalidModelException,
 )
 # TODO: naming
 from ..lora import LoRAModel as LoRAModelRaw
@@ -56,10 +57,18 @@ class LoRAModel(ModelBase):
 
     @classmethod
     def detect_format(cls, path: str):
+        if not os.path.exists(path):
+            raise ModelNotFoundException()
+
         if os.path.isdir(path):
-            return LoRAModelFormat.Diffusers
-        else:
-            return LoRAModelFormat.LyCORIS
+            if os.path.exists(os.path.join(path, "pytorch_lora_weights.bin")):
+                return LoRAModelFormat.Diffusers
+
+        if os.path.isfile(path):
+            if any([path.endswith(f".{ext}") for ext in ["safetensors", "ckpt", "pt"]]):
+                return LoRAModelFormat.LyCORIS
+
+        raise InvalidModelException(f"Not a valid model: {path}")
 
     @classmethod
     def convert_if_required(

invokeai/backend/model_management/models/sdxl.py

@@ -0,0 +1,114 @@
+import os
+import json
+from enum import Enum
+from pydantic import Field
+from typing import Literal, Optional
+from .base import (
+    ModelConfigBase,
+    BaseModelType,
+    ModelType,
+    ModelVariantType,
+    DiffusersModel,
+    read_checkpoint_meta,
+    classproperty,
+)
+from omegaconf import OmegaConf
+
+class StableDiffusionXLModelFormat(str, Enum):
+    Checkpoint = "checkpoint"
+    Diffusers = "diffusers"
+    
+class StableDiffusionXLModel(DiffusersModel):
+
+    # TODO: check that configs overwriten properly
+    class DiffusersConfig(ModelConfigBase):
+        model_format: Literal[StableDiffusionXLModelFormat.Diffusers]
+        vae: Optional[str] = Field(None)
+        variant: ModelVariantType
+
+    class CheckpointConfig(ModelConfigBase):
+        model_format: Literal[StableDiffusionXLModelFormat.Checkpoint]
+        vae: Optional[str] = Field(None)
+        config: str
+        variant: ModelVariantType
+
+    def __init__(self, model_path: str, base_model: BaseModelType, model_type: ModelType):
+        assert base_model in {BaseModelType.StableDiffusionXL, BaseModelType.StableDiffusionXLRefiner}
+        assert model_type == ModelType.Main
+        super().__init__(
+            model_path=model_path,
+            base_model=BaseModelType.StableDiffusionXL,
+            model_type=ModelType.Main,
+        )
+
+    @classmethod
+    def probe_config(cls, path: str, **kwargs):
+        model_format = cls.detect_format(path)
+        ckpt_config_path = kwargs.get("config", None)
+        if model_format == StableDiffusionXLModelFormat.Checkpoint:
+            if ckpt_config_path:
+                ckpt_config = OmegaConf.load(ckpt_config_path)
+                ckpt_config["model"]["params"]["unet_config"]["params"]["in_channels"]
+
+            else:
+                checkpoint = read_checkpoint_meta(path)
+                checkpoint = checkpoint.get('state_dict', checkpoint)
+                in_channels = checkpoint["model.diffusion_model.input_blocks.0.0.weight"].shape[1]
+
+        elif model_format == StableDiffusionXLModelFormat.Diffusers:
+            unet_config_path = os.path.join(path, "unet", "config.json")
+            if os.path.exists(unet_config_path):
+                with open(unet_config_path, "r") as f:
+                    unet_config = json.loads(f.read())
+                in_channels = unet_config['in_channels']
+
+            else:
+                raise Exception("Not supported stable diffusion diffusers format(possibly onnx?)")
+
+        else:
+            raise NotImplementedError(f"Unknown stable diffusion 2.* format: {model_format}")
+
+        if in_channels == 9:
+            variant = ModelVariantType.Inpaint
+        elif in_channels == 5:
+            variant = ModelVariantType.Depth
+        elif in_channels == 4:
+            variant = ModelVariantType.Normal
+        else:
+            raise Exception("Unkown stable diffusion 2.* model format")
+
+        if ckpt_config_path is None:
+            # TO DO: implement picking
+            pass
+        
+        return cls.create_config(
+            path=path,
+            model_format=model_format,
+
+            config=ckpt_config_path,
+            variant=variant,
+        )
+
+    @classproperty
+    def save_to_config(cls) -> bool:
+        return True
+
+    @classmethod
+    def detect_format(cls, model_path: str):
+        if os.path.isdir(model_path):
+            return StableDiffusionXLModelFormat.Diffusers
+        else:
+            return StableDiffusionXLModelFormat.Checkpoint
+
+    @classmethod
+    def convert_if_required(
+        cls,
+        model_path: str,
+        output_path: str,
+        config: ModelConfigBase,
+        base_model: BaseModelType,
+    ) -> str:
+        if isinstance(config, cls.CheckpointConfig):
+            raise NotImplementedError('conversion of SDXL checkpoint models to diffusers format is not yet supported')
+        else:
+            return model_path

invokeai/backend/model_management/models/stable_diffusion.py

@@ -5,17 +5,15 @@ from pydantic import Field
 from pathlib import Path
 from typing import Literal, Optional, Union
 from .base import (
-    ModelBase,
     ModelConfigBase,
     BaseModelType,
     ModelType,
-    SubModelType,
     ModelVariantType,
     DiffusersModel,
-    SchedulerPredictionType,
     SilenceWarnings,
     read_checkpoint_meta,
     classproperty,
+    InvalidModelException,
 )
 from invokeai.app.services.config import InvokeAIAppConfig
 from omegaconf import OmegaConf
@@ -36,8 +34,7 @@ class StableDiffusion1Model(DiffusersModel):
         vae: Optional[str] = Field(None)
         config: str
         variant: ModelVariantType
-
-
+        
     def __init__(self, model_path: str, base_model: BaseModelType, model_type: ModelType):
         assert base_model == BaseModelType.StableDiffusion1
         assert model_type == ModelType.Main
@@ -98,10 +95,18 @@ class StableDiffusion1Model(DiffusersModel):
 
     @classmethod
     def detect_format(cls, model_path: str):
+        if not os.path.exists(model_path):
+            raise ModelNotFoundException()
+
         if os.path.isdir(model_path):
-            return StableDiffusion1ModelFormat.Diffusers
-        else:
-            return StableDiffusion1ModelFormat.Checkpoint
+            if os.path.exists(os.path.join(model_path, "model_index.json")):
+                return StableDiffusion1ModelFormat.Diffusers
+
+        if os.path.isfile(model_path):
+            if any([model_path.endswith(f".{ext}") for ext in ["safetensors", "ckpt", "pt"]]):
+                return StableDiffusion1ModelFormat.Checkpoint
+
+        raise InvalidModelException(f"Not a valid model: {model_path}")
 
     @classmethod
     def convert_if_required(
@@ -200,10 +205,18 @@ class StableDiffusion2Model(DiffusersModel):
 
     @classmethod
     def detect_format(cls, model_path: str):
+        if not os.path.exists(model_path):
+            raise ModelNotFoundException()
+
         if os.path.isdir(model_path):
-            return StableDiffusion2ModelFormat.Diffusers
-        else:
-            return StableDiffusion2ModelFormat.Checkpoint
+            if os.path.exists(os.path.join(model_path, "model_index.json")):
+                return StableDiffusion2ModelFormat.Diffusers
+
+        if os.path.isfile(model_path):
+            if any([model_path.endswith(f".{ext}") for ext in ["safetensors", "ckpt", "pt"]]):
+                return StableDiffusion2ModelFormat.Checkpoint
+
+        raise InvalidModelException(f"Not a valid model: {model_path}")
 
     @classmethod
     def convert_if_required(
@@ -232,6 +245,12 @@ def _select_ckpt_config(version: BaseModelType, variant: ModelVariantType):
             ModelVariantType.Normal: "v2-inference-v.yaml", # best guess, as we can't differentiate with base(512)
             ModelVariantType.Inpaint: "v2-inpainting-inference.yaml",
             ModelVariantType.Depth: "v2-midas-inference.yaml",
+        },
+        # note that these .yaml files don't yet exist!
+        BaseModelType.StableDiffusionXL: {
+            ModelVariantType.Normal: "xl-inference-v.yaml",
+            ModelVariantType.Inpaint: "xl-inpainting-inference.yaml",
+            ModelVariantType.Depth: "xl-midas-inference.yaml",
         }
     }
 
@@ -247,6 +266,7 @@ def _select_ckpt_config(version: BaseModelType, variant: ModelVariantType):
 
 
 # TODO: rework
+# Note that convert_ckpt_to_diffuses does not currently support conversion of SDXL models
 def _convert_ckpt_and_cache(
     version: BaseModelType,
     model_config: Union[StableDiffusion1Model.CheckpointConfig, StableDiffusion2Model.CheckpointConfig],

invokeai/backend/model_management/models/stable_diffusion_onnx.py

@@ -0,0 +1,156 @@
+import os
+import json
+from enum import Enum
+from pydantic import Field
+from pathlib import Path
+from typing import Literal, Optional, Union
+from .base import (
+    ModelBase,
+    ModelConfigBase,
+    BaseModelType,
+    ModelType,
+    SubModelType,
+    ModelVariantType,
+    DiffusersModel,
+    SchedulerPredictionType,
+    SilenceWarnings,
+    read_checkpoint_meta,
+    classproperty,
+    OnnxRuntimeModel,
+    IAIOnnxRuntimeModel,
+)
+from invokeai.app.services.config import InvokeAIAppConfig
+
+class ONNXStableDiffusion1Model(DiffusersModel):
+
+    class Config(ModelConfigBase):
+        model_format: None
+        variant: ModelVariantType
+
+
+    def __init__(self, model_path: str, base_model: BaseModelType, model_type: ModelType):
+        assert base_model == BaseModelType.StableDiffusion1
+        assert model_type == ModelType.ONNX
+        super().__init__(
+            model_path=model_path,
+            base_model=BaseModelType.StableDiffusion1,
+            model_type=ModelType.ONNX,
+        )
+
+        for child_name, child_type in self.child_types.items():
+            if child_type is OnnxRuntimeModel:
+                self.child_types[child_name] = IAIOnnxRuntimeModel
+
+            # TODO: check that no optimum models provided
+
+    @classmethod
+    def probe_config(cls, path: str, **kwargs):
+        model_format = cls.detect_format(path)
+        in_channels = 4 # TODO:
+
+        if in_channels == 9:
+            variant = ModelVariantType.Inpaint
+        elif in_channels == 4:
+            variant = ModelVariantType.Normal
+        else:
+            raise Exception("Unkown stable diffusion 1.* model format")
+
+
+        return cls.create_config(
+            path=path,
+            model_format=model_format,
+
+            variant=variant,
+        )
+
+    @classproperty
+    def save_to_config(cls) -> bool:
+        return True
+
+    @classmethod
+    def detect_format(cls, model_path: str):
+        return None
+
+    @classmethod
+    def convert_if_required(
+        cls,
+        model_path: str,
+        output_path: str,
+        config: ModelConfigBase,
+        base_model: BaseModelType,
+    ) -> str:
+        return model_path
+
+class ONNXStableDiffusion2Model(DiffusersModel):
+
+    # TODO: check that configs overwriten properly
+    class Config(ModelConfigBase):
+        model_format: None
+        variant: ModelVariantType
+        prediction_type: SchedulerPredictionType
+        upcast_attention: bool
+
+
+    def __init__(self, model_path: str, base_model: BaseModelType, model_type: ModelType):
+        assert base_model == BaseModelType.StableDiffusion2
+        assert model_type == ModelType.ONNX
+        super().__init__(
+            model_path=model_path,
+            base_model=BaseModelType.StableDiffusion2,
+            model_type=ModelType.ONNX,
+        )
+
+        for child_name, child_type in self.child_types.items():
+            if child_type is OnnxRuntimeModel:
+                self.child_types[child_name] = IAIOnnxRuntimeModel
+            # TODO: check that no optimum models provided
+
+    @classmethod
+    def probe_config(cls, path: str, **kwargs):
+        model_format = cls.detect_format(path)
+        in_channels = 4 # TODO:
+
+        if in_channels == 9:
+            variant = ModelVariantType.Inpaint
+        elif in_channels == 5:
+            variant = ModelVariantType.Depth
+        elif in_channels == 4:
+            variant = ModelVariantType.Normal
+        else:
+            raise Exception("Unkown stable diffusion 2.* model format")
+
+        if variant == ModelVariantType.Normal:
+            prediction_type = SchedulerPredictionType.VPrediction
+            upcast_attention = True
+
+        else:
+            prediction_type = SchedulerPredictionType.Epsilon
+            upcast_attention = False
+
+        return cls.create_config(
+            path=path,
+            model_format=model_format,
+
+            variant=variant,
+            prediction_type=prediction_type,
+            upcast_attention=upcast_attention,
+        )
+
+    @classproperty
+    def save_to_config(cls) -> bool:
+        return True
+
+    @classmethod
+    def detect_format(cls, model_path: str):
+        return None
+
+    @classmethod
+    def convert_if_required(
+        cls,
+        model_path: str,
+        output_path: str,
+        config: ModelConfigBase,
+        base_model: BaseModelType,
+    ) -> str:
+        return model_path
+

invokeai/backend/model_management/models/textual_inversion.py

@@ -9,6 +9,7 @@ from .base import (
     SubModelType,
     classproperty,
     ModelNotFoundException,
+    InvalidModelException,
 )
 # TODO: naming
 from ..lora import TextualInversionModel as TextualInversionModelRaw
@@ -59,7 +60,18 @@ class TextualInversionModel(ModelBase):
 
     @classmethod
     def detect_format(cls, path: str):
-        return None
+        if not os.path.exists(path):
+            raise ModelNotFoundException()
+
+        if os.path.isdir(path):
+            if os.path.exists(os.path.join(path, "learned_embeds.bin")):
+                return None # diffusers-ti
+
+        if os.path.isfile(path):
+            if any([path.endswith(f".{ext}") for ext in ["safetensors", "ckpt", "pt", "bin"]]):
+                return None
+
+        raise InvalidModelException(f"Not a valid model: {path}")
 
     @classmethod
     def convert_if_required(

invokeai/backend/model_management/models/vae.py

@@ -15,6 +15,8 @@ from .base import (
     calc_model_size_by_fs,
     calc_model_size_by_data,
     classproperty,
+    InvalidModelException,
+    ModelNotFoundException,
 )
 from invokeai.app.services.config import InvokeAIAppConfig
 from diffusers.utils import is_safetensors_available
@@ -75,10 +77,18 @@ class VaeModel(ModelBase):
 
     @classmethod
     def detect_format(cls, path: str):
+        if not os.path.exists(path):
+            raise ModelNotFoundException()
+
         if os.path.isdir(path):
-            return VaeModelFormat.Diffusers
-        else:
-            return VaeModelFormat.Checkpoint
+            if os.path.exists(os.path.join(path, "config.json")):
+                return VaeModelFormat.Diffusers
+
+        if os.path.isfile(path):
+            if any([path.endswith(f".{ext}") for ext in ["safetensors", "ckpt", "pt"]]):
+                return VaeModelFormat.Checkpoint
+
+        raise InvalidModelException(f"Not a valid model: {path}")
 
     @classmethod
     def convert_if_required(

invokeai/backend/restoration/__init__.py

@@ -1,4 +0,0 @@
-"""
-Initialization file for the invokeai.backend.restoration package
-"""
-from .base import Restoration

invokeai/backend/restoration/base.py

@@ -1,45 +0,0 @@
-import invokeai.backend.util.logging as logger
-
-class Restoration:
-    def __init__(self) -> None:
-        pass
-
-    def load_face_restore_models(
-        self, gfpgan_model_path="./models/core/face_restoration/gfpgan/GFPGANv1.4.pth"
-    ):
-        # Load GFPGAN
-        gfpgan = self.load_gfpgan(gfpgan_model_path)
-        if gfpgan.gfpgan_model_exists:
-            logger.info("GFPGAN Initialized")
-        else:
-            logger.info("GFPGAN Disabled")
-            gfpgan = None
-
-        # Load CodeFormer
-        codeformer = self.load_codeformer()
-        if codeformer.codeformer_model_exists:
-            logger.info("CodeFormer Initialized")
-        else:
-            logger.info("CodeFormer Disabled")
-            codeformer = None
-
-        return gfpgan, codeformer
-
-    # Face Restore Models
-    def load_gfpgan(self, gfpgan_model_path):
-        from .gfpgan import GFPGAN
-
-        return GFPGAN(gfpgan_model_path)
-
-    def load_codeformer(self):
-        from .codeformer import CodeFormerRestoration
-
-        return CodeFormerRestoration()
-
-    # Upscale Models
-    def load_esrgan(self, esrgan_bg_tile=400):
-        from .realesrgan import ESRGAN
-
-        esrgan = ESRGAN(esrgan_bg_tile)
-        logger.info("ESRGAN Initialized")
-        return esrgan