chore(ui): typgen

to avoid disrupting already-working LoRA

.dockerignore

@@ -1,9 +1,11 @@
 *
 !invokeai
 !pyproject.toml
+!uv.lock
 !docker/docker-entrypoint.sh
 !LICENSE
 
+**/dist
 **/node_modules
 **/__pycache__
-**/*.egg-info
+**/*.egg-info

.git-blame-ignore-revs

@@ -1,2 +1,5 @@
 b3dccfaeb636599c02effc377cdd8a87d658256c
 218b6d0546b990fc449c876fb99f44b50c4daa35
+182580ff6970caed400be178c5b888514b75d7f2
+8e9d5c1187b0d36da80571ce4c8ba9b3a37b6c46
+99aac5870e1092b182e6c5f21abcaab6936a4ad1

.gitattributes

@@ -2,4 +2,5 @@
 # Only affects text files and ignores other file types.
 # For more info see: https://www.aleksandrhovhannisyan.com/blog/crlf-vs-lf-normalizing-line-endings-in-git/
 * text=auto
-docker/** text eol=lf
+docker/** text eol=lf
+tests/test_model_probe/stripped_models/** filter=lfs diff=lfs merge=lfs -text

.github/CODEOWNERS

@@ -1,32 +1,31 @@
 # continuous integration
-/.github/workflows/  @lstein @blessedcoolant @hipsterusername @ebr @jazzhaiku
+/.github/workflows/  @lstein @blessedcoolant @hipsterusername @ebr @jazzhaiku @psychedelicious
 
 # documentation
-/docs/ @lstein @blessedcoolant @hipsterusername @Millu
-/mkdocs.yml @lstein  @blessedcoolant @hipsterusername @Millu
+/docs/ @lstein @blessedcoolant @hipsterusername @psychedelicious
+/mkdocs.yml @lstein @blessedcoolant @hipsterusername @psychedelicious
 
 # nodes
-/invokeai/app/ @Kyle0654 @blessedcoolant @psychedelicious @brandonrising @hipsterusername @jazzhaiku
+/invokeai/app/ @blessedcoolant @psychedelicious @hipsterusername @jazzhaiku
 
 # installation and configuration
-/pyproject.toml  @lstein @blessedcoolant @hipsterusername
-/docker/  @lstein @blessedcoolant @hipsterusername @ebr
-/scripts/ @ebr @lstein @hipsterusername
-/installer/ @lstein @ebr @hipsterusername
-/invokeai/assets @lstein @ebr @hipsterusername
-/invokeai/configs @lstein @hipsterusername
-/invokeai/version @lstein @blessedcoolant @hipsterusername
+/pyproject.toml  @lstein @blessedcoolant @psychedelicious @hipsterusername
+/docker/  @lstein @blessedcoolant @psychedelicious @hipsterusername @ebr
+/scripts/ @ebr @lstein @psychedelicious @hipsterusername
+/installer/ @lstein @ebr @psychedelicious @hipsterusername
+/invokeai/assets @lstein @ebr @psychedelicious @hipsterusername
+/invokeai/configs @lstein @psychedelicious @hipsterusername
+/invokeai/version @lstein @blessedcoolant @psychedelicious @hipsterusername
 
 # web ui
 /invokeai/frontend @blessedcoolant @psychedelicious @lstein @maryhipp @hipsterusername
-/invokeai/backend @blessedcoolant @psychedelicious @lstein @maryhipp @hipsterusername
 
 # generation, model management, postprocessing
-/invokeai/backend  @damian0815 @lstein @blessedcoolant @gregghelt2 @StAlKeR7779 @brandonrising @ryanjdick @hipsterusername @jazzhaiku
+/invokeai/backend  @lstein @blessedcoolant @hipsterusername @jazzhaiku @psychedelicious @maryhipp 
 
 # front ends
-/invokeai/frontend/CLI @lstein @hipsterusername
-/invokeai/frontend/install @lstein @ebr @hipsterusername
-/invokeai/frontend/merge @lstein @blessedcoolant @hipsterusername
-/invokeai/frontend/training @lstein @blessedcoolant @hipsterusername
+/invokeai/frontend/CLI @lstein @psychedelicious @hipsterusername
+/invokeai/frontend/install @lstein @ebr @psychedelicious @hipsterusername
+/invokeai/frontend/merge @lstein @blessedcoolant @psychedelicious @hipsterusername
+/invokeai/frontend/training @lstein @blessedcoolant @psychedelicious @hipsterusername
 /invokeai/frontend/web @psychedelicious @blessedcoolant @maryhipp @hipsterusername

.github/workflows/build-container.yml

@@ -97,6 +97,8 @@ jobs:
           context: .
           file: docker/Dockerfile
           platforms: ${{ env.PLATFORMS }}
+          build-args: |
+            GPU_DRIVER=${{ matrix.gpu-driver }}
           push: ${{ github.ref == 'refs/heads/main' || github.ref_type == 'tag' || github.event.inputs.push-to-registry }}
           tags: ${{ steps.meta.outputs.tags }}
           labels: ${{ steps.meta.outputs.labels }}

.github/workflows/build-wheel.yml

@@ -1,6 +1,6 @@
-# Builds and uploads the installer and python build artifacts.
+# Builds and uploads python build artifacts.
 
-name: build installer
+name: build wheel
 
 on:
   workflow_dispatch:
@@ -17,7 +17,7 @@ jobs:
       - name: setup python
         uses: actions/setup-python@v5
         with:
-          python-version: '3.10'
+          python-version: '3.12'
           cache: pip
           cache-dependency-path: pyproject.toml
 
@@ -27,19 +27,12 @@ jobs:
       - name: setup frontend
         uses: ./.github/actions/install-frontend-deps
 
-      - name: create installer
-        id: create_installer
-        run: ./create_installer.sh
-        working-directory: installer
+      - name: build wheel
+        id: build_wheel
+        run: ./scripts/build_wheel.sh
 
       - name: upload python distribution artifact
         uses: actions/upload-artifact@v4
         with:
           name: dist
-          path: ${{ steps.create_installer.outputs.DIST_PATH }}
-
-      - name: upload installer artifact
-        uses: actions/upload-artifact@v4
-        with:
-          name: installer
-          path: ${{ steps.create_installer.outputs.INSTALLER_PATH }}
+          path: ${{ steps.build_wheel.outputs.DIST_PATH }}

.github/workflows/python-checks.yml

@@ -34,6 +34,9 @@ on:
 
 jobs:
   python-checks:
+    env:
+      # uv requires a venv by default - but for this, we can simply use the system python
+      UV_SYSTEM_PYTHON: 1
     runs-on: ubuntu-latest
     timeout-minutes: 5 # expected run time: <1 min
     steps:
@@ -57,25 +60,23 @@ jobs:
               - '!invokeai/frontend/web/**'
               - 'tests/**'
 
-      - name: setup python
+      - name: setup uv
         if: ${{ steps.changed-files.outputs.python_any_changed == 'true' || inputs.always_run == true }}
-        uses: actions/setup-python@v5
+        uses: astral-sh/setup-uv@v5
         with:
-          python-version: '3.10'
-          cache: pip
-          cache-dependency-path: pyproject.toml
+          version: '0.6.10'
+          enable-cache: true
 
-      - name: install ruff
+      - name: check pypi classifiers
         if: ${{ steps.changed-files.outputs.python_any_changed == 'true' || inputs.always_run == true }}
-        run: pip install ruff==0.9.9
-        shell: bash
+        run: uv run --no-project scripts/check_classifiers.py ./pyproject.toml
 
       - name: ruff check
         if: ${{ steps.changed-files.outputs.python_any_changed == 'true' || inputs.always_run == true }}
-        run: ruff check --output-format=github .
+        run: uv tool run ruff@0.11.2 check --output-format=github .
         shell: bash
 
       - name: ruff format
         if: ${{ steps.changed-files.outputs.python_any_changed == 'true' || inputs.always_run == true }}
-        run: ruff format --check .
+        run: uv tool run ruff@0.11.2 format --check .
         shell: bash

.github/workflows/python-tests.yml

@@ -39,24 +39,15 @@ jobs:
     strategy:
       matrix:
         python-version:
-          - '3.10'
           - '3.11'
+          - '3.12'
         platform:
-          - linux-cuda-11_7
-          - linux-rocm-5_2
           - linux-cpu
           - macos-default
           - windows-cpu
         include:
-          - platform: linux-cuda-11_7
-            os: ubuntu-22.04
-            github-env: $GITHUB_ENV
-          - platform: linux-rocm-5_2
-            os: ubuntu-22.04
-            extra-index-url: 'https://download.pytorch.org/whl/rocm5.2'
-            github-env: $GITHUB_ENV
           - platform: linux-cpu
-            os: ubuntu-22.04
+            os: ubuntu-24.04
             extra-index-url: 'https://download.pytorch.org/whl/cpu'
             github-env: $GITHUB_ENV
           - platform: macos-default
@@ -70,9 +61,12 @@ jobs:
     timeout-minutes: 15 # expected run time: 2-6 min, depending on platform
     env:
       PIP_USE_PEP517: '1'
+      UV_SYSTEM_PYTHON: 1
+
     steps:
       - name: checkout
-        uses: actions/checkout@v4
+        # https://github.com/nschloe/action-cached-lfs-checkout
+        uses: nschloe/action-cached-lfs-checkout@f46300cd8952454b9f0a21a3d133d4bd5684cfc2
 
       - name: check for changed python files
         if: ${{ inputs.always_run != true }}
@@ -91,20 +85,25 @@ jobs:
               - '!invokeai/frontend/web/**'
               - 'tests/**'
 
+      - name: setup uv
+        if: ${{ steps.changed-files.outputs.python_any_changed == 'true' || inputs.always_run == true }}
+        uses: astral-sh/setup-uv@v5
+        with:
+          version: '0.6.10'
+          enable-cache: true
+          python-version: ${{ matrix.python-version }}
+
       - name: setup python
         if: ${{ steps.changed-files.outputs.python_any_changed == 'true' || inputs.always_run == true }}
         uses: actions/setup-python@v5
         with:
           python-version: ${{ matrix.python-version }}
-          cache: pip
-          cache-dependency-path: pyproject.toml
 
       - name: install dependencies
         if: ${{ steps.changed-files.outputs.python_any_changed == 'true' || inputs.always_run == true }}
         env:
-          PIP_EXTRA_INDEX_URL: ${{ matrix.extra-index-url }}
-        run: >
-          pip3 install --editable=".[test]"
+          UV_INDEX: ${{ matrix.extra-index-url }}
+        run: uv pip install --editable ".[test]"
 
       - name: run pytest
         if: ${{ steps.changed-files.outputs.python_any_changed == 'true' || inputs.always_run == true }}

.github/workflows/release.yml

@@ -49,7 +49,7 @@ jobs:
       always_run: true
 
   build:
-    uses: ./.github/workflows/build-installer.yml
+    uses: ./.github/workflows/build-wheel.yml
 
   publish-testpypi:
     runs-on: ubuntu-latest

.github/workflows/typegen-checks.yml

@@ -54,17 +54,25 @@ jobs:
               - 'pyproject.toml'
               - 'invokeai/**'
 
+      - name: setup uv
+        if: ${{ steps.changed-files.outputs.src_any_changed == 'true' || inputs.always_run == true }}
+        uses: astral-sh/setup-uv@v5
+        with:
+          version: '0.6.10'
+          enable-cache: true
+          python-version: '3.11'
+
       - name: setup python
         if: ${{ steps.changed-files.outputs.src_any_changed == 'true' || inputs.always_run == true }}
         uses: actions/setup-python@v5
         with:
-          python-version: '3.10'
-          cache: pip
-          cache-dependency-path: pyproject.toml
+          python-version: '3.11'
 
-      - name: install python dependencies
+      - name: install dependencies
         if: ${{ steps.changed-files.outputs.src_any_changed == 'true' || inputs.always_run == true }}
-        run: pip3 install --use-pep517 --editable="."
+        env:
+          UV_INDEX: ${{ matrix.extra-index-url }}
+        run: uv pip install --editable .
 
       - name: install frontend dependencies
         if: ${{ steps.changed-files.outputs.src_any_changed == 'true' || inputs.always_run == true }}
@@ -77,7 +85,7 @@ jobs:
 
       - name: generate schema
         if: ${{ steps.changed-files.outputs.src_any_changed == 'true' || inputs.always_run == true }}
-        run: make frontend-typegen
+        run: cd invokeai/frontend/web && uv run ../../../scripts/generate_openapi_schema.py | pnpm typegen
         shell: bash
 
       - name: compare files

.github/workflows/uv-lock-checks.yml

@@ -0,0 +1,68 @@
+# Check the `uv` lockfile for consistency with `pyproject.toml`.
+#
+# If this check fails, you should run `uv lock` to update the lockfile.
+
+name: 'uv lock checks'
+
+on:
+  push:
+    branches:
+      - 'main'
+  pull_request:
+    types:
+      - 'ready_for_review'
+      - 'opened'
+      - 'synchronize'
+  merge_group:
+  workflow_dispatch:
+    inputs:
+      always_run:
+        description: 'Always run the checks'
+        required: true
+        type: boolean
+        default: true
+  workflow_call:
+    inputs:
+      always_run:
+        description: 'Always run the checks'
+        required: true
+        type: boolean
+        default: true
+
+jobs:
+  uv-lock-checks:
+    env:
+      # uv requires a venv by default - but for this, we can simply use the system python
+      UV_SYSTEM_PYTHON: 1
+    runs-on: ubuntu-latest
+    timeout-minutes: 5 # expected run time: <1 min
+    steps:
+      - name: checkout
+        uses: actions/checkout@v4
+
+      - name: check for changed python files
+        if: ${{ inputs.always_run != true }}
+        id: changed-files
+        # Pinned to the _hash_ for v45.0.9 to prevent supply-chain attacks.
+        # See:
+        # - CVE-2025-30066
+        # - https://www.stepsecurity.io/blog/harden-runner-detection-tj-actions-changed-files-action-is-compromised
+        # - https://github.com/tj-actions/changed-files/issues/2463
+        uses: tj-actions/changed-files@a284dc1814e3fd07f2e34267fc8f81227ed29fb8
+        with:
+          files_yaml: |
+            uvlock-pyprojecttoml:
+              - 'pyproject.toml'
+              - 'uv.lock'
+
+      - name: setup uv
+        if: ${{ steps.changed-files.outputs.uvlock-pyprojecttoml_any_changed == 'true' || inputs.always_run == true }}
+        uses: astral-sh/setup-uv@v5
+        with:
+          version: '0.6.10'
+          enable-cache: true
+
+      - name: check lockfile
+        if: ${{ steps.changed-files.outputs.uvlock-pyprojecttoml_any_changed == 'true' || inputs.always_run == true }}
+        run: uv lock --locked # this will exit with 1 if the lockfile is not consistent with pyproject.toml
+        shell: bash

.gitignore

@@ -188,3 +188,4 @@ installer/install.sh
 installer/update.bat
 installer/update.sh
 installer/InvokeAI-Installer/
+.aider*

.nvmrc

@@ -1 +1 @@
-v22.12.0
+v22.14.0

.pre-commit-config.yaml

@@ -4,21 +4,29 @@ repos:
   hooks:
   - id: black
     name: black
-    stages: [commit]
+    stages: [pre-commit]
     language: system
     entry: black
     types: [python]
 
   - id: flake8
     name: flake8
-    stages: [commit]
+    stages: [pre-commit]
     language: system
     entry: flake8
     types: [python]
 
   - id: isort
     name: isort
-    stages: [commit]
+    stages: [pre-commit]
     language: system
     entry: isort
-    types: [python]
+    types: [python]
+
+  - id: uvlock
+    name: uv lock
+    stages: [pre-commit]
+    language: system
+    entry: uv lock
+    files: ^pyproject\.toml$
+    pass_filenames: false

Makefile

@@ -16,7 +16,7 @@ help:
 	@echo "frontend-build           Build the frontend in order to run on localhost:9090"
 	@echo "frontend-dev             Run the frontend in developer mode on localhost:5173"
 	@echo "frontend-typegen         Generate types for the frontend from the OpenAPI schema"
-	@echo "installer-zip            Build the installer .zip file for the current version"
+	@echo "wheel            				Build the wheel for the current version"
 	@echo "tag-release              Tag the GitHub repository with the current version (use at release time only!)"
 	@echo "openapi                  Generate the OpenAPI schema for the app, outputting to stdout"
 	@echo "docs                     Serve the mkdocs site with live reload"
@@ -64,13 +64,13 @@ frontend-dev:
 frontend-typegen:
 	cd invokeai/frontend/web && python ../../../scripts/generate_openapi_schema.py | pnpm typegen
 
-# Installer zip file
-installer-zip:
-	cd installer && ./create_installer.sh
+# Tag the release
+wheel:
+	cd scripts && ./build_wheel.sh
 
 # Tag the release
 tag-release:
-	cd installer && ./tag_release.sh
+	cd scripts && ./tag_release.sh
 
 # Generate the OpenAPI Schema for the app
 openapi:

docker/Dockerfile

@@ -1,77 +1,6 @@
 # syntax=docker/dockerfile:1.4
 
-## Builder stage
-
-FROM library/ubuntu:24.04 AS builder
-
-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 \
-        build-essential \
-        git
-
-# Install `uv` for package management
-COPY --from=ghcr.io/astral-sh/uv:0.6.0 /uv /uvx /bin/
-
-ENV VIRTUAL_ENV=/opt/venv
-ENV PATH="$VIRTUAL_ENV/bin:$PATH"
-ENV INVOKEAI_SRC=/opt/invokeai
-ENV PYTHON_VERSION=3.11
-ENV UV_PYTHON=3.11
-ENV UV_COMPILE_BYTECODE=1
-ENV UV_LINK_MODE=copy
-ENV UV_PROJECT_ENVIRONMENT="$VIRTUAL_ENV"
-ENV UV_INDEX="https://download.pytorch.org/whl/cu124"
-
-ARG GPU_DRIVER=cuda
-# unused but available
-ARG BUILDPLATFORM
-
-# Switch to the `ubuntu` user to work around dependency issues with uv-installed python
-RUN mkdir -p ${VIRTUAL_ENV} && \
-    mkdir -p ${INVOKEAI_SRC} && \
-    chmod -R a+w /opt && \
-    mkdir ~ubuntu/.cache && chown ubuntu: ~ubuntu/.cache
-USER ubuntu
-
-# Install python
-RUN --mount=type=cache,target=/home/ubuntu/.cache/uv,uid=1000,gid=1000 \
-  uv python install ${PYTHON_VERSION}
-
-WORKDIR ${INVOKEAI_SRC}
-
-# Install project's dependencies as a separate layer so they aren't rebuilt every commit.
-# bind-mount instead of copy to defer adding sources to the image until next layer.
-#
-# NOTE: there are no pytorch builds for arm64 + cuda, only cpu
-# x86_64/CUDA is the default
-RUN --mount=type=cache,target=/home/ubuntu/.cache/uv,uid=1000,gid=1000 \
-    --mount=type=bind,source=pyproject.toml,target=pyproject.toml \
-    --mount=type=bind,source=invokeai/version,target=invokeai/version \
-    if [ "$TARGETPLATFORM" = "linux/arm64" ] || [ "$GPU_DRIVER" = "cpu" ]; then \
-      UV_INDEX="https://download.pytorch.org/whl/cpu"; \
-    elif [ "$GPU_DRIVER" = "rocm" ]; then \
-      UV_INDEX="https://download.pytorch.org/whl/rocm6.1"; \
-    fi && \
-    uv sync --no-install-project
-
-# Now that the bulk of the dependencies have been installed, copy in the project files that change more frequently.
-COPY invokeai invokeai
-COPY pyproject.toml .
-
-RUN --mount=type=cache,target=/home/ubuntu/.cache/uv,uid=1000,gid=1000 \
-    --mount=type=bind,source=pyproject.toml,target=pyproject.toml \
-    if [ "$TARGETPLATFORM" = "linux/arm64" ] || [ "$GPU_DRIVER" = "cpu" ]; then \
-      UV_INDEX="https://download.pytorch.org/whl/cpu"; \
-    elif [ "$GPU_DRIVER" = "rocm" ]; then \
-      UV_INDEX="https://download.pytorch.org/whl/rocm6.1"; \
-    fi && \
-    uv sync
-
-
-#### Build the Web UI ------------------------------------
+#### Web UI ------------------------------------
 
 FROM docker.io/node:22-slim AS web-builder
 ENV PNPM_HOME="/pnpm"
@@ -85,69 +14,100 @@ RUN --mount=type=cache,target=/pnpm/store \
     pnpm install --frozen-lockfile
 RUN npx vite build
 
-#### Runtime stage ---------------------------------------
+## Backend ---------------------------------------
 
-FROM library/ubuntu:24.04 AS runtime
+FROM library/ubuntu:24.04
 
 ARG DEBIAN_FRONTEND=noninteractive
-ENV PYTHONUNBUFFERED=1
-ENV PYTHONDONTWRITEBYTECODE=1
+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 \
+    --mount=type=cache,target=/var/lib/apt \
+    apt update && apt install -y --no-install-recommends \
+    ca-certificates \
+    git \
+    gosu \
+    libglib2.0-0 \
+    libgl1 \
+    libglx-mesa0 \
+    build-essential \
+    libopencv-dev \
+    libstdc++-10-dev
 
-RUN apt update && apt install -y --no-install-recommends \
-        git \
-        curl \
-        vim \
-        tmux \
-        ncdu \
-        iotop \
-        bzip2 \
-        gosu \
-        magic-wormhole \
-        libglib2.0-0 \
-        libgl1 \
-        libglx-mesa0 \
-        build-essential \
-        libopencv-dev \
-        libstdc++-10-dev &&\
-    apt-get clean && apt-get autoclean
+ENV \
+    PYTHONUNBUFFERED=1 \
+    PYTHONDONTWRITEBYTECODE=1 \
+    VIRTUAL_ENV=/opt/venv \
+    INVOKEAI_SRC=/opt/invokeai \
+    PYTHON_VERSION=3.12 \
+    UV_PYTHON=3.12 \
+    UV_COMPILE_BYTECODE=1 \
+    UV_MANAGED_PYTHON=1 \
+    UV_LINK_MODE=copy \
+    UV_PROJECT_ENVIRONMENT=/opt/venv \
+    UV_INDEX="https://download.pytorch.org/whl/cu124" \
+    INVOKEAI_ROOT=/invokeai \
+    INVOKEAI_HOST=0.0.0.0 \
+    INVOKEAI_PORT=9090 \
+    PATH="/opt/venv/bin:$PATH" \
+    CONTAINER_UID=${CONTAINER_UID:-1000} \
+    CONTAINER_GID=${CONTAINER_GID:-1000}
 
-ENV INVOKEAI_SRC=/opt/invokeai
-ENV VIRTUAL_ENV=/opt/venv
-ENV UV_PROJECT_ENVIRONMENT="$VIRTUAL_ENV"
-ENV PYTHON_VERSION=3.11
-ENV INVOKEAI_ROOT=/invokeai
-ENV INVOKEAI_HOST=0.0.0.0
-ENV INVOKEAI_PORT=9090
-ENV PATH="$VIRTUAL_ENV/bin:$INVOKEAI_SRC:$PATH"
-ENV CONTAINER_UID=${CONTAINER_UID:-1000}
-ENV CONTAINER_GID=${CONTAINER_GID:-1000}
+ARG GPU_DRIVER=cuda
 
 # Install `uv` for package management
-# and install python for the ubuntu user (expected to exist on ubuntu >=24.x)
-# this is too tiny to optimize with multi-stage builds, but maybe we'll come back to it
-COPY --from=ghcr.io/astral-sh/uv:0.6.0 /uv /uvx /bin/
-USER ubuntu
-RUN uv python install ${PYTHON_VERSION}
-USER root
+COPY --from=ghcr.io/astral-sh/uv:0.6.9 /uv /uvx /bin/
 
-# --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"
+# Install python & allow non-root user to use it by traversing the /root dir without read permissions
+RUN --mount=type=cache,target=/root/.cache/uv \
+    uv python install ${PYTHON_VERSION} && \
+    # chmod --recursive a+rX /root/.local/share/uv/python
+    chmod 711 /root
 
 WORKDIR ${INVOKEAI_SRC}
 
+# Install project's dependencies as a separate layer so they aren't rebuilt every commit.
+# bind-mount instead of copy to defer adding sources to the image until next layer.
+#
+# NOTE: there are no pytorch builds for arm64 + cuda, only cpu
+# x86_64/CUDA is the default
+RUN --mount=type=cache,target=/root/.cache/uv \
+    --mount=type=bind,source=pyproject.toml,target=pyproject.toml \
+    --mount=type=bind,source=uv.lock,target=uv.lock \
+    # this is just to get the package manager to recognize that the project exists, without making changes to the docker layer
+    --mount=type=bind,source=invokeai/version,target=invokeai/version \
+    if [ "$TARGETPLATFORM" = "linux/arm64" ] || [ "$GPU_DRIVER" = "cpu" ]; then UV_INDEX="https://download.pytorch.org/whl/cpu"; \
+    elif [ "$GPU_DRIVER" = "rocm" ]; then UV_INDEX="https://download.pytorch.org/whl/rocm6.2"; \
+    fi && \
+    uv sync --frozen
+
 # build patchmatch
 RUN cd /usr/lib/$(uname -p)-linux-gnu/pkgconfig/ && ln -sf opencv4.pc opencv.pc
 RUN python -c "from patchmatch import patch_match"
 
+# 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"
+
 RUN mkdir -p ${INVOKEAI_ROOT} && chown -R ${CONTAINER_UID}:${CONTAINER_GID} ${INVOKEAI_ROOT}
 
 COPY docker/docker-entrypoint.sh ./
 ENTRYPOINT ["/opt/invokeai/docker-entrypoint.sh"]
 CMD ["invokeai-web"]
+
+# --link requires buldkit w/ dockerfile syntax 1.4, does not work with podman
+COPY --link --from=web-builder /build/dist ${INVOKEAI_SRC}/invokeai/frontend/web/dist
+
+# add sources last to minimize image changes on code changes
+COPY invokeai ${INVOKEAI_SRC}/invokeai
+
+# this should not increase image size because we've already installed dependencies
+# in a previous layer
+RUN --mount=type=cache,target=/root/.cache/uv \
+    --mount=type=bind,source=pyproject.toml,target=pyproject.toml \
+    --mount=type=bind,source=uv.lock,target=uv.lock \
+    if [ "$TARGETPLATFORM" = "linux/arm64" ] || [ "$GPU_DRIVER" = "cpu" ]; then UV_INDEX="https://download.pytorch.org/whl/cpu"; \
+    elif [ "$GPU_DRIVER" = "rocm" ]; then UV_INDEX="https://download.pytorch.org/whl/rocm6.2"; \
+    fi && \
+    uv pip install -e .
+

docs/RELEASE.md

@@ -60,16 +60,11 @@ Next, these jobs run and must pass. They are the same jobs that are run for ever
 - **`frontend-checks`**: runs `prettier` (format), `eslint` (lint), `dpdm` (circular refs), `tsc` (static type check) and `knip` (unused imports)
 - **`typegen-checks`**: ensures the frontend and backend types are synced
 
-#### `build-installer` Job
+#### `build-wheel` Job
 
-This sets up both python and frontend dependencies and builds the python package. Internally, this runs `installer/create_installer.sh` and uploads two artifacts:
+This sets up both python and frontend dependencies and builds the python package. Internally, this runs `./scripts/build_wheel.sh` and uploads `dist.zip`, which contains the wheel and unarchived build.
 
-- **`dist`**: the python distribution, to be published on PyPI
-- **`InvokeAI-installer-${VERSION}.zip`**: the legacy install scripts
-
-You don't need to download either of these files.
-
-> The legacy install scripts are no longer used, but we haven't updated the workflow to skip building them.
+You don't need to download or test these artifacts.
 
 #### Sanity Check & Smoke Test
 
@@ -79,7 +74,7 @@ It's possible to test the python package before it gets published to PyPI. We've
 
 But, if you want to be extra-super careful, here's how to test it:
 
-- Download the `dist.zip` build artifact from the `build-installer` job
+- Download the `dist.zip` build artifact from the `build-wheel` job
 - Unzip it and find the wheel file
 - Create a fresh Invoke install by following the [manual install guide](https://invoke-ai.github.io/InvokeAI/installation/manual/) - but instead of installing from PyPI, install from the wheel
 - Test the app

docs/contributing/INVOCATIONS.md

@@ -39,7 +39,7 @@ nodes imported in the `__init__.py` file are loaded. See the README in the nodes
 folder for more examples:
 
 ```py
-from .cool_node import CoolInvocation
+from .cool_node import ResizeInvocation
 ```
 
 ## Creating A New Invocation
@@ -69,7 +69,10 @@ The first set of things we need to do when creating a new Invocation are -
 So let us do that.
 
 ```python
-from invokeai.app.invocations.baseinvocation import BaseInvocation, invocation
+from invokeai.invocation_api import (
+    BaseInvocation,
+    invocation,
+)
 
 @invocation('resize')
 class ResizeInvocation(BaseInvocation):
@@ -103,8 +106,12 @@ create your own custom field types later in this guide. For now, let's go ahead
 and use it.
 
 ```python
-from invokeai.app.invocations.baseinvocation import BaseInvocation, InputField, invocation
-from invokeai.app.invocations.primitives import ImageField
+from invokeai.invocation_api import (
+    BaseInvocation,
+    ImageField,
+    InputField,
+    invocation,
+)
 
 @invocation('resize')
 class ResizeInvocation(BaseInvocation):
@@ -128,8 +135,12 @@ image: ImageField = InputField(description="The input image")
 Great. Now let us create our other inputs for `width` and `height`
 
 ```python
-from invokeai.app.invocations.baseinvocation import BaseInvocation, InputField, invocation
-from invokeai.app.invocations.primitives import ImageField
+from invokeai.invocation_api import (
+    BaseInvocation,
+    ImageField,
+    InputField,
+    invocation,
+)
 
 @invocation('resize')
 class ResizeInvocation(BaseInvocation):
@@ -163,8 +174,13 @@ that are provided by it by InvokeAI.
 Let us create this function first.
 
 ```python
-from invokeai.app.invocations.baseinvocation import BaseInvocation, InputField, invocation, InvocationContext
-from invokeai.app.invocations.primitives import ImageField
+from invokeai.invocation_api import (
+    BaseInvocation,
+    ImageField,
+    InputField,
+    InvocationContext,
+    invocation,
+)
 
 @invocation('resize')
 class ResizeInvocation(BaseInvocation):
@@ -191,8 +207,14 @@ 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 invokeai.app.invocations.baseinvocation import BaseInvocation, InputField, invocation, InvocationContext
-from invokeai.app.invocations.primitives import ImageField
+from invokeai.invocation_api import (
+    BaseInvocation,
+    ImageField,
+    InputField,
+    InvocationContext,
+    invocation,
+)
+
 from invokeai.app.invocations.image import ImageOutput
 
 @invocation('resize')
@@ -217,9 +239,15 @@ Perfect. Now that we have our Invocation setup, let us do what we want to do.
 So let's do that.
 
 ```python
-from invokeai.app.invocations.baseinvocation import BaseInvocation, InputField, invocation, InvocationContext
-from invokeai.app.invocations.primitives import ImageField
-from invokeai.app.invocations.image import ImageOutput, ResourceOrigin, ImageCategory
+from invokeai.invocation_api import (
+    BaseInvocation,
+    ImageField,
+    InputField,
+    InvocationContext,
+    invocation,
+)
+
+from invokeai.app.invocations.image import ImageOutput
 
 @invocation("resize")
 class ResizeInvocation(BaseInvocation):

docs/contributing/dev-environment.md

@@ -18,9 +18,19 @@ If you just want to use Invoke, you should use the [launcher][launcher link].
 
 2. [Fork and clone][forking link] the [InvokeAI repo][repo link].
 
-3. Create an directory for user data (images, models, db, etc). This is typically at `~/invokeai`, but if you already have a non-dev install, you may want to create a separate directory for the dev install.
+3. This repository uses Git LFS to manage large files. To ensure all assets are downloaded:
+      - Install git-lfs → [Download here](https://git-lfs.com/)
+      - Enable automatic LFS fetching for this repository:
+        ```shell
+        git config lfs.fetchinclude "*"
+        ```
+        - Fetch files from LFS (only needs to be done once; subsequent `git pull` will fetch changes automatically):
+        ```
+        git lfs pull
+        ```
+4. Create an directory for user data (images, models, db, etc). This is typically at `~/invokeai`, but if you already have a non-dev install, you may want to create a separate directory for the dev install.
 
-4. Follow the [manual install][manual install link] guide, with some modifications to the install command:
+5. Follow the [manual install][manual install link] guide, with some modifications to the install command:
 
       - Use `.` instead of `invokeai` to install from the current directory. You don't need to specify the version.
 
@@ -31,22 +41,22 @@ If you just want to use Invoke, you should use the [launcher][launcher link].
      With the modifications made, the install command should look something like this:
 
       ```sh
-      uv pip install -e ".[dev,test,docs,xformers]" --python 3.11 --python-preference only-managed --index=https://download.pytorch.org/whl/cu124 --reinstall
+      uv pip install -e ".[dev,test,docs,xformers]" --python 3.12 --python-preference only-managed --index=https://download.pytorch.org/whl/cu126 --reinstall
       ```
 
-5. At this point, you should have Invoke installed, a venv set up and activated, and the server running. But you will see a warning in the terminal that no UI was found. If you go to the URL for the server, you won't get a UI.
+6. At this point, you should have Invoke installed, a venv set up and activated, and the server running. But you will see a warning in the terminal that no UI was found. If you go to the URL for the server, you won't get a UI.
 
       This is because the UI build is not distributed with the source code. You need to build it manually. End the running server instance.
 
       If you only want to edit the docs, you can stop here and skip to the **Documentation** section below.
 
-6. Install the frontend dev toolchain:
+7. Install the frontend dev toolchain:
 
       - [`nodejs`](https://nodejs.org/) (v20+)
 
       - [`pnpm`](https://pnpm.io/8.x/installation) (must be v8 - not v9!)
 
-7. Do a production build of the frontend:
+8. Do a production build of the frontend:
 
       ```sh
       cd <PATH_TO_INVOKEAI_REPO>/invokeai/frontend/web
@@ -54,7 +64,7 @@ If you just want to use Invoke, you should use the [launcher][launcher link].
       pnpm build
       ```
 
-8. Restart the server and navigate to the URL. You should get a UI. After making changes to the python code, restart the server to see those changes.
+9. Restart the server and navigate to the URL. You should get a UI. After making changes to the python code, restart the server to see those changes.
 
 ## Updating the UI
 

docs/installation/legacy_scripts.md

@@ -1,121 +0,0 @@
-# Legacy Scripts
-
-!!! warning "Legacy Scripts"
-
-    We recommend using the Invoke Launcher to install and update Invoke. It's a desktop application for Windows, macOS and Linux. It takes care of a lot of nitty gritty details for you.
-
-    Follow the [quick start guide](./quick_start.md) to get started.
-
-!!! tip "Use the installer to update"
-
-    Using the installer for updates will not erase any of your data (images, models, boards, etc). It only updates the core libraries used to run Invoke.
-
-    Simply use the same path you installed to originally to update your existing installation.
-
-Both release and pre-release versions can be installed using the installer. It also supports install through a wheel if needed.
-
-Be sure to review the [installation requirements] and ensure your system has everything it needs to install Invoke.
-
-## Getting the Latest Installer
-
-Download the `InvokeAI-installer-vX.Y.Z.zip` file from the [latest release] page. It is at the bottom of the page, under **Assets**.
-
-After unzipping the installer, you should have a `InvokeAI-Installer` folder with some files inside, including `install.bat` and `install.sh`.
-
-## Running the Installer
-
-!!! tip
-
-    Windows users should first double-click the `WinLongPathsEnabled.reg` file to prevent a failed installation due to long file paths.
-
-Double-click the install script:
-
-=== "Windows"
-
-    ```sh
-    install.bat
-    ```
-
-=== "Linux/macOS"
-
-    ```sh
-    install.sh
-    ```
-
-!!! info "Running the Installer from the commandline"
-
-    You can also run the install script from cmd/powershell (Windows) or terminal (Linux/macOS).
-
-!!! warning "Untrusted Publisher (Windows)"
-
-    You may get a popup saying the file comes from an `Untrusted Publisher`. Click `More Info` and `Run Anyway` to get past this.
-
-The installation process is simple, with a few prompts:
-
-- Select the version to install. Unless you have a specific reason to install a specific version, select the default (the latest version).
-- Select location for the install. Be sure you have enough space in this folder for the base application, as described in the [installation requirements].
-- Select a GPU device.
-
-!!! info "Slow Installation"
-
-    The installer needs to download several GB of data and install it all. It may appear to get stuck at 99.9% when installing `pytorch` or during a step labeled "Installing collected packages".
-
-    If it is stuck for over 10 minutes, something has probably gone wrong and you should close the window and restart.
-
-## Running the Application
-
-Find the install location you selected earlier. Double-click the launcher script to run the app:
-
-=== "Windows"
-
-    ```sh
-    invoke.bat
-    ```
-
-=== "Linux/macOS"
-
-    ```sh
-    invoke.sh
-    ```
-
-Choose the first option to run the UI. After a series of startup messages, you'll see something like this:
-
-```sh
-Uvicorn running on http://127.0.0.1:9090 (Press CTRL+C to quit)
-```
-
-Copy the URL into your browser and you should see the UI.
-
-## Improved Outpainting with PatchMatch
-
-PatchMatch is an extra add-on that can improve outpainting. Windows users are in luck - it works out of the box.
-
-On macOS and Linux, a few extra steps are needed to set it up. See the [PatchMatch installation guide](./patchmatch.md).
-
-## First-time Setup
-
-You will need to [install some models] before you can generate.
-
-Check the [configuration docs] for details on configuring the application.
-
-## Updating
-
-Updating is exactly the same as installing - download the latest installer, choose the latest version, enter your existing installation path, and the app will update. None of your data (images, models, boards, etc) will be erased.
-
-!!! info "Dependency Resolution Issues"
-
-    We've found that pip's dependency resolution can cause issues when upgrading packages. One very common problem was pip "downgrading" torch from CUDA to CPU, but things broke in other novel ways.
-
-    The installer doesn't have this kind of problem, so we use it for updating as well.
-
-## Installation Issues
-
-If you have installation issues, please review the [FAQ]. You can also [create an issue] or ask for help on [discord].
-
-[installation requirements]: ./requirements.md
-[FAQ]: ../faq.md
-[install some models]: ./models.md
-[configuration docs]: ../configuration.md
-[latest release]: https://github.com/invoke-ai/InvokeAI/releases/latest
-[create an issue]: https://github.com/invoke-ai/InvokeAI/issues
-[discord]: https://discord.gg/ZmtBAhwWhy

docs/installation/manual.md

@@ -43,10 +43,10 @@ The following commands vary depending on the version of Invoke being installed a
 3. Create a virtual environment in that directory:
 
     ```sh
-    uv venv --relocatable --prompt invoke --python 3.11 --python-preference only-managed .venv
+    uv venv --relocatable --prompt invoke --python 3.12 --python-preference only-managed .venv
     ```
 
-    This command creates a portable virtual environment at `.venv` complete with a portable python 3.11. It doesn't matter if your system has no python installed, or has a different version - `uv` will handle everything.
+    This command creates a portable virtual environment at `.venv` complete with a portable python 3.12. It doesn't matter if your system has no python installed, or has a different version - `uv` will handle everything.
 
 4. Activate the virtual environment:
 
@@ -64,14 +64,28 @@ The following commands vary depending on the version of Invoke being installed a
 
 5. Choose a version to install. Review the [GitHub releases page](https://github.com/invoke-ai/InvokeAI/releases).
 
-6. Determine the package package specifier to use when installing. This is a performance optimization.
+6. Determine the package specifier to use when installing. This is a performance optimization.
 
     - If you have an Nvidia 20xx series GPU or older, use `invokeai[xformers]`.
     - If you have an Nvidia 30xx series GPU or newer, or do not have an Nvidia GPU, use `invokeai`.
 
 7. Determine the `PyPI` index URL to use for installation, if any. This is necessary to get the right version of torch installed.
 
-    === "Invoke v5 or later"
+    === "Invoke v5.12 and later"
+
+        - If you are on Windows or Linux with an Nvidia GPU, use `https://download.pytorch.org/whl/cu128`.
+        - If you are on Linux with no GPU, use `https://download.pytorch.org/whl/cpu`.
+        - If you are on Linux with an AMD GPU, use `https://download.pytorch.org/whl/rocm6.2.4`.
+        - **In all other cases, do not use an index.**
+
+    === "Invoke v5.10.0 to v5.11.0"
+
+        - If you are on Windows or Linux with an Nvidia GPU, use `https://download.pytorch.org/whl/cu126`.
+        - If you are on Linux with no GPU, use `https://download.pytorch.org/whl/cpu`.
+        - If you are on Linux with an AMD GPU, use `https://download.pytorch.org/whl/rocm6.2.4`.
+        - **In all other cases, do not use an index.**
+
+    === "Invoke v5.0.0 to v5.9.1"
 
         - If you are on Windows with an Nvidia GPU, use `https://download.pytorch.org/whl/cu124`.
         - If you are on Linux with no GPU, use `https://download.pytorch.org/whl/cpu`.
@@ -88,13 +102,13 @@ The following commands vary depending on the version of Invoke being installed a
 8. Install the `invokeai` package. Substitute the package specifier and version.
 
     ```sh
-    uv pip install <PACKAGE_SPECIFIER>==<VERSION> --python 3.11 --python-preference only-managed --force-reinstall
+    uv pip install <PACKAGE_SPECIFIER>==<VERSION> --python 3.12 --python-preference only-managed --force-reinstall
     ```
 
     If you determined you needed to use a `PyPI` index URL in the previous step, you'll need to add `--index=<INDEX_URL>` like this:
 
     ```sh
-    uv pip install <PACKAGE_SPECIFIER>==<VERSION> --python 3.11 --python-preference only-managed --index=<INDEX_URL> --force-reinstall
+    uv pip install <PACKAGE_SPECIFIER>==<VERSION> --python 3.12 --python-preference only-managed --index=<INDEX_URL> --force-reinstall
     ```
 
 9. Deactivate and reactivate your venv so that the invokeai-specific commands become available in the environment:

docs/installation/quick_start.md

@@ -49,9 +49,9 @@ If you have an existing Invoke installation, you can select it and let the launc
 
 !!! warning "Problem running the launcher on macOS"
 
-    macOS may not allow you to run the launcher. We are working to resolve this by signing the launcher executable. Until that is done, you can either use the [legacy scripts](./legacy_scripts.md) to install, or manually flag the launcher as safe:
+    macOS may not allow you to run the launcher. We are working to resolve this by signing the launcher executable. Until that is done, you can manually flag the launcher as safe:
 
-    - Open the **Invoke-Installer-mac-arm64.dmg** file.
+    - Open the **Invoke Community Edition.dmg** file.
     - Drag the launcher to **Applications**.
     - Open a terminal.
     - Run `xattr -d 'com.apple.quarantine' /Applications/Invoke\ Community\ Edition.app`.
@@ -117,7 +117,6 @@ If you still have problems, ask for help on the Invoke [discord](https://discord
 
 - You can install the Invoke application as a python package. See our [manual install](./manual.md) docs.
 - You can run Invoke with docker. See our [docker install](./docker.md) docs.
-- You can still use our legacy scripts to install and run Invoke. See the [legacy scripts](./legacy_scripts.md) docs.
 
 ## Need Help?
 

docs/installation/requirements.md

@@ -41,7 +41,7 @@ The requirements below are rough guidelines for best performance. GPUs with less
 
     You don't need to do this if you are installing with the [Invoke Launcher](./quick_start.md).
 
-Invoke requires python 3.10 or 3.11. If you don't already have one of these versions installed, we suggest installing 3.11, as it will be supported for longer.
+Invoke requires python 3.10 through 3.12. If you don't already have one of these versions installed, we suggest installing 3.12, as it will be supported for longer.
 
 Check that your system has an up-to-date Python installed by running `python3 --version` in the terminal (Linux, macOS) or cmd/powershell (Windows).
 
@@ -49,19 +49,19 @@ Check that your system has an up-to-date Python installed by running `python3 --
 
     === "Windows"
 
-        - Install python 3.11 with [an official installer].
+        - Install python with [an official installer].
         - The installer includes an option to add python to your PATH. Be sure to enable this. If you missed it, re-run the installer, choose to modify an existing installation, and tick that checkbox.
         - You may need to install [Microsoft Visual C++ Redistributable].
 
     === "macOS"
 
-        - Install python 3.11 with [an official installer].
+        - Install python with [an official installer].
         - If model installs fail with a certificate error, you may need to run this command (changing the python version to match what you have installed): `/Applications/Python\ 3.10/Install\ Certificates.command`
         - If you haven't already, you will need to install the XCode CLI Tools by running `xcode-select --install` in a terminal.
 
     === "Linux"
 
-        - Installing python varies depending on your system. On Ubuntu, you can use the [deadsnakes PPA](https://launchpad.net/~deadsnakes/+archive/ubuntu/ppa).
+        - Installing python varies depending on your system. We recommend [using `uv` to manage your python installation](https://docs.astral.sh/uv/concepts/python-versions/#installing-a-python-version).
         - You'll need to install `libglib2.0-0` and `libgl1-mesa-glx` for OpenCV to work. For example, on a Debian system: `sudo apt update && sudo apt install -y libglib2.0-0 libgl1-mesa-glx`
 
 ## Drivers

docs/nodes/communityNodes.md

@@ -13,6 +13,7 @@ If you'd prefer, you can also just download the whole node folder from the linke
 To use a community workflow, download the `.json` node graph file and load it into Invoke AI via the **Load Workflow** button in the Workflow Editor. 
 
 - Community Nodes
+    + [Anamorphic Tools](#anamorphic-tools)
     + [Adapters-Linked](#adapters-linked-nodes)
     + [Autostereogram](#autostereogram-nodes)
     + [Average Images](#average-images)
@@ -20,9 +21,12 @@ To use a community workflow, download the `.json` node graph file and load it in
     + [Close Color Mask](#close-color-mask) 
     + [Clothing Mask](#clothing-mask)
     + [Contrast Limited Adaptive Histogram Equalization](#contrast-limited-adaptive-histogram-equalization)
+    + [Curves](#curves)
     + [Depth Map from Wavefront OBJ](#depth-map-from-wavefront-obj)
     + [Enhance Detail](#enhance-detail)
     + [Film Grain](#film-grain)
+    + [Flip Pose](#flip-pose)
+    + [Flux Ideal Size](#flux-ideal-size)
     + [Generative Grammar-Based Prompt Nodes](#generative-grammar-based-prompt-nodes)
     + [GPT2RandomPromptMaker](#gpt2randompromptmaker)
     + [Grid to Gif](#grid-to-gif)
@@ -61,6 +65,13 @@ To use a community workflow, download the `.json` node graph file and load it in
 - [Help](#help)
 
 
+--------------------------------
+### Anamorphic Tools
+
+**Description:** A set of nodes to perform anamorphic modifications to images, like lens blur, streaks, spherical distortion, and vignetting.
+
+**Node Link:** https://github.com/JPPhoto/anamorphic-tools
+
 --------------------------------
 ### Adapters Linked Nodes
 
@@ -132,6 +143,13 @@ Node Link: https://github.com/VeyDlin/clahe-node
 View:
 </br><img src="https://raw.githubusercontent.com/VeyDlin/clahe-node/master/.readme/node.png" width="500" />
 
+--------------------------------
+### Curves
+
+**Description:** Adjust an image's curve based on a user-defined string.
+
+**Node Link:** https://github.com/JPPhoto/curves-node
+
 --------------------------------
 ### Depth Map from Wavefront OBJ
 
@@ -162,6 +180,20 @@ To be imported, an .obj must use triangulated meshes, so make sure to enable tha
 
 **Node Link:** https://github.com/JPPhoto/film-grain-node
 
+--------------------------------
+### Flip Pose
+
+**Description:** This node will flip an openpose image horizontally, recoloring it to make sure that it isn't facing the wrong direction. Note that it does not work with openpose hands.
+
+**Node Link:** https://github.com/JPPhoto/flip-pose-node
+
+--------------------------------
+### Flux Ideal Size
+
+**Description:** This node returns an ideal size to use for the first stage of a Flux image generation pipeline. Generating at the right size helps limit duplication and odd subject placement.
+
+**Node Link:** https://github.com/JPPhoto/flux-ideal-size
+
 --------------------------------
 ### Generative Grammar-Based Prompt Nodes
 

installer/install.bat.in

@@ -1,128 +0,0 @@
-@echo off
-setlocal EnableExtensions EnableDelayedExpansion
-
-@rem This script requires the user to install Python 3.10 or higher. All other
-@rem requirements are downloaded as needed.
-
-@rem change to the script's directory
-PUSHD "%~dp0"
-
-set "no_cache_dir=--no-cache-dir"
-if "%1" == "use-cache" (
-    set "no_cache_dir="
-)
-
-@rem Config
-@rem The version in the next line is replaced by an up to date release number
-@rem when create_installer.sh is run. Change the release number there.
-set INSTRUCTIONS=https://invoke-ai.github.io/InvokeAI/installation/INSTALL_AUTOMATED/
-set TROUBLESHOOTING=https://invoke-ai.github.io/InvokeAI/help/FAQ/
-set PYTHON_URL=https://www.python.org/downloads/windows/
-set MINIMUM_PYTHON_VERSION=3.10.0
-set PYTHON_URL=https://www.python.org/downloads/release/python-3109/
-
-set err_msg=An error has occurred and the script could not continue.
-
-@rem --------------------------- Intro -------------------------------
-echo This script will install InvokeAI and its dependencies.
-echo.
-echo BEFORE YOU START PLEASE MAKE SURE TO DO THE FOLLOWING
-echo 1. Install python 3.10 or 3.11. Python version 3.9 is no longer supported.
-echo 2. Double-click on the file WinLongPathsEnabled.reg in order to
-echo    enable long path support on your system.
-echo 3. Install the Visual C++ core libraries.
-echo    Please download and install the libraries from:
-echo    https://learn.microsoft.com/en-US/cpp/windows/latest-supported-vc-redist?view=msvc-170
-echo.
-echo See %INSTRUCTIONS% for more details.
-echo.
-echo FOR THE BEST USER EXPERIENCE WE SUGGEST MAXIMIZING THIS WINDOW NOW.
-pause
-
-@rem ---------------------------- check Python version ---------------
-echo ***** Checking and Updating Python *****
-
-call python --version >.tmp1 2>.tmp2
-if %errorlevel% == 1 (
-   set err_msg=Please install Python 3.10-11. See %INSTRUCTIONS% for details.
-   goto err_exit
-)
-
-for /f "tokens=2" %%i in (.tmp1) do set python_version=%%i
-if "%python_version%" == "" (
-   set err_msg=No python was detected on your system. Please install Python version %MINIMUM_PYTHON_VERSION% or higher. We recommend Python 3.10.12 from %PYTHON_URL%
-   goto err_exit
-)
-
-call :compareVersions %MINIMUM_PYTHON_VERSION% %python_version%
-if %errorlevel% == 1 (
-   set err_msg=Your version of Python is too low. You need at least %MINIMUM_PYTHON_VERSION% but you have %python_version%. We recommend Python 3.10.12 from %PYTHON_URL%
-   goto err_exit
-)
-
-@rem Cleanup
-del /q .tmp1 .tmp2
-
-@rem -------------- Install and Configure ---------------
-
-call python .\lib\main.py
-pause
-exit /b
-
-@rem ------------------------ Subroutines ---------------
-@rem routine to do comparison of semantic version numbers
-@rem found at https://stackoverflow.com/questions/15807762/compare-version-numbers-in-batch-file
-:compareVersions
-::
-:: Compares two version numbers and returns the result in the ERRORLEVEL
-::
-:: Returns 1 if version1 > version2
-::         0 if version1 = version2
-::        -1 if version1 < version2
-::
-:: The nodes must be delimited by . or , or -
-::
-:: Nodes are normally strictly numeric, without a 0 prefix. A letter suffix
-:: is treated as a separate node
-::
-setlocal enableDelayedExpansion
-set "v1=%~1"
-set "v2=%~2"
-call :divideLetters v1
-call :divideLetters v2
-:loop
-call :parseNode "%v1%" n1 v1
-call :parseNode "%v2%" n2 v2
-if %n1% gtr %n2% exit /b 1
-if %n1% lss %n2% exit /b -1
-if not defined v1 if not defined v2 exit /b 0
-if not defined v1 exit /b -1
-if not defined v2 exit /b 1
-goto :loop
-
-
-:parseNode  version  nodeVar  remainderVar
-for /f "tokens=1* delims=.,-" %%A in ("%~1") do (
-  set "%~2=%%A"
-  set "%~3=%%B"
-)
-exit /b
-
-
-:divideLetters  versionVar
-for %%C in (a b c d e f g h i j k l m n o p q r s t u v w x y z) do set "%~1=!%~1:%%C=.%%C!"
-exit /b
-
-:err_exit
-echo %err_msg%
-echo The installer will exit now.
-pause
-exit /b
-
-pause
-
-:Trim
-SetLocal EnableDelayedExpansion
-set Params=%*
-for /f "tokens=1*" %%a in ("!Params!") do EndLocal & set %1=%%b
-exit /b

installer/install.sh.in

@@ -1,40 +0,0 @@
-#!/bin/bash
-
-# make sure we are not already in a venv
-# (don't need to check status)
-deactivate >/dev/null 2>&1
-scriptdir=$(dirname "$0")
-cd $scriptdir
-
-function version { echo "$@" | awk -F. '{ printf("%d%03d%03d%03d\n", $1,$2,$3,$4); }'; }
-
-MINIMUM_PYTHON_VERSION=3.10.0
-MAXIMUM_PYTHON_VERSION=3.11.100
-PYTHON=""
-for candidate in python3.11 python3.10 python3 python ; do
-    if ppath=`which $candidate 2>/dev/null`; then
-        # when using `pyenv`, the executable for an inactive Python version will exist but will not be operational
-        # we check that this found executable can actually run
-        if [ $($candidate --version &>/dev/null; echo ${PIPESTATUS}) -gt 0 ]; then continue; fi
-
-        python_version=$($ppath -V | awk '{ print $2 }')
-        if [ $(version $python_version) -ge $(version "$MINIMUM_PYTHON_VERSION") ]; then
-            if [ $(version $python_version) -le $(version "$MAXIMUM_PYTHON_VERSION") ]; then
-                PYTHON=$ppath
-                break
-            fi
-        fi
-    fi
-done
-
-if [ -z "$PYTHON" ]; then
-    echo "A suitable Python interpreter could not be found"
-    echo "Please install Python $MINIMUM_PYTHON_VERSION or higher (maximum $MAXIMUM_PYTHON_VERSION) before running this script. See instructions at $INSTRUCTIONS for help."
-    read -p "Press any key to exit"
-    exit -1
-fi
-
-echo "For the best user experience we suggest enlarging or maximizing this window now."
-
-exec $PYTHON ./lib/main.py ${@}
-read -p "Press any key to exit"

installer/lib/installer.py

@@ -1,438 +0,0 @@
-# Copyright (c) 2023 Eugene Brodsky (https://github.com/ebr)
-"""
-InvokeAI installer script
-"""
-
-import locale
-import os
-import platform
-import re
-import shutil
-import subprocess
-import sys
-import venv
-from pathlib import Path
-from tempfile import TemporaryDirectory
-from typing import Optional, Tuple
-
-SUPPORTED_PYTHON = ">=3.10.0,<=3.11.100"
-INSTALLER_REQS = ["rich", "semver", "requests", "plumbum", "prompt-toolkit"]
-BOOTSTRAP_VENV_PREFIX = "invokeai-installer-tmp"
-DOCS_URL = "https://invoke-ai.github.io/InvokeAI/"
-DISCORD_URL = "https://discord.gg/ZmtBAhwWhy"
-
-OS = platform.uname().system
-ARCH = platform.uname().machine
-VERSION = "latest"
-
-
-def get_version_from_wheel_filename(wheel_filename: str) -> str:
-    match = re.search(r"-(\d+\.\d+\.\d+)", wheel_filename)
-    if match:
-        version = match.group(1)
-        return version
-    else:
-        raise ValueError(f"Could not extract version from wheel filename: {wheel_filename}")
-
-
-class Installer:
-    """
-    Deploys an InvokeAI installation into a given path
-    """
-
-    reqs: list[str] = INSTALLER_REQS
-
-    def __init__(self) -> None:
-        if os.getenv("VIRTUAL_ENV") is not None:
-            print("A virtual environment is already activated. Please 'deactivate' before installation.")
-            sys.exit(-1)
-        self.bootstrap()
-        self.available_releases = get_github_releases()
-
-    def mktemp_venv(self) -> TemporaryDirectory[str]:
-        """
-        Creates a temporary virtual environment for the installer itself
-
-        :return: path to the created virtual environment directory
-        :rtype: TemporaryDirectory
-        """
-
-        # Cleaning up temporary directories on Windows results in a race condition
-        # and a stack trace.
-        # `ignore_cleanup_errors` was only added in Python 3.10
-        if OS == "Windows" and int(platform.python_version_tuple()[1]) >= 10:
-            venv_dir = TemporaryDirectory(prefix=BOOTSTRAP_VENV_PREFIX, ignore_cleanup_errors=True)
-        else:
-            venv_dir = TemporaryDirectory(prefix=BOOTSTRAP_VENV_PREFIX)
-
-        venv.create(venv_dir.name, with_pip=True)
-        self.venv_dir = venv_dir
-        set_sys_path(Path(venv_dir.name))
-
-        return venv_dir
-
-    def bootstrap(self, verbose: bool = False) -> TemporaryDirectory[str] | None:
-        """
-        Bootstrap the installer venv with packages required at install time
-        """
-
-        print("Initializing the installer. This may take a minute - please wait...")
-
-        venv_dir = self.mktemp_venv()
-        pip = get_pip_from_venv(Path(venv_dir.name))
-
-        cmd = [pip, "install", "--require-virtualenv", "--use-pep517"]
-        cmd.extend(self.reqs)
-
-        try:
-            # upgrade pip to the latest version to avoid a confusing message
-            res = upgrade_pip(Path(venv_dir.name))
-            if verbose:
-                print(res)
-
-            # run the install prerequisites installation
-            res = subprocess.check_output(cmd).decode()
-
-            if verbose:
-                print(res)
-
-            return venv_dir
-        except subprocess.CalledProcessError as e:
-            print(e)
-
-    def app_venv(self, venv_parent: Path) -> Path:
-        """
-        Create a virtualenv for the InvokeAI installation
-        """
-
-        venv_dir = venv_parent / ".venv"
-
-        # Prefer to copy python executables
-        # so that updates to system python don't break InvokeAI
-        try:
-            venv.create(venv_dir, with_pip=True)
-        # If installing over an existing environment previously created with symlinks,
-        # the executables will fail to copy. Keep symlinks in that case
-        except shutil.SameFileError:
-            venv.create(venv_dir, with_pip=True, symlinks=True)
-
-        return venv_dir
-
-    def install(
-        self,
-        root: str = "~/invokeai",
-        yes_to_all: bool = False,
-        find_links: Optional[str] = None,
-        wheel: Optional[Path] = None,
-    ) -> None:
-        """Install the InvokeAI application into the given runtime path
-
-        Args:
-            root: Destination path for the installation
-            yes_to_all: Accept defaults to all questions
-            find_links: A local directory to search for requirement wheels before going to remote indexes
-            wheel: A wheel file to install
-        """
-
-        import messages
-
-        if wheel:
-            messages.installing_from_wheel(wheel.name)
-            version = get_version_from_wheel_filename(wheel.name)
-        else:
-            messages.welcome(self.available_releases)
-            version = messages.choose_version(self.available_releases)
-
-        auto_dest = Path(os.environ.get("INVOKEAI_ROOT", root)).expanduser().resolve()
-        destination = auto_dest if yes_to_all else messages.dest_path(root)
-        if destination is None:
-            print("Could not find or create the destination directory. Installation cancelled.")
-            sys.exit(0)
-
-        # create the venv for the app
-        self.venv = self.app_venv(venv_parent=destination)
-
-        self.instance = InvokeAiInstance(runtime=destination, venv=self.venv, version=version)
-
-        # install dependencies and the InvokeAI application
-        (extra_index_url, optional_modules) = get_torch_source() if not yes_to_all else (None, None)
-        self.instance.install(extra_index_url, optional_modules, find_links, wheel)
-
-        # install the launch/update scripts into the runtime directory
-        self.instance.install_user_scripts()
-
-        message = f"""
-*** Installation Successful ***
-
-To start the application, run:
-    {destination}/invoke.{"bat" if sys.platform == "win32" else "sh"}
-
-For more information, troubleshooting and support, visit our docs at:
-    {DOCS_URL}
-
-Join the community on Discord:
-    {DISCORD_URL}
-"""
-        print(message)
-
-
-class InvokeAiInstance:
-    """
-    Manages an installed instance of InvokeAI, comprising a virtual environment and a runtime directory.
-    The virtual environment *may* reside within the runtime directory.
-    A single runtime directory *may* be shared by multiple virtual environments, though this isn't currently tested or supported.
-    """
-
-    def __init__(self, runtime: Path, venv: Path, version: str = "stable") -> None:
-        self.runtime = runtime
-        self.venv = venv
-        self.pip = get_pip_from_venv(venv)
-        self.version = version
-
-        set_sys_path(venv)
-        os.environ["INVOKEAI_ROOT"] = str(self.runtime.expanduser().resolve())
-        os.environ["VIRTUAL_ENV"] = str(self.venv.expanduser().resolve())
-        upgrade_pip(venv)
-
-    def get(self) -> tuple[Path, Path]:
-        """
-        Get the location of the virtualenv directory for this installation
-
-        :return: Paths of the runtime and the venv directory
-        :rtype: tuple[Path, Path]
-        """
-
-        return (self.runtime, self.venv)
-
-    def install(
-        self,
-        extra_index_url: Optional[str] = None,
-        optional_modules: Optional[str] = None,
-        find_links: Optional[str] = None,
-        wheel: Optional[Path] = None,
-    ):
-        """Install the package from PyPi or a wheel, if provided.
-
-        Args:
-            extra_index_url: the "--extra-index-url ..." line for pip to look in extra indexes.
-            optional_modules: optional modules to install using "[module1,module2]" format.
-            find_links: path to a directory containing wheels to be searched prior to going to the internet
-            wheel: a wheel file to install
-        """
-
-        import messages
-
-        # not currently used, but may be useful for "install most recent version" option
-        if self.version == "prerelease":
-            version = None
-            pre_flag = "--pre"
-        elif self.version == "stable":
-            version = None
-            pre_flag = None
-        else:
-            version = self.version
-            pre_flag = None
-
-        src = "invokeai"
-        if optional_modules:
-            src += optional_modules
-        if version:
-            src += f"=={version}"
-
-        messages.simple_banner("Installing the InvokeAI Application :art:")
-
-        from plumbum import FG, ProcessExecutionError, local
-
-        pip = local[self.pip]
-
-        # Uninstall xformers if it is present; the correct version of it will be reinstalled if needed
-        _ = pip["uninstall", "-yqq", "xformers"] & FG
-
-        pipeline = pip[
-            "install",
-            "--require-virtualenv",
-            "--force-reinstall",
-            "--use-pep517",
-            str(src) if not wheel else str(wheel),
-            "--find-links" if find_links is not None else None,
-            find_links,
-            "--extra-index-url" if extra_index_url is not None else None,
-            extra_index_url,
-            pre_flag if not wheel else None,  # Ignore the flag if we are installing a wheel
-        ]
-
-        try:
-            _ = pipeline & FG
-        except ProcessExecutionError as e:
-            print(f"Error: {e}")
-            print(
-                "Could not install InvokeAI. Please try downloading the latest version of the installer and install again."
-            )
-            sys.exit(1)
-
-    def install_user_scripts(self):
-        """
-        Copy the launch and update scripts to the runtime dir
-        """
-
-        ext = "bat" if OS == "Windows" else "sh"
-
-        scripts = ["invoke"]
-
-        for script in scripts:
-            src = Path(__file__).parent / ".." / "templates" / f"{script}.{ext}.in"
-            dest = self.runtime / f"{script}.{ext}"
-            shutil.copy(src, dest)
-            os.chmod(dest, 0o0755)
-
-
-### Utility functions ###
-
-
-def get_pip_from_venv(venv_path: Path) -> str:
-    """
-    Given a path to a virtual environment, get the absolute path to the `pip` executable
-    in a cross-platform fashion. Does not validate that the pip executable
-    actually exists in the virtualenv.
-
-    :param venv_path: Path to the virtual environment
-    :type venv_path: Path
-    :return: Absolute path to the pip executable
-    :rtype: str
-    """
-
-    pip = "Scripts\\pip.exe" if OS == "Windows" else "bin/pip"
-    return str(venv_path.expanduser().resolve() / pip)
-
-
-def upgrade_pip(venv_path: Path) -> str | None:
-    """
-    Upgrade the pip executable in the given virtual environment
-    """
-
-    python = "Scripts\\python.exe" if OS == "Windows" else "bin/python"
-    python = str(venv_path.expanduser().resolve() / python)
-
-    try:
-        result = subprocess.check_output([python, "-m", "pip", "install", "--upgrade", "pip"]).decode(
-            encoding=locale.getpreferredencoding()
-        )
-    except subprocess.CalledProcessError as e:
-        print(e)
-        result = None
-
-    return result
-
-
-def set_sys_path(venv_path: Path) -> None:
-    """
-    Given a path to a virtual environment, set the sys.path, in a cross-platform fashion,
-    such that packages from the given venv may be imported in the current process.
-    Ensure that the packages from system environment are not visible (emulate
-    the virtual env 'activate' script) - this doesn't work on Windows yet.
-
-    :param venv_path: Path to the virtual environment
-    :type venv_path: Path
-    """
-
-    # filter out any paths in sys.path that may be system- or user-wide
-    # but leave the temporary bootstrap virtualenv as it contains packages we
-    # temporarily need at install time
-    sys.path = list(filter(lambda p: not p.endswith("-packages") or p.find(BOOTSTRAP_VENV_PREFIX) != -1, sys.path))
-
-    # determine site-packages/lib directory location for the venv
-    lib = "Lib" if OS == "Windows" else f"lib/python{sys.version_info.major}.{sys.version_info.minor}"
-
-    # add the site-packages location to the venv
-    sys.path.append(str(Path(venv_path, lib, "site-packages").expanduser().resolve()))
-
-
-def get_github_releases() -> tuple[list[str], list[str]] | None:
-    """
-    Query Github for published (pre-)release versions.
-    Return a tuple where the first element is a list of stable releases and the second element is a list of pre-releases.
-    Return None if the query fails for any reason.
-    """
-
-    import requests
-
-    ## get latest releases using github api
-    url = "https://api.github.com/repos/invoke-ai/InvokeAI/releases"
-    releases: list[str] = []
-    pre_releases: list[str] = []
-    try:
-        res = requests.get(url)
-        res.raise_for_status()
-        tag_info = res.json()
-        for tag in tag_info:
-            if not tag["prerelease"]:
-                releases.append(tag["tag_name"].lstrip("v"))
-            else:
-                pre_releases.append(tag["tag_name"].lstrip("v"))
-    except requests.HTTPError as e:
-        print(f"Error: {e}")
-        print("Could not fetch version information from GitHub. Please check your network connection and try again.")
-        return
-    except Exception as e:
-        print(f"Error: {e}")
-        print("An unexpected error occurred while trying to fetch version information from GitHub. Please try again.")
-        return
-
-    releases.sort(reverse=True)
-    pre_releases.sort(reverse=True)
-
-    return releases, pre_releases
-
-
-def get_torch_source() -> Tuple[str | None, str | None]:
-    """
-    Determine the extra index URL for pip to use for torch installation.
-    This depends on the OS and the graphics accelerator in use.
-    This is only applicable to Windows and Linux, since PyTorch does not
-    offer accelerated builds for macOS.
-
-    Prefer CUDA-enabled wheels if the user wasn't sure of their GPU, as it will fallback to CPU if possible.
-
-    A NoneType return means just go to PyPi.
-
-    :return: tuple consisting of (extra index url or None, optional modules to load or None)
-    :rtype: list
-    """
-
-    from messages import GpuType, select_gpu
-
-    # device can be one of: "cuda", "rocm", "cpu", "cuda_and_dml, autodetect"
-    device = select_gpu()
-
-    # The correct extra index URLs for torch are inconsistent, see https://pytorch.org/get-started/locally/#start-locally
-
-    url = None
-    optional_modules: str | None = None
-    if OS == "Linux":
-        if device == GpuType.ROCM:
-            url = "https://download.pytorch.org/whl/rocm6.1"
-        elif device == GpuType.CPU:
-            url = "https://download.pytorch.org/whl/cpu"
-        elif device == GpuType.CUDA:
-            url = "https://download.pytorch.org/whl/cu124"
-            optional_modules = "[onnx-cuda]"
-        elif device == GpuType.CUDA_WITH_XFORMERS:
-            url = "https://download.pytorch.org/whl/cu124"
-            optional_modules = "[xformers,onnx-cuda]"
-    elif OS == "Windows":
-        if device == GpuType.CUDA:
-            url = "https://download.pytorch.org/whl/cu124"
-            optional_modules = "[onnx-cuda]"
-        elif device == GpuType.CUDA_WITH_XFORMERS:
-            url = "https://download.pytorch.org/whl/cu124"
-            optional_modules = "[xformers,onnx-cuda]"
-        elif device.value == "cpu":
-            # CPU  uses the default PyPi index, no optional modules
-            pass
-    elif OS == "Darwin":
-        # macOS uses the default PyPi index, no optional modules
-        pass
-
-    # Fall back to defaults
-
-    return (url, optional_modules)

installer/lib/main.py

@@ -1,57 +0,0 @@
-"""
-InvokeAI Installer
-"""
-
-import argparse
-import os
-from pathlib import Path
-
-from installer import Installer
-
-if __name__ == "__main__":
-    parser = argparse.ArgumentParser()
-
-    parser.add_argument(
-        "-r",
-        "--root",
-        dest="root",
-        type=str,
-        help="Destination path for installation",
-        default=os.environ.get("INVOKEAI_ROOT") or "~/invokeai",
-    )
-    parser.add_argument(
-        "-y",
-        "--yes",
-        "--yes-to-all",
-        dest="yes_to_all",
-        action="store_true",
-        help="Assume default answers to all questions",
-        default=False,
-    )
-
-    parser.add_argument(
-        "--find-links",
-        dest="find_links",
-        help="Specifies a directory of local wheel files to be searched prior to searching the online repositories.",
-        type=Path,
-        default=None,
-    )
-
-    parser.add_argument(
-        "--wheel",
-        dest="wheel",
-        help="Specifies a wheel for the InvokeAI package. Used for troubleshooting or testing prereleases.",
-        type=Path,
-        default=None,
-    )
-
-    args = parser.parse_args()
-
-    inst = Installer()
-
-    try:
-        inst.install(**args.__dict__)
-    except KeyboardInterrupt:
-        print("\n")
-        print("Ctrl-C pressed. Aborting.")
-        print("Come back soon!")

installer/lib/messages.py

@@ -1,342 +0,0 @@
-# Copyright (c) 2023 Eugene Brodsky (https://github.com/ebr)
-"""
-Installer user interaction
-"""
-
-import os
-import platform
-from enum import Enum
-from pathlib import Path
-from typing import Optional
-
-from prompt_toolkit import prompt
-from prompt_toolkit.completion import FuzzyWordCompleter, PathCompleter
-from prompt_toolkit.validation import Validator
-from rich import box, print
-from rich.console import Console, Group, group
-from rich.panel import Panel
-from rich.prompt import Confirm
-from rich.style import Style
-from rich.syntax import Syntax
-from rich.text import Text
-
-OS = platform.uname().system
-ARCH = platform.uname().machine
-
-if OS == "Windows":
-    # Windows terminals look better without a background colour
-    console = Console(style=Style(color="grey74"))
-else:
-    console = Console(style=Style(color="grey74", bgcolor="grey19"))
-
-
-def welcome(available_releases: tuple[list[str], list[str]] | None = None) -> None:
-    @group()
-    def text():
-        if (platform_specific := _platform_specific_help()) is not None:
-            yield platform_specific
-            yield ""
-        yield Text.from_markup(
-            "Some of the installation steps take a long time to run. Please be patient. If the script appears to hang for more than 10 minutes, please interrupt with [i]Control-C[/] and retry.",
-            justify="center",
-        )
-        if available_releases is not None:
-            latest_stable = available_releases[0][0]
-            last_pre = available_releases[1][0]
-            yield ""
-            yield Text.from_markup(
-                f"[red3]🠶[/] Latest stable release (recommended): [b bright_white]{latest_stable}", justify="center"
-            )
-            yield Text.from_markup(
-                f"[red3]🠶[/] Last published pre-release version: [b bright_white]{last_pre}", justify="center"
-            )
-
-    console.rule()
-    print(
-        Panel(
-            title="[bold wheat1]Welcome to the InvokeAI Installer",
-            renderable=text(),
-            box=box.DOUBLE,
-            expand=True,
-            padding=(1, 2),
-            style=Style(bgcolor="grey23", color="orange1"),
-            subtitle=f"[bold grey39]{OS}-{ARCH}",
-        )
-    )
-    console.line()
-
-
-def installing_from_wheel(wheel_filename: str) -> None:
-    """Display a message about installing from a wheel"""
-
-    @group()
-    def text():
-        yield Text.from_markup(f"You are installing from a wheel file: [bold]{wheel_filename}\n")
-        yield Text.from_markup(
-            "[bold orange3]If you are not sure why you are doing this, you should cancel and install InvokeAI normally."
-        )
-
-    console.print(
-        Panel(
-            title="Installing from Wheel",
-            renderable=text(),
-            box=box.DOUBLE,
-            expand=True,
-            padding=(1, 2),
-        )
-    )
-
-    should_proceed = Confirm.ask("Do you want to proceed?")
-
-    if not should_proceed:
-        console.print("Installation cancelled.")
-        exit()
-
-
-def choose_version(available_releases: tuple[list[str], list[str]] | None = None) -> str:
-    """
-    Prompt the user to choose an Invoke version to install
-    """
-
-    # short circuit if we couldn't get a version list
-    # still try to install the latest stable version
-    if available_releases is None:
-        return "stable"
-
-    console.print(":grey_question: [orange3]Please choose an Invoke version to install.")
-
-    choices = available_releases[0] + available_releases[1]
-
-    response = prompt(
-        message=f"   <Enter> to install the recommended release ({choices[0]}). <Tab> or type to pick a version: ",
-        complete_while_typing=True,
-        completer=FuzzyWordCompleter(choices),
-    )
-    console.print(f"   Version {choices[0] if response == '' else response} will be installed.")
-
-    console.line()
-
-    return "stable" if response == "" else response
-
-
-def confirm_install(dest: Path) -> bool:
-    if dest.exists():
-        print(f":stop_sign: Directory {dest} already exists!")
-        print("   Is this location correct?")
-        default = False
-    else:
-        print(f":file_folder: InvokeAI will be installed in {dest}")
-        default = True
-
-    dest_confirmed = Confirm.ask("   Please confirm:", default=default)
-
-    console.line()
-
-    return dest_confirmed
-
-
-def dest_path(dest: Optional[str | Path] = None) -> Path | None:
-    """
-    Prompt the user for the destination path and create the path
-
-    :param dest: a filesystem path, defaults to None
-    :type dest: str, optional
-    :return: absolute path to the created installation directory
-    :rtype: Path
-    """
-
-    if dest is not None:
-        dest = Path(dest).expanduser().resolve()
-    else:
-        dest = Path.cwd().expanduser().resolve()
-    prev_dest = init_path = dest
-    dest_confirmed = False
-
-    while not dest_confirmed:
-        browse_start = (dest or Path.cwd()).expanduser().resolve()
-
-        path_completer = PathCompleter(
-            only_directories=True,
-            expanduser=True,
-            get_paths=lambda: [str(browse_start)],  # noqa: B023
-            # get_paths=lambda: [".."].extend(list(browse_start.iterdir()))
-        )
-
-        console.line()
-
-        console.print(f":grey_question: [orange3]Please select the install destination:[/] \\[{browse_start}]: ")
-        selected = prompt(
-            ">>> ",
-            complete_in_thread=True,
-            completer=path_completer,
-            default=str(browse_start) + os.sep,
-            vi_mode=True,
-            complete_while_typing=True,
-            # Test that this is not needed on Windows
-            # complete_style=CompleteStyle.READLINE_LIKE,
-        )
-        prev_dest = dest
-        dest = Path(selected)
-
-        console.line()
-
-        dest_confirmed = confirm_install(dest.expanduser().resolve())
-
-        if not dest_confirmed:
-            dest = prev_dest
-
-    dest = dest.expanduser().resolve()
-
-    try:
-        dest.mkdir(exist_ok=True, parents=True)
-        return dest
-    except PermissionError:
-        console.print(
-            f"Failed to create directory {dest} due to insufficient permissions",
-            style=Style(color="red"),
-            highlight=True,
-        )
-    except OSError:
-        console.print_exception()
-
-    if Confirm.ask("Would you like to try again?"):
-        dest_path(init_path)
-    else:
-        console.rule("Goodbye!")
-
-
-class GpuType(Enum):
-    CUDA_WITH_XFORMERS = "xformers"
-    CUDA = "cuda"
-    ROCM = "rocm"
-    CPU = "cpu"
-
-
-def select_gpu() -> GpuType:
-    """
-    Prompt the user to select the GPU driver
-    """
-
-    if ARCH == "arm64" and OS != "Darwin":
-        print(f"Only CPU acceleration is available on {ARCH} architecture. Proceeding with that.")
-        return GpuType.CPU
-
-    nvidia = (
-        "an [gold1 b]NVIDIA[/] RTX 3060 or newer GPU using CUDA",
-        GpuType.CUDA,
-    )
-    vintage_nvidia = (
-        "an [gold1 b]NVIDIA[/] RTX 20xx or older GPU using CUDA+xFormers",
-        GpuType.CUDA_WITH_XFORMERS,
-    )
-    amd = (
-        "an [gold1 b]AMD[/] GPU using ROCm",
-        GpuType.ROCM,
-    )
-    cpu = (
-        "Do not install any GPU support, use CPU for generation (slow)",
-        GpuType.CPU,
-    )
-
-    options = []
-    if OS == "Windows":
-        options = [nvidia, vintage_nvidia, cpu]
-    if OS == "Linux":
-        options = [nvidia, vintage_nvidia, amd, cpu]
-    elif OS == "Darwin":
-        options = [cpu]
-
-    if len(options) == 1:
-        return options[0][1]
-
-    options = {str(i): opt for i, opt in enumerate(options, 1)}
-
-    console.rule(":space_invader: GPU (Graphics Card) selection :space_invader:")
-    console.print(
-        Panel(
-            Group(
-                "\n".join(
-                    [
-                        f"Detected the [gold1]{OS}-{ARCH}[/] platform",
-                        "",
-                        "See [deep_sky_blue1]https://invoke-ai.github.io/InvokeAI/installation/requirements/[/] to ensure your system meets the minimum requirements.",
-                        "",
-                        "[red3]🠶[/] [b]Your GPU drivers must be correctly installed before using InvokeAI![/] [red3]🠴[/]",
-                    ]
-                ),
-                "",
-                "Please select the type of GPU installed in your computer.",
-                Panel(
-                    "\n".join([f"[dark_goldenrod b i]{i}[/] [dark_red]🢒[/]{opt[0]}" for (i, opt) in options.items()]),
-                    box=box.MINIMAL,
-                ),
-            ),
-            box=box.MINIMAL,
-            padding=(1, 1),
-        )
-    )
-    choice = prompt(
-        "Please make your selection: ",
-        validator=Validator.from_callable(
-            lambda n: n in options.keys(), error_message="Please select one the above options"
-        ),
-    )
-
-    return options[choice][1]
-
-
-def simple_banner(message: str) -> None:
-    """
-    A simple banner with a message, defined here for styling consistency
-
-    :param message: The message to display
-    :type message: str
-    """
-
-    console.rule(message)
-
-
-# TODO this does not yet work correctly
-def windows_long_paths_registry() -> None:
-    """
-    Display a message about applying the Windows long paths registry fix
-    """
-
-    with open(str(Path(__file__).parent / "WinLongPathsEnabled.reg"), "r", encoding="utf-16le") as code:
-        syntax = Syntax(code.read(), line_numbers=True, lexer="regedit")
-
-    console.print(
-        Panel(
-            Group(
-                "\n".join(
-                    [
-                        "We will now apply a registry fix to enable long paths on Windows. InvokeAI needs this to function correctly. We are asking your permission to modify the Windows Registry on your behalf.",
-                        "",
-                        "This is the change that will be applied:",
-                        str(syntax),
-                    ]
-                )
-            ),
-            title="Windows Long Paths registry fix",
-            box=box.HORIZONTALS,
-            padding=(1, 1),
-        )
-    )
-
-
-def _platform_specific_help() -> Text | None:
-    if OS == "Darwin":
-        text = Text.from_markup(
-            """[b wheat1]macOS Users![/]\n\nPlease be sure you have the [b wheat1]Xcode command-line tools[/] installed before continuing.\nIf not, cancel with [i]Control-C[/] and follow the Xcode install instructions at [deep_sky_blue1]https://www.freecodecamp.org/news/install-xcode-command-line-tools/[/]."""
-        )
-    elif OS == "Windows":
-        text = Text.from_markup(
-            """[b wheat1]Windows Users![/]\n\nBefore you start, please do the following:
-  1. Double-click on the file [b wheat1]WinLongPathsEnabled.reg[/] in order to
-     enable long path support on your system.
-  2. Make sure you have the [b wheat1]Visual C++ core libraries[/] installed. If not, install from
-     [deep_sky_blue1]https://learn.microsoft.com/en-US/cpp/windows/latest-supported-vc-redist?view=msvc-170[/]"""
-        )
-    else:
-        return
-    return text

installer/readme.txt

@@ -1,52 +0,0 @@
-InvokeAI
-
-Project homepage: https://github.com/invoke-ai/InvokeAI
-
-Preparations:
-
-  You will need to install Python 3.10 or higher for this installer
-  to work. Instructions are given here:
-  https://invoke-ai.github.io/InvokeAI/installation/INSTALL_AUTOMATED/
-
-  Before you start the installer, please open up your system's command
-  line window (Terminal or Command) and type the commands:
-
-     python --version
-
-  If all is well, it will print "Python 3.X.X", where the version number
-  is at least 3.10.*, and not higher than 3.11.*.
-
-  If this works, check the version of the Python package manager, pip:
-
-     pip --version
-
-  You should get a message that indicates that the pip package
-  installer was derived from Python 3.10 or 3.11. For example:
-  "pip 22.0.1 from /usr/bin/pip (python 3.10)"
-
-Long Paths on Windows:
-
-  If you are on Windows, you will need to enable Windows Long Paths to
-  run InvokeAI successfully. If you're not sure what this is, you
-  almost certainly need to do this.
-
-  Simply double-click the "WinLongPathsEnabled.reg" file located in
-  this directory, and approve the Windows warnings. Note that you will
-  need to have admin privileges in order to do this.
-
-Launching the installer:
-
-  Windows: double-click the 'install.bat' file (while keeping it inside
-  the InvokeAI-Installer folder).
-
-  Linux and Mac: Please open the terminal application and run
-  './install.sh' (while keeping it inside the InvokeAI-Installer
-  folder).
-
-The installer will create a directory of your choice and install the
-InvokeAI application within it. This directory contains everything you need to run
-invokeai. Once InvokeAI is up and running, you may delete the
-InvokeAI-Installer folder at your convenience.
-
-For more information, please see
-https://invoke-ai.github.io/InvokeAI/installation/INSTALL_AUTOMATED/

installer/templates/invoke.bat.in

@@ -1,54 +0,0 @@
-@echo off
-
-PUSHD "%~dp0"
-setlocal
-
-call .venv\Scripts\activate.bat
-set INVOKEAI_ROOT=.
-
-:start
-echo Desired action:
-echo 1. Generate images with the browser-based interface
-echo 2. Open the developer console
-echo 3. Command-line help
-echo Q - Quit
-echo.
-echo To update, download and run the installer from https://github.com/invoke-ai/InvokeAI/releases/latest
-echo.
-set /P choice="Please enter 1-4, 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 %*
-) ELSE IF /I "%choice%" == "2" (
-    echo Developer Console
-    echo Python command is:
-    where python
-    echo Python version is:
-    python --version
-    echo *************************
-    echo You are now in the system shell, with the local InvokeAI Python virtual environment activated,
-    echo so that you can troubleshoot this InvokeAI installation as necessary.
-    echo *************************
-    echo *** Type `exit` to quit this shell and deactivate the Python virtual environment ***
-    call cmd /k
-) ELSE IF /I "%choice%" == "3" (
-    echo Displaying command line help...
-    python .venv\Scripts\invokeai-web.exe --help %*
-    pause
-    exit /b
-) ELSE IF /I "%choice%" == "q" (
-    echo Goodbye!
-    goto ending
-) ELSE (
-    echo Invalid selection
-    pause
-    exit /b
-)
-goto start
-
-endlocal
-pause
-
-:ending
-exit /b

installer/templates/invoke.sh.in

@@ -1,87 +0,0 @@
-#!/bin/bash
-
-# MIT License
-
-# Coauthored by Lincoln Stein, Eugene Brodsky and Joshua Kimsey
-# Copyright 2023, The InvokeAI Development Team
-
-####
-# This launch script assumes that:
-# 1. it is located in the runtime directory,
-# 2. the .venv is also located in the runtime directory and is named exactly that
-#
-# If both of the above are not true, this script will likely not work as intended.
-# Activate the virtual environment and run `invoke.py` directly.
-####
-
-set -eu
-
-# Ensure we're in the correct folder in case user's CWD is somewhere else
-scriptdir=$(dirname $(readlink -f "$0"))
-cd "$scriptdir"
-
-. .venv/bin/activate
-
-export INVOKEAI_ROOT="$scriptdir"
-
-# Stash the CLI args - when we prompt for user input, `$@` is overwritten
-PARAMS=$@
-
-# This setting allows torch to fall back to CPU for operations that are not supported by MPS on macOS.
-if [ "$(uname -s)" == "Darwin" ]; then
-    export PYTORCH_ENABLE_MPS_FALLBACK=1
-fi
-
-# Primary function for the case statement to determine user input
-do_choice() {
-    case $1 in
-    1)
-        clear
-        printf "Generate images with a browser-based interface\n"
-        invokeai-web $PARAMS
-        ;;
-    2)
-        clear
-        printf "Open the developer console\n"
-        file_name=$(basename "${BASH_SOURCE[0]}")
-        bash --init-file "$file_name"
-        ;;
-    3)
-        clear
-        printf "Command-line help\n"
-        invokeai-web --help
-        ;;
-    *)
-        clear
-        printf "Exiting...\n"
-        exit
-        ;;
-    esac
-    clear
-}
-
-# Command-line interface for launching Invoke functions
-do_line_input() {
-    clear
-    printf "What would you like to do?\n"
-    printf "1: Generate images using the browser-based interface\n"
-    printf "2: Open the developer console\n"
-    printf "3: Command-line help\n"
-    printf "Q: Quit\n\n"
-    printf "To update, download and run the installer from https://github.com/invoke-ai/InvokeAI/releases/latest\n\n"
-    read -p "Please enter 1-4, Q: [1] " yn
-    choice=${yn:='1'}
-    do_choice $choice
-    clear
-}
-
-# Main IF statement for launching Invoke, and for checking if the user is in the developer console
-if [ "$0" != "bash" ]; then
-    while true; do
-        do_line_input
-    done
-else # in developer console
-    python --version
-    printf "Press ^D to exit\n"
-    export PS1="(InvokeAI) \u@\h \w> "
-fi

invokeai/app/api/dependencies.py

@@ -23,6 +23,10 @@ from invokeai.app.services.invoker import Invoker
 from invokeai.app.services.model_images.model_images_default import ModelImageFileStorageDisk
 from invokeai.app.services.model_manager.model_manager_default import ModelManagerService
 from invokeai.app.services.model_records.model_records_sql import ModelRecordServiceSQL
+from invokeai.app.services.model_relationship_records.model_relationship_records_sqlite import (
+    SqliteModelRelationshipRecordStorage,
+)
+from invokeai.app.services.model_relationships.model_relationships_default import ModelRelationshipsService
 from invokeai.app.services.names.names_default import SimpleNameService
 from invokeai.app.services.object_serializer.object_serializer_disk import ObjectSerializerDisk
 from invokeai.app.services.object_serializer.object_serializer_forward_cache import ObjectSerializerForwardCache
@@ -37,7 +41,14 @@ from invokeai.app.services.style_preset_records.style_preset_records_sqlite impo
 from invokeai.app.services.urls.urls_default import LocalUrlService
 from invokeai.app.services.workflow_records.workflow_records_sqlite import SqliteWorkflowRecordsStorage
 from invokeai.app.services.workflow_thumbnails.workflow_thumbnails_disk import WorkflowThumbnailFileStorageDisk
-from invokeai.backend.stable_diffusion.diffusion.conditioning_data import ConditioningFieldData
+from invokeai.backend.stable_diffusion.diffusion.conditioning_data import (
+    BasicConditioningInfo,
+    CogView4ConditioningInfo,
+    ConditioningFieldData,
+    FLUXConditioningInfo,
+    SD3ConditioningInfo,
+    SDXLConditioningInfo,
+)
 from invokeai.backend.util.logging import InvokeAILogger
 from invokeai.version.invokeai_version import __version__
 
@@ -101,10 +112,25 @@ class ApiDependencies:
         images = ImageService()
         invocation_cache = MemoryInvocationCache(max_cache_size=config.node_cache_size)
         tensors = ObjectSerializerForwardCache(
-            ObjectSerializerDisk[torch.Tensor](output_folder / "tensors", ephemeral=True)
+            ObjectSerializerDisk[torch.Tensor](
+                output_folder / "tensors",
+                safe_globals=[torch.Tensor],
+                ephemeral=True,
+            ),
         )
         conditioning = ObjectSerializerForwardCache(
-            ObjectSerializerDisk[ConditioningFieldData](output_folder / "conditioning", ephemeral=True)
+            ObjectSerializerDisk[ConditioningFieldData](
+                output_folder / "conditioning",
+                safe_globals=[
+                    ConditioningFieldData,
+                    BasicConditioningInfo,
+                    SDXLConditioningInfo,
+                    FLUXConditioningInfo,
+                    SD3ConditioningInfo,
+                    CogView4ConditioningInfo,
+                ],
+                ephemeral=True,
+            ),
         )
         download_queue_service = DownloadQueueService(app_config=configuration, event_bus=events)
         model_images_service = ModelImageFileStorageDisk(model_images_folder / "model_images")
@@ -114,6 +140,8 @@ class ApiDependencies:
             download_queue=download_queue_service,
             events=events,
         )
+        model_relationships = ModelRelationshipsService()
+        model_relationship_records = SqliteModelRelationshipRecordStorage(db=db)
         names = SimpleNameService()
         performance_statistics = InvocationStatsService()
         session_processor = DefaultSessionProcessor(session_runner=DefaultSessionRunner())
@@ -139,6 +167,8 @@ class ApiDependencies:
             logger=logger,
             model_images=model_images_service,
             model_manager=model_manager,
+            model_relationships=model_relationships,
+            model_relationship_records=model_relationship_records,
             download_queue=download_queue_service,
             names=names,
             performance_statistics=performance_statistics,

invokeai/app/api/routers/app_info.py

@@ -1,8 +1,7 @@
 import typing
 from enum import Enum
-from importlib.metadata import PackageNotFoundError, version
+from importlib.metadata import distributions
 from pathlib import Path
-from platform import python_version
 from typing import Optional
 
 import torch
@@ -12,6 +11,7 @@ from pydantic import BaseModel, Field
 
 from invokeai.app.api.dependencies import ApiDependencies
 from invokeai.app.invocations.upscale import ESRGAN_MODELS
+from invokeai.app.services.config.config_default import InvokeAIAppConfig, get_config
 from invokeai.app.services.invocation_cache.invocation_cache_common import InvocationCacheStatus
 from invokeai.backend.image_util.infill_methods.patchmatch import PatchMatch
 from invokeai.backend.util.logging import logging
@@ -43,24 +43,6 @@ class AppVersion(BaseModel):
     highlights: Optional[list[str]] = Field(default=None, description="Highlights of release")
 
 
-class AppDependencyVersions(BaseModel):
-    """App depencency Versions Response"""
-
-    accelerate: str = Field(description="accelerate version")
-    compel: str = Field(description="compel version")
-    cuda: Optional[str] = Field(description="CUDA version")
-    diffusers: str = Field(description="diffusers version")
-    numpy: str = Field(description="Numpy version")
-    opencv: str = Field(description="OpenCV version")
-    onnx: str = Field(description="ONNX version")
-    pillow: str = Field(description="Pillow (PIL) version")
-    python: str = Field(description="Python version")
-    torch: str = Field(description="PyTorch version")
-    torchvision: str = Field(description="PyTorch Vision version")
-    transformers: str = Field(description="transformers version")
-    xformers: Optional[str] = Field(description="xformers version")
-
-
 class AppConfig(BaseModel):
     """App Config Response"""
 
@@ -75,31 +57,23 @@ async def get_version() -> AppVersion:
     return AppVersion(version=__version__)
 
 
-@app_router.get("/app_deps", operation_id="get_app_deps", status_code=200, response_model=AppDependencyVersions)
-async def get_app_deps() -> AppDependencyVersions:
+@app_router.get("/app_deps", operation_id="get_app_deps", status_code=200, response_model=dict[str, str])
+async def get_app_deps() -> dict[str, str]:
+    deps: dict[str, str] = {dist.metadata["Name"]: dist.version for dist in distributions()}
     try:
-        xformers = version("xformers")
-    except PackageNotFoundError:
-        xformers = None
-    return AppDependencyVersions(
-        accelerate=version("accelerate"),
-        compel=version("compel"),
-        cuda=torch.version.cuda,
-        diffusers=version("diffusers"),
-        numpy=version("numpy"),
-        opencv=version("opencv-python"),
-        onnx=version("onnx"),
-        pillow=version("pillow"),
-        python=python_version(),
-        torch=torch.version.__version__,
-        torchvision=version("torchvision"),
-        transformers=version("transformers"),
-        xformers=xformers,
-    )
+        cuda = torch.version.cuda or "N/A"
+    except Exception:
+        cuda = "N/A"
+
+    deps["CUDA"] = cuda
+
+    sorted_deps = dict(sorted(deps.items(), key=lambda item: item[0].lower()))
+
+    return sorted_deps
 
 
 @app_router.get("/config", operation_id="get_config", status_code=200, response_model=AppConfig)
-async def get_config() -> AppConfig:
+async def get_config_() -> AppConfig:
     infill_methods = ["lama", "tile", "cv2", "color"]  # TODO: add mosaic back
     if PatchMatch.patchmatch_available():
         infill_methods.append("patchmatch")
@@ -121,6 +95,21 @@ async def get_config() -> AppConfig:
     )
 
 
+class InvokeAIAppConfigWithSetFields(BaseModel):
+    """InvokeAI App Config with model fields set"""
+
+    set_fields: set[str] = Field(description="The set fields")
+    config: InvokeAIAppConfig = Field(description="The InvokeAI App Config")
+
+
+@app_router.get(
+    "/runtime_config", operation_id="get_runtime_config", status_code=200, response_model=InvokeAIAppConfigWithSetFields
+)
+async def get_runtime_config() -> InvokeAIAppConfigWithSetFields:
+    config = get_config()
+    return InvokeAIAppConfigWithSetFields(set_fields=config.model_fields_set, config=config)
+
+
 @app_router.get(
     "/logging",
     operation_id="get_log_level",

invokeai/app/api/routers/boards.py

@@ -146,7 +146,7 @@ async def list_boards(
     response_model=list[str],
 )
 async def list_all_board_image_names(
-    board_id: str = Path(description="The id of the board"),
+    board_id: str = Path(description="The id of the board or 'none' for uncategorized images"),
     categories: list[ImageCategory] | None = Query(default=None, description="The categories of image to include."),
     is_intermediate: bool | None = Query(default=None, description="Whether to list intermediate images."),
 ) -> list[str]:

invokeai/app/api/routers/images.py

@@ -1,12 +1,13 @@
 import io
+import json
 import traceback
-from typing import Optional
+from typing import ClassVar, Optional
 
 from fastapi import BackgroundTasks, Body, HTTPException, Path, Query, Request, Response, UploadFile
 from fastapi.responses import FileResponse
 from fastapi.routing import APIRouter
 from PIL import Image
-from pydantic import BaseModel, Field
+from pydantic import BaseModel, Field, model_validator
 
 from invokeai.app.api.dependencies import ApiDependencies
 from invokeai.app.api.extract_metadata_from_image import extract_metadata_from_image
@@ -19,6 +20,8 @@ from invokeai.app.services.image_records.image_records_common import (
 from invokeai.app.services.images.images_common import ImageDTO, ImageUrlsDTO
 from invokeai.app.services.shared.pagination import OffsetPaginatedResults
 from invokeai.app.services.shared.sqlite.sqlite_common import SQLiteDirection
+from invokeai.app.util.controlnet_utils import heuristic_resize_fast
+from invokeai.backend.image_util.util import np_to_pil, pil_to_np
 
 images_router = APIRouter(prefix="/v1/images", tags=["images"])
 
@@ -27,6 +30,19 @@ images_router = APIRouter(prefix="/v1/images", tags=["images"])
 IMAGE_MAX_AGE = 31536000
 
 
+class ResizeToDimensions(BaseModel):
+    width: int = Field(..., gt=0)
+    height: int = Field(..., gt=0)
+
+    MAX_SIZE: ClassVar[int] = 4096 * 4096
+
+    @model_validator(mode="after")
+    def validate_total_output_size(self):
+        if self.width * self.height > self.MAX_SIZE:
+            raise ValueError(f"Max total output size for resizing is {self.MAX_SIZE} pixels")
+        return self
+
+
 @images_router.post(
     "/upload",
     operation_id="upload_image",
@@ -46,6 +62,11 @@ async def upload_image(
     board_id: Optional[str] = Query(default=None, description="The board to add this image to, if any"),
     session_id: Optional[str] = Query(default=None, description="The session ID associated with this upload, if any"),
     crop_visible: Optional[bool] = Query(default=False, description="Whether to crop the image"),
+    resize_to: Optional[str] = Body(
+        default=None,
+        description=f"Dimensions to resize the image to, must be stringified tuple of 2 integers. Max total pixel count: {ResizeToDimensions.MAX_SIZE}",
+        example='"[1024,1024]"',
+    ),
     metadata: Optional[str] = Body(
         default=None,
         description="The metadata to associate with the image, must be a stringified JSON dict",
@@ -59,13 +80,33 @@ async def upload_image(
     contents = await file.read()
     try:
         pil_image = Image.open(io.BytesIO(contents))
-        if crop_visible:
-            bbox = pil_image.getbbox()
-            pil_image = pil_image.crop(bbox)
     except Exception:
         ApiDependencies.invoker.services.logger.error(traceback.format_exc())
         raise HTTPException(status_code=415, detail="Failed to read image")
 
+    if crop_visible:
+        try:
+            bbox = pil_image.getbbox()
+            pil_image = pil_image.crop(bbox)
+        except Exception:
+            raise HTTPException(status_code=500, detail="Failed to crop image")
+
+    if resize_to:
+        try:
+            dims = json.loads(resize_to)
+            resize_dims = ResizeToDimensions(**dims)
+        except Exception:
+            raise HTTPException(status_code=400, detail="Invalid resize_to format or size")
+
+        try:
+            # heuristic_resize_fast expects an RGB or RGBA image
+            pil_rgba = pil_image.convert("RGBA")
+            np_image = pil_to_np(pil_rgba)
+            np_image = heuristic_resize_fast(np_image, (resize_dims.width, resize_dims.height))
+            pil_image = np_to_pil(np_image)
+        except Exception:
+            raise HTTPException(status_code=500, detail="Failed to resize image")
+
     extracted_metadata = extract_metadata_from_image(
         pil_image=pil_image,
         invokeai_metadata_override=metadata,
@@ -96,6 +137,22 @@ async def upload_image(
         raise HTTPException(status_code=500, detail="Failed to create image")
 
 
+class ImageUploadEntry(BaseModel):
+    image_dto: ImageDTO = Body(description="The image DTO")
+    presigned_url: str = Body(description="The URL to get the presigned URL for the image upload")
+
+
+@images_router.post("/", operation_id="create_image_upload_entry")
+async def create_image_upload_entry(
+    width: int = Body(description="The width of the image"),
+    height: int = Body(description="The height of the image"),
+    board_id: Optional[str] = Body(default=None, description="The board to add this image to, if any"),
+) -> ImageUploadEntry:
+    """Uploads an image from a URL, not implemented"""
+
+    raise HTTPException(status_code=501, detail="Not implemented")
+
+
 @images_router.delete("/i/{image_name}", operation_id="delete_image")
 async def delete_image(
     image_name: str = Path(description="The name of the image to delete"),
@@ -340,6 +397,29 @@ async def delete_images_from_list(
         raise HTTPException(status_code=500, detail="Failed to delete images")
 
 
+@images_router.delete(
+    "/uncategorized", operation_id="delete_uncategorized_images", response_model=DeleteImagesFromListResult
+)
+async def delete_uncategorized_images() -> DeleteImagesFromListResult:
+    """Deletes all images that are uncategorized"""
+
+    image_names = ApiDependencies.invoker.services.board_images.get_all_board_image_names_for_board(
+        board_id="none", categories=None, is_intermediate=None
+    )
+
+    try:
+        deleted_images: list[str] = []
+        for image_name in image_names:
+            try:
+                ApiDependencies.invoker.services.images.delete(image_name)
+                deleted_images.append(image_name)
+            except Exception:
+                pass
+        return DeleteImagesFromListResult(deleted_images=deleted_images)
+    except Exception:
+        raise HTTPException(status_code=500, detail="Failed to delete images")
+
+
 class ImagesUpdatedFromListResult(BaseModel):
     updated_image_names: list[str] = Field(description="The image names that were updated")
 

invokeai/app/api/routers/model_manager.py

@@ -28,12 +28,10 @@ from invokeai.app.services.model_records import (
     UnknownModelException,
 )
 from invokeai.app.util.suppress_output import SuppressOutput
+from invokeai.backend.model_manager import BaseModelType, ModelFormat, ModelType
 from invokeai.backend.model_manager.config import (
     AnyModelConfig,
-    BaseModelType,
     MainCheckpointConfig,
-    ModelFormat,
-    ModelType,
 )
 from invokeai.backend.model_manager.load.model_cache.cache_stats import CacheStats
 from invokeai.backend.model_manager.metadata.fetch.huggingface import HuggingFaceMetadataFetch
@@ -87,6 +85,7 @@ example_model_config = {
     "config_path": "string",
     "key": "string",
     "hash": "string",
+    "file_size": 1,
     "description": "string",
     "source": "string",
     "converted_at": 0,
@@ -894,6 +893,12 @@ class HFTokenHelper:
             huggingface_hub.login(token=token, add_to_git_credential=False)
         return cls.get_status()
 
+    @classmethod
+    def reset_token(cls) -> HFTokenStatus:
+        with SuppressOutput(), contextlib.suppress(Exception):
+            huggingface_hub.logout()
+        return cls.get_status()
+
 
 @model_manager_router.get("/hf_login", operation_id="get_hf_login_status", response_model=HFTokenStatus)
 async def get_hf_login_status() -> HFTokenStatus:
@@ -916,3 +921,8 @@ async def do_hf_login(
         ApiDependencies.invoker.services.logger.warning("Unable to verify HF token")
 
     return token_status
+
+
+@model_manager_router.delete("/hf_login", operation_id="reset_hf_token", response_model=HFTokenStatus)
+async def reset_hf_token() -> HFTokenStatus:
+    return HFTokenHelper.reset_token()

invokeai/app/api/routers/model_relationships.py

@@ -0,0 +1,215 @@
+"""FastAPI route for model relationship records."""
+
+from typing import List
+
+from fastapi import APIRouter, Body, HTTPException, Path, status
+from pydantic import BaseModel, Field
+
+from invokeai.app.api.dependencies import ApiDependencies
+
+model_relationships_router = APIRouter(prefix="/v1/model_relationships", tags=["model_relationships"])
+
+# === Schemas ===
+
+
+class ModelRelationshipCreateRequest(BaseModel):
+    model_key_1: str = Field(
+        ...,
+        description="The key of the first model in the relationship",
+        examples=[
+            "aa3b247f-90c9-4416-bfcd-aeaa57a5339e",
+            "ac32b914-10ab-496e-a24a-3068724b9c35",
+            "d944abfd-c7c3-42e2-a4ff-da640b29b8b4",
+            "b1c2d3e4-f5a6-7890-abcd-ef1234567890",
+            "12345678-90ab-cdef-1234-567890abcdef",
+            "fedcba98-7654-3210-fedc-ba9876543210",
+        ],
+    )
+    model_key_2: str = Field(
+        ...,
+        description="The key of the second model in the relationship",
+        examples=[
+            "3bb7c0eb-b6c8-469c-ad8c-4d69c06075e4",
+            "f0c3da4e-d9ff-42b5-a45c-23be75c887c9",
+            "38170dd8-f1e5-431e-866c-2c81f1277fcc",
+            "c57fea2d-7646-424c-b9ad-c0ba60fc68be",
+            "10f7807b-ab54-46a9-ab03-600e88c630a1",
+            "f6c1d267-cf87-4ee0-bee0-37e791eacab7",
+        ],
+    )
+
+
+class ModelRelationshipBatchRequest(BaseModel):
+    model_keys: List[str] = Field(
+        ...,
+        description="List of model keys to fetch related models for",
+        examples=[
+            [
+                "aa3b247f-90c9-4416-bfcd-aeaa57a5339e",
+                "ac32b914-10ab-496e-a24a-3068724b9c35",
+            ],
+            [
+                "b1c2d3e4-f5a6-7890-abcd-ef1234567890",
+                "12345678-90ab-cdef-1234-567890abcdef",
+                "fedcba98-7654-3210-fedc-ba9876543210",
+            ],
+            [
+                "3bb7c0eb-b6c8-469c-ad8c-4d69c06075e4",
+            ],
+        ],
+    )
+
+
+# === Routes ===
+
+
+@model_relationships_router.get(
+    "/i/{model_key}",
+    operation_id="get_related_models",
+    response_model=list[str],
+    responses={
+        200: {
+            "description": "A list of related model keys was retrieved successfully",
+            "content": {
+                "application/json": {
+                    "example": [
+                        "15e9eb28-8cfe-47c9-b610-37907a79fc3c",
+                        "71272e82-0e5f-46d5-bca9-9a61f4bd8a82",
+                        "a5d7cd49-1b98-4534-a475-aeee4ccf5fa2",
+                    ]
+                }
+            },
+        },
+        404: {"description": "The specified model could not be found"},
+        422: {"description": "Validation error"},
+    },
+)
+async def get_related_models(
+    model_key: str = Path(..., description="The key of the model to get relationships for"),
+) -> list[str]:
+    """
+    Get a list of model keys related to a given model.
+    """
+    try:
+        return ApiDependencies.invoker.services.model_relationships.get_related_model_keys(model_key)
+    except Exception as e:
+        raise HTTPException(status_code=500, detail=str(e))
+
+
+@model_relationships_router.post(
+    "/",
+    status_code=status.HTTP_204_NO_CONTENT,
+    responses={
+        204: {"description": "The relationship was successfully created"},
+        400: {"description": "Invalid model keys or self-referential relationship"},
+        409: {"description": "The relationship already exists"},
+        422: {"description": "Validation error"},
+        500: {"description": "Internal server error"},
+    },
+    summary="Add Model Relationship",
+    description="Creates a **bidirectional** relationship between two models, allowing each to reference the other as related.",
+)
+async def add_model_relationship(
+    req: ModelRelationshipCreateRequest = Body(..., description="The model keys to relate"),
+) -> None:
+    """
+    Add a relationship between two models.
+
+    Relationships are bidirectional and will be accessible from both models.
+
+    - Raises 400 if keys are invalid or identical.
+    - Raises 409 if the relationship already exists.
+    """
+    try:
+        if req.model_key_1 == req.model_key_2:
+            raise HTTPException(status_code=400, detail="Cannot relate a model to itself.")
+
+        ApiDependencies.invoker.services.model_relationships.add_model_relationship(
+            req.model_key_1,
+            req.model_key_2,
+        )
+    except ValueError as e:
+        raise HTTPException(status_code=409, detail=str(e))
+    except Exception as e:
+        raise HTTPException(status_code=500, detail=str(e))
+
+
+@model_relationships_router.delete(
+    "/",
+    status_code=status.HTTP_204_NO_CONTENT,
+    responses={
+        204: {"description": "The relationship was successfully removed"},
+        400: {"description": "Invalid model keys or self-referential relationship"},
+        404: {"description": "The relationship does not exist"},
+        422: {"description": "Validation error"},
+        500: {"description": "Internal server error"},
+    },
+    summary="Remove Model Relationship",
+    description="Removes a **bidirectional** relationship between two models. The relationship must already exist.",
+)
+async def remove_model_relationship(
+    req: ModelRelationshipCreateRequest = Body(..., description="The model keys to disconnect"),
+) -> None:
+    """
+    Removes a bidirectional relationship between two model keys.
+
+    - Raises 400 if attempting to unlink a model from itself.
+    - Raises 404 if the relationship was not found.
+    """
+    try:
+        if req.model_key_1 == req.model_key_2:
+            raise HTTPException(status_code=400, detail="Cannot unlink a model from itself.")
+
+        ApiDependencies.invoker.services.model_relationships.remove_model_relationship(
+            req.model_key_1,
+            req.model_key_2,
+        )
+    except ValueError as e:
+        raise HTTPException(status_code=404, detail=str(e))
+    except Exception as e:
+        raise HTTPException(status_code=500, detail=str(e))
+
+
+@model_relationships_router.post(
+    "/batch",
+    operation_id="get_related_models_batch",
+    response_model=List[str],
+    responses={
+        200: {
+            "description": "Related model keys retrieved successfully",
+            "content": {
+                "application/json": {
+                    "example": [
+                        "ca562b14-995e-4a42-90c1-9528f1a5921d",
+                        "cc0c2b8a-c62e-41d6-878e-cc74dde5ca8f",
+                        "18ca7649-6a9e-47d5-bc17-41ab1e8cec81",
+                        "7c12d1b2-0ef9-4bec-ba55-797b2d8f2ee1",
+                        "c382eaa3-0e28-4ab0-9446-408667699aeb",
+                        "71272e82-0e5f-46d5-bca9-9a61f4bd8a82",
+                        "a5d7cd49-1b98-4534-a475-aeee4ccf5fa2",
+                    ]
+                }
+            },
+        },
+        422: {"description": "Validation error"},
+        500: {"description": "Internal server error"},
+    },
+    summary="Get Related Model Keys (Batch)",
+    description="Retrieves all **unique related model keys** for a list of given models. This is useful for contextual suggestions or filtering.",
+)
+async def get_related_models_batch(
+    req: ModelRelationshipBatchRequest = Body(..., description="Model keys to check for related connections"),
+) -> list[str]:
+    """
+    Accepts multiple model keys and returns a flat list of all unique related keys.
+
+    Useful when working with multiple selections in the UI or cross-model comparisons.
+    """
+    try:
+        all_related: set[str] = set()
+        for key in req.model_keys:
+            related = ApiDependencies.invoker.services.model_relationships.get_related_model_keys(key)
+            all_related.update(related)
+        return list(all_related)
+    except Exception as e:
+        raise HTTPException(status_code=500, detail=str(e))

invokeai/app/api/routers/session_queue.py

@@ -2,7 +2,7 @@ from typing import Optional
 
 from fastapi import Body, Path, Query
 from fastapi.routing import APIRouter
-from pydantic import BaseModel
+from pydantic import BaseModel, Field
 
 from invokeai.app.api.dependencies import ApiDependencies
 from invokeai.app.services.session_processor.session_processor_common import SessionProcessorStatus
@@ -15,6 +15,7 @@ from invokeai.app.services.session_queue.session_queue_common import (
     CancelByDestinationResult,
     ClearResult,
     EnqueueBatchResult,
+    FieldIdentifier,
     PruneResult,
     RetryItemsResult,
     SessionQueueCountsByDestination,
@@ -34,6 +35,12 @@ class SessionQueueAndProcessorStatus(BaseModel):
     processor: SessionProcessorStatus
 
 
+class ValidationRunData(BaseModel):
+    workflow_id: str = Field(description="The id of the workflow being published.")
+    input_fields: list[FieldIdentifier] = Body(description="The input fields for the published workflow")
+    output_fields: list[FieldIdentifier] = Body(description="The output fields for the published workflow")
+
+
 @session_queue_router.post(
     "/{queue_id}/enqueue_batch",
     operation_id="enqueue_batch",
@@ -45,6 +52,10 @@ async def enqueue_batch(
     queue_id: str = Path(description="The queue id to perform this operation on"),
     batch: Batch = Body(description="Batch to process"),
     prepend: bool = Body(default=False, description="Whether or not to prepend this batch in the queue"),
+    validation_run_data: Optional[ValidationRunData] = Body(
+        default=None,
+        description="The validation run data to use for this batch. This is only used if this is a validation run.",
+    ),
 ) -> EnqueueBatchResult:
     """Processes a batch and enqueues the output graphs for execution."""
 

invokeai/app/api/routers/workflows.py

@@ -106,6 +106,7 @@ async def list_workflows(
     tags: Optional[list[str]] = Query(default=None, description="The tags of workflow to get"),
     query: Optional[str] = Query(default=None, description="The text to query by (matches name and description)"),
     has_been_opened: Optional[bool] = Query(default=None, description="Whether to include/exclude recent workflows"),
+    is_published: Optional[bool] = Query(default=None, description="Whether to include/exclude published workflows"),
 ) -> PaginatedResults[WorkflowRecordListItemWithThumbnailDTO]:
     """Gets a page of workflows"""
     workflows_with_thumbnails: list[WorkflowRecordListItemWithThumbnailDTO] = []
@@ -118,6 +119,7 @@ async def list_workflows(
         categories=categories,
         tags=tags,
         has_been_opened=has_been_opened,
+        is_published=is_published,
     )
     for workflow in workflows.items:
         workflows_with_thumbnails.append(

invokeai/app/api_app.py

@@ -22,6 +22,7 @@ from invokeai.app.api.routers import (
     download_queue,
     images,
     model_manager,
+    model_relationships,
     session_queue,
     style_presets,
     utilities,
@@ -125,6 +126,7 @@ app.include_router(download_queue.download_queue_router, prefix="/api")
 app.include_router(images.images_router, prefix="/api")
 app.include_router(boards.boards_router, prefix="/api")
 app.include_router(board_images.board_images_router, prefix="/api")
+app.include_router(model_relationships.model_relationships_router, prefix="/api")
 app.include_router(app_info.app_router, prefix="/api")
 app.include_router(session_queue.session_queue_router, prefix="/api")
 app.include_router(workflows.workflows_router, prefix="/api")
@@ -156,7 +158,7 @@ web_root_path = Path(list(web_dir.__path__)[0])
 try:
     app.mount("/", NoCacheStaticFiles(directory=Path(web_root_path, "dist"), html=True), name="ui")
 except RuntimeError:
-    logger.warn(f"No UI found at {web_root_path}/dist, skipping UI mount")
+    logger.warning(f"No UI found at {web_root_path}/dist, skipping UI mount")
 app.mount(
     "/static", NoCacheStaticFiles(directory=Path(web_root_path, "static/")), name="static"
 )  # docs favicon is in here

invokeai/app/invocations/baseinvocation.py

@@ -5,9 +5,12 @@ from __future__ import annotations
 import inspect
 import re
 import sys
+import types
+import typing
 import warnings
 from abc import ABC, abstractmethod
 from enum import Enum
+from functools import lru_cache
 from inspect import signature
 from typing import (
     TYPE_CHECKING,
@@ -19,15 +22,16 @@ from typing import (
     Literal,
     Optional,
     Type,
+    TypedDict,
     TypeVar,
     Union,
+    cast,
 )
 
 import semver
-from pydantic import BaseModel, ConfigDict, Field, TypeAdapter, create_model
+from pydantic import BaseModel, ConfigDict, Field, JsonValue, TypeAdapter, create_model
 from pydantic.fields import FieldInfo
 from pydantic_core import PydanticUndefined
-from typing_extensions import TypeAliasType
 
 from invokeai.app.invocations.fields import (
     FieldKind,
@@ -72,13 +76,24 @@ class Classification(str, Enum, metaclass=MetaEnum):
     Special = "special"
 
 
+class Bottleneck(str, Enum, metaclass=MetaEnum):
+    """
+    The bottleneck of an invocation.
+    - `Network`: The invocation's execution is network-bound.
+    - `GPU`: The invocation's execution is GPU-bound.
+    """
+
+    Network = "network"
+    GPU = "gpu"
+
+
 class UIConfigBase(BaseModel):
     """
     Provides additional node configuration to the UI.
     This is used internally by the @invocation decorator logic. Do not use this directly.
     """
 
-    tags: Optional[list[str]] = Field(default_factory=None, description="The node's tags")
+    tags: Optional[list[str]] = Field(default=None, description="The node's tags")
     title: Optional[str] = Field(default=None, description="The node's display name")
     category: Optional[str] = Field(default=None, description="The node's category")
     version: str = Field(
@@ -93,6 +108,11 @@ class UIConfigBase(BaseModel):
     )
 
 
+class OriginalModelField(TypedDict):
+    annotation: Any
+    field_info: FieldInfo
+
+
 class BaseInvocationOutput(BaseModel):
     """
     Base class for all invocation outputs.
@@ -100,36 +120,11 @@ class BaseInvocationOutput(BaseModel):
     All invocation outputs must use the `@invocation_output` decorator to provide their unique type.
     """
 
-    _output_classes: ClassVar[set[BaseInvocationOutput]] = set()
-    _typeadapter: ClassVar[Optional[TypeAdapter[Any]]] = None
-    _typeadapter_needs_update: ClassVar[bool] = False
-
-    @classmethod
-    def register_output(cls, output: BaseInvocationOutput) -> None:
-        """Registers an invocation output."""
-        cls._output_classes.add(output)
-        cls._typeadapter_needs_update = True
-
-    @classmethod
-    def get_outputs(cls) -> Iterable[BaseInvocationOutput]:
-        """Gets all invocation outputs."""
-        return cls._output_classes
-
-    @classmethod
-    def get_typeadapter(cls) -> TypeAdapter[Any]:
-        """Gets a pydantc TypeAdapter for the union of all invocation output types."""
-        if not cls._typeadapter or cls._typeadapter_needs_update:
-            AnyInvocationOutput = TypeAliasType(
-                "AnyInvocationOutput", Annotated[Union[tuple(cls._output_classes)], Field(discriminator="type")]
-            )
-            cls._typeadapter = TypeAdapter(AnyInvocationOutput)
-            cls._typeadapter_needs_update = False
-        return cls._typeadapter
-
-    @classmethod
-    def get_output_types(cls) -> Iterable[str]:
-        """Gets all invocation output types."""
-        return (i.get_type() for i in BaseInvocationOutput.get_outputs())
+    output_meta: Optional[dict[str, JsonValue]] = Field(
+        default=None,
+        description="Optional dictionary of metadata for the invocation output, unrelated to the invocation's actual output value. This is not exposed as an output field.",
+        json_schema_extra={"field_kind": FieldKind.NodeAttribute},
+    )
 
     @staticmethod
     def json_schema_extra(schema: dict[str, Any], model_class: Type[BaseInvocationOutput]) -> None:
@@ -146,6 +141,9 @@ class BaseInvocationOutput(BaseModel):
         """Gets the invocation output's type, as provided by the `@invocation_output` decorator."""
         return cls.model_fields["type"].default
 
+    _original_model_fields: ClassVar[dict[str, OriginalModelField]] = {}
+    """The original model fields, before any modifications were made by the @invocation_output decorator."""
+
     model_config = ConfigDict(
         protected_namespaces=(),
         validate_assignment=True,
@@ -173,76 +171,16 @@ class BaseInvocation(ABC, BaseModel):
     All invocations must use the `@invocation` decorator to provide their unique type.
     """
 
-    _invocation_classes: ClassVar[set[BaseInvocation]] = set()
-    _typeadapter: ClassVar[Optional[TypeAdapter[Any]]] = None
-    _typeadapter_needs_update: ClassVar[bool] = False
-
     @classmethod
     def get_type(cls) -> str:
         """Gets the invocation's type, as provided by the `@invocation` decorator."""
         return cls.model_fields["type"].default
 
     @classmethod
-    def register_invocation(cls, invocation: BaseInvocation) -> None:
-        """Registers an invocation."""
-        cls._invocation_classes.add(invocation)
-        cls._typeadapter_needs_update = True
-
-    @classmethod
-    def get_typeadapter(cls) -> TypeAdapter[Any]:
-        """Gets a pydantc TypeAdapter for the union of all invocation types."""
-        if not cls._typeadapter or cls._typeadapter_needs_update:
-            AnyInvocation = TypeAliasType(
-                "AnyInvocation", Annotated[Union[tuple(cls.get_invocations())], Field(discriminator="type")]
-            )
-            cls._typeadapter = TypeAdapter(AnyInvocation)
-            cls._typeadapter_needs_update = False
-        return cls._typeadapter
-
-    @classmethod
-    def invalidate_typeadapter(cls) -> None:
-        """Invalidates the typeadapter, forcing it to be rebuilt on next access. If the invocation allowlist or
-        denylist is changed, this should be called to ensure the typeadapter is updated and validation respects
-        the updated allowlist and denylist."""
-        cls._typeadapter_needs_update = True
-
-    @classmethod
-    def get_invocations(cls) -> Iterable[BaseInvocation]:
-        """Gets all invocations, respecting the allowlist and denylist."""
-        app_config = get_config()
-        allowed_invocations: set[BaseInvocation] = set()
-        for sc in cls._invocation_classes:
-            invocation_type = sc.get_type()
-            is_in_allowlist = (
-                invocation_type in app_config.allow_nodes if isinstance(app_config.allow_nodes, list) else True
-            )
-            is_in_denylist = (
-                invocation_type in app_config.deny_nodes if isinstance(app_config.deny_nodes, list) else False
-            )
-            if is_in_allowlist and not is_in_denylist:
-                allowed_invocations.add(sc)
-        return allowed_invocations
-
-    @classmethod
-    def get_invocations_map(cls) -> dict[str, BaseInvocation]:
-        """Gets a map of all invocation types to their invocation classes."""
-        return {i.get_type(): i for i in BaseInvocation.get_invocations()}
-
-    @classmethod
-    def get_invocation_types(cls) -> Iterable[str]:
-        """Gets all invocation types."""
-        return (i.get_type() for i in BaseInvocation.get_invocations())
-
-    @classmethod
-    def get_output_annotation(cls) -> BaseInvocationOutput:
+    def get_output_annotation(cls) -> Type[BaseInvocationOutput]:
         """Gets the invocation's output annotation (i.e. the return annotation of its `invoke()` method)."""
         return signature(cls.invoke).return_annotation
 
-    @classmethod
-    def get_invocation_for_type(cls, invocation_type: str) -> BaseInvocation | None:
-        """Gets the invocation class for a given invocation type."""
-        return cls.get_invocations_map().get(invocation_type)
-
     @staticmethod
     def json_schema_extra(schema: dict[str, Any], model_class: Type[BaseInvocation]) -> None:
         """Adds various UI-facing attributes to the invocation's OpenAPI schema."""
@@ -271,7 +209,7 @@ class BaseInvocation(ABC, BaseModel):
         Internal invoke method, calls `invoke()` after some prep.
         Handles optional fields that are required to call `invoke()` and invocation cache.
         """
-        for field_name, field in self.model_fields.items():
+        for field_name, field in type(self).model_fields.items():
             if not field.json_schema_extra or callable(field.json_schema_extra):
                 # something has gone terribly awry, we should always have this and it should be a dict
                 continue
@@ -286,9 +224,9 @@ class BaseInvocation(ABC, BaseModel):
                 setattr(self, field_name, orig_default)
             if orig_required and orig_default is PydanticUndefined and getattr(self, field_name) is None:
                 if input_ == Input.Connection:
-                    raise RequiredConnectionException(self.model_fields["type"].default, field_name)
+                    raise RequiredConnectionException(type(self).model_fields["type"].default, field_name)
                 elif input_ == Input.Any:
-                    raise MissingInputException(self.model_fields["type"].default, field_name)
+                    raise MissingInputException(type(self).model_fields["type"].default, field_name)
 
         # skip node cache codepath if it's disabled
         if services.configuration.node_cache_size == 0:
@@ -326,6 +264,8 @@ class BaseInvocation(ABC, BaseModel):
         json_schema_extra={"field_kind": FieldKind.NodeAttribute},
     )
 
+    bottleneck: ClassVar[Bottleneck]
+
     UIConfig: ClassVar[UIConfigBase]
 
     model_config = ConfigDict(
@@ -336,21 +276,163 @@ class BaseInvocation(ABC, BaseModel):
         coerce_numbers_to_str=True,
     )
 
+    _original_model_fields: ClassVar[dict[str, OriginalModelField]] = {}
+    """The original model fields, before any modifications were made by the @invocation decorator."""
+
 
 TBaseInvocation = TypeVar("TBaseInvocation", bound=BaseInvocation)
 
 
+class InvocationRegistry:
+    _invocation_classes: ClassVar[set[type[BaseInvocation]]] = set()
+    _output_classes: ClassVar[set[type[BaseInvocationOutput]]] = set()
+
+    @classmethod
+    def register_invocation(cls, invocation: type[BaseInvocation]) -> None:
+        """Registers an invocation."""
+
+        invocation_type = invocation.get_type()
+        node_pack = invocation.UIConfig.node_pack
+
+        # Log a warning when an existing invocation is being clobbered by the one we are registering
+        clobbered_invocation = InvocationRegistry.get_invocation_for_type(invocation_type)
+        if clobbered_invocation is not None:
+            # This should always be true - we just checked if the invocation type was in the set
+            clobbered_node_pack = clobbered_invocation.UIConfig.node_pack
+
+            if clobbered_node_pack == "invokeai":
+                # The invocation being clobbered is a core invocation
+                logger.warning(f'Overriding core node "{invocation_type}" with node from "{node_pack}"')
+            else:
+                # The invocation being clobbered is a custom invocation
+                logger.warning(
+                    f'Overriding node "{invocation_type}" from "{node_pack}" with node from "{clobbered_node_pack}"'
+                )
+            cls._invocation_classes.remove(clobbered_invocation)
+
+        cls._invocation_classes.add(invocation)
+        cls.invalidate_invocation_typeadapter()
+
+    @classmethod
+    @lru_cache(maxsize=1)
+    def get_invocation_typeadapter(cls) -> TypeAdapter[Any]:
+        """Gets a pydantic TypeAdapter for the union of all invocation types.
+
+        This is used to parse serialized invocations into the correct invocation class.
+
+        This method is cached to avoid rebuilding the TypeAdapter on every access. If the invocation allowlist or
+        denylist is changed, the cache should be cleared to ensure the TypeAdapter is updated and validation respects
+        the updated allowlist and denylist.
+
+        @see https://docs.pydantic.dev/latest/concepts/type_adapter/
+        """
+        return TypeAdapter(Annotated[Union[tuple(cls.get_invocation_classes())], Field(discriminator="type")])
+
+    @classmethod
+    def invalidate_invocation_typeadapter(cls) -> None:
+        """Invalidates the cached invocation type adapter."""
+        cls.get_invocation_typeadapter.cache_clear()
+
+    @classmethod
+    def get_invocation_classes(cls) -> Iterable[type[BaseInvocation]]:
+        """Gets all invocations, respecting the allowlist and denylist."""
+        app_config = get_config()
+        allowed_invocations: set[type[BaseInvocation]] = set()
+        for sc in cls._invocation_classes:
+            invocation_type = sc.get_type()
+            is_in_allowlist = (
+                invocation_type in app_config.allow_nodes if isinstance(app_config.allow_nodes, list) else True
+            )
+            is_in_denylist = (
+                invocation_type in app_config.deny_nodes if isinstance(app_config.deny_nodes, list) else False
+            )
+            if is_in_allowlist and not is_in_denylist:
+                allowed_invocations.add(sc)
+        return allowed_invocations
+
+    @classmethod
+    def get_invocations_map(cls) -> dict[str, type[BaseInvocation]]:
+        """Gets a map of all invocation types to their invocation classes."""
+        return {i.get_type(): i for i in cls.get_invocation_classes()}
+
+    @classmethod
+    def get_invocation_types(cls) -> Iterable[str]:
+        """Gets all invocation types."""
+        return (i.get_type() for i in cls.get_invocation_classes())
+
+    @classmethod
+    def get_invocation_for_type(cls, invocation_type: str) -> type[BaseInvocation] | None:
+        """Gets the invocation class for a given invocation type."""
+        return cls.get_invocations_map().get(invocation_type)
+
+    @classmethod
+    def register_output(cls, output: "type[TBaseInvocationOutput]") -> None:
+        """Registers an invocation output."""
+        output_type = output.get_type()
+
+        # Log a warning when an existing invocation is being clobbered by the one we are registering
+        clobbered_output = InvocationRegistry.get_output_for_type(output_type)
+        if clobbered_output is not None:
+            # TODO(psyche): We do not record the node pack of the output, so we cannot log it here
+            logger.warning(f'Overriding invocation output "{output_type}"')
+            cls._output_classes.remove(clobbered_output)
+
+        cls._output_classes.add(output)
+        cls.invalidate_output_typeadapter()
+
+    @classmethod
+    def get_output_classes(cls) -> Iterable[type[BaseInvocationOutput]]:
+        """Gets all invocation outputs."""
+        return cls._output_classes
+
+    @classmethod
+    def get_outputs_map(cls) -> dict[str, type[BaseInvocationOutput]]:
+        """Gets a map of all output types to their output classes."""
+        return {i.get_type(): i for i in cls.get_output_classes()}
+
+    @classmethod
+    @lru_cache(maxsize=1)
+    def get_output_typeadapter(cls) -> TypeAdapter[Any]:
+        """Gets a pydantic TypeAdapter for the union of all invocation output types.
+
+        This is used to parse serialized invocation outputs into the correct invocation output class.
+
+        This method is cached to avoid rebuilding the TypeAdapter on every access. If the invocation allowlist or
+        denylist is changed, the cache should be cleared to ensure the TypeAdapter is updated and validation respects
+        the updated allowlist and denylist.
+
+        @see https://docs.pydantic.dev/latest/concepts/type_adapter/
+        """
+        return TypeAdapter(Annotated[Union[tuple(cls._output_classes)], Field(discriminator="type")])
+
+    @classmethod
+    def invalidate_output_typeadapter(cls) -> None:
+        """Invalidates the cached invocation output type adapter."""
+        cls.get_output_typeadapter.cache_clear()
+
+    @classmethod
+    def get_output_types(cls) -> Iterable[str]:
+        """Gets all invocation output types."""
+        return (i.get_type() for i in cls.get_output_classes())
+
+    @classmethod
+    def get_output_for_type(cls, output_type: str) -> type[BaseInvocationOutput] | None:
+        """Gets the output class for a given output type."""
+        return cls.get_outputs_map().get(output_type)
+
+
 RESERVED_NODE_ATTRIBUTE_FIELD_NAMES = {
     "id",
     "is_intermediate",
     "use_cache",
     "type",
     "workflow",
+    "bottleneck",
 }
 
 RESERVED_INPUT_FIELD_NAMES = {"metadata", "board"}
 
-RESERVED_OUTPUT_FIELD_NAMES = {"type"}
+RESERVED_OUTPUT_FIELD_NAMES = {"type", "output_meta"}
 
 
 class _Model(BaseModel):
@@ -417,11 +499,53 @@ def validate_fields(model_fields: dict[str, FieldInfo], model_type: str) -> None
 
         ui_type = field.json_schema_extra.get("ui_type", None)
         if isinstance(ui_type, str) and ui_type.startswith("DEPRECATED_"):
-            logger.warn(f'"UIType.{ui_type.split("_")[-1]}" is deprecated, ignoring')
+            logger.warning(f'"UIType.{ui_type.split("_")[-1]}" is deprecated, ignoring')
             field.json_schema_extra.pop("ui_type")
     return None
 
 
+class NoDefaultSentinel:
+    pass
+
+
+def validate_field_default(
+    cls_name: str, field_name: str, invocation_type: str, annotation: Any, field_info: FieldInfo
+) -> None:
+    """Validates the default value of a field against its pydantic field definition."""
+
+    assert isinstance(field_info.json_schema_extra, dict), "json_schema_extra is not a dict"
+
+    # By the time we are doing this, we've already done some pydantic magic by overriding the original default value.
+    # We store the original default value in the json_schema_extra dict, so we can validate it here.
+    orig_default = field_info.json_schema_extra.get("orig_default", NoDefaultSentinel)
+
+    if orig_default is NoDefaultSentinel:
+        return
+
+    # To validate the default value, we can create a temporary pydantic model with the field we are validating as its
+    # only field. Then validate the default value against this temporary model.
+    TempDefaultValidator = cast(BaseModel, create_model(cls_name, **{field_name: (annotation, field_info)}))
+
+    try:
+        TempDefaultValidator.model_validate({field_name: orig_default})
+    except Exception as e:
+        raise InvalidFieldError(
+            f'Default value for field "{field_name}" on invocation "{invocation_type}" is invalid, {e}'
+        ) from e
+
+
+def is_optional(annotation: Any) -> bool:
+    """
+    Checks if the given annotation is optional (i.e. Optional[X], Union[X, None] or X | None).
+    """
+    origin = typing.get_origin(annotation)
+    # PEP 604 unions (int|None) have origin types.UnionType
+    is_union = origin is typing.Union or origin is types.UnionType
+    if not is_union:
+        return False
+    return any(arg is type(None) for arg in typing.get_args(annotation))
+
+
 def invocation(
     invocation_type: str,
     title: Optional[str] = None,
@@ -430,6 +554,7 @@ def invocation(
     version: Optional[str] = None,
     use_cache: Optional[bool] = True,
     classification: Classification = Classification.Stable,
+    bottleneck: Bottleneck = Bottleneck.GPU,
 ) -> Callable[[Type[TBaseInvocation]], Type[TBaseInvocation]]:
     """
     Registers an invocation.
@@ -441,6 +566,7 @@ def invocation(
     :param Optional[str] version: Adds a version to the invocation. Must be a valid semver string. Defaults to None.
     :param Optional[bool] use_cache: Whether or not to use the invocation cache. Defaults to True. The user may override this in the workflow editor.
     :param Classification classification: The classification of the invocation. Defaults to FeatureClassification.Stable. Use Beta or Prototype if the invocation is unstable.
+    :param Bottleneck bottleneck: The bottleneck of the invocation. Defaults to Bottleneck.GPU. Use Network if the invocation is network-bound.
     """
 
     def wrapper(cls: Type[TBaseInvocation]) -> Type[TBaseInvocation]:
@@ -452,27 +578,26 @@ def invocation(
         # The node pack is the module name - will be "invokeai" for built-in nodes
         node_pack = cls.__module__.split(".")[0]
 
-        # Handle the case where an existing node is being clobbered by the one we are registering
-        if invocation_type in BaseInvocation.get_invocation_types():
-            clobbered_invocation = BaseInvocation.get_invocation_for_type(invocation_type)
-            # This should always be true - we just checked if the invocation type was in the set
-            assert clobbered_invocation is not None
-
-            clobbered_node_pack = clobbered_invocation.UIConfig.node_pack
-
-            if clobbered_node_pack == "invokeai":
-                # The node being clobbered is a core node
-                raise ValueError(
-                    f'Cannot load node "{invocation_type}" from node pack "{node_pack}" - a core node with the same type already exists'
-                )
-            else:
-                # The node being clobbered is a custom node
-                raise ValueError(
-                    f'Cannot load node "{invocation_type}" from node pack "{node_pack}" - a node with the same type already exists in node pack "{clobbered_node_pack}"'
-                )
-
         validate_fields(cls.model_fields, invocation_type)
 
+        fields: dict[str, tuple[Any, FieldInfo]] = {}
+
+        for field_name, field_info in cls.model_fields.items():
+            annotation = field_info.annotation
+            assert annotation is not None, f"{field_name} on invocation {invocation_type} has no type annotation."
+            assert isinstance(field_info.json_schema_extra, dict), (
+                f"{field_name} on invocation {invocation_type} has a non-dict json_schema_extra, did you forget to use InputField?"
+            )
+
+            cls._original_model_fields[field_name] = OriginalModelField(annotation=annotation, field_info=field_info)
+
+            validate_field_default(cls.__name__, field_name, invocation_type, annotation, field_info)
+
+            if field_info.default is None and not is_optional(annotation):
+                annotation = annotation | None
+
+            fields[field_name] = (annotation, field_info)
+
         # Add OpenAPI schema extras
         uiconfig: dict[str, Any] = {}
         uiconfig["title"] = title
@@ -488,7 +613,7 @@ def invocation(
                 raise InvalidVersionError(f'Invalid version string for node "{invocation_type}": "{version}"') from e
             uiconfig["version"] = version
         else:
-            logger.warn(f'No version specified for node "{invocation_type}", using "1.0.0"')
+            logger.warning(f'No version specified for node "{invocation_type}", using "1.0.0"')
             uiconfig["version"] = "1.0.0"
 
         cls.UIConfig = UIConfigBase(**uiconfig)
@@ -496,6 +621,8 @@ def invocation(
         if use_cache is not None:
             cls.model_fields["use_cache"].default = use_cache
 
+        cls.bottleneck = bottleneck
+
         # Add the invocation type to the model.
 
         # You'd be tempted to just add the type field and rebuild the model, like this:
@@ -505,11 +632,27 @@ def invocation(
         # Unfortunately, because the `GraphInvocation` uses a forward ref in its `graph` field's annotation, this does
         # not work. Instead, we have to create a new class with the type field and patch the original class with it.
 
-        invocation_type_annotation = Literal[invocation_type]  # type: ignore
-        invocation_type_field = Field(
-            title="type", default=invocation_type, json_schema_extra={"field_kind": FieldKind.NodeAttribute}
+        invocation_type_annotation = Literal[invocation_type]
+
+        # Field() returns an instance of FieldInfo, but thanks to a pydantic implementation detail, it is _typed_ as Any.
+        # This cast makes the type annotation match the class's true type.
+        invocation_type_field_info = cast(
+            FieldInfo,
+            Field(title="type", default=invocation_type, json_schema_extra={"field_kind": FieldKind.NodeAttribute}),
         )
 
+        fields["type"] = (invocation_type_annotation, invocation_type_field_info)
+
+        # Invocation outputs must be registered using the @invocation_output decorator, but it is possible that the
+        # output is registered _after_ this invocation is registered. It depends on module import ordering.
+        #
+        # We can only confirm the output for an invocation is registered after all modules are imported. There's
+        # only really one good time to do that - during application startup, in `run_app.py`, after loading all
+        # custom nodes.
+        #
+        # We can still do some basic validation here - ensure the invoke method is defined and returns an instance
+        # of BaseInvocationOutput.
+
         # Validate the `invoke()` method is implemented
         if "invoke" in cls.__abstractmethods__:
             raise ValueError(f'Invocation "{invocation_type}" must implement the "invoke" method')
@@ -531,18 +674,12 @@ def invocation(
             )
 
         docstring = cls.__doc__
-        cls = create_model(
-            cls.__qualname__,
-            __base__=cls,
-            __module__=cls.__module__,
-            type=(invocation_type_annotation, invocation_type_field),
-        )
-        cls.__doc__ = docstring
+        new_class = create_model(cls.__qualname__, __base__=cls, __module__=cls.__module__, **fields)  # type: ignore
+        new_class.__doc__ = docstring
 
-        # TODO: how to type this correctly? it's typed as ModelMetaclass, a private class in pydantic
-        BaseInvocation.register_invocation(cls)  # type: ignore
+        InvocationRegistry.register_invocation(new_class)
 
-        return cls
+        return new_class
 
     return wrapper
 
@@ -565,29 +702,41 @@ def invocation_output(
         if re.compile(r"^\S+$").match(output_type) is None:
             raise ValueError(f'"output_type" must consist of non-whitespace characters, got "{output_type}"')
 
-        if output_type in BaseInvocationOutput.get_output_types():
-            raise ValueError(f'Invocation type "{output_type}" already exists')
-
         validate_fields(cls.model_fields, output_type)
 
-        # Add the output type to the model.
+        fields: dict[str, tuple[Any, FieldInfo]] = {}
 
-        output_type_annotation = Literal[output_type]  # type: ignore
-        output_type_field = Field(
-            title="type", default=output_type, json_schema_extra={"field_kind": FieldKind.NodeAttribute}
+        for field_name, field_info in cls.model_fields.items():
+            annotation = field_info.annotation
+            assert annotation is not None, f"{field_name} on invocation output {output_type} has no type annotation."
+            assert isinstance(field_info.json_schema_extra, dict), (
+                f"{field_name} on invocation output {output_type} has a non-dict json_schema_extra, did you forget to use InputField?"
+            )
+
+            cls._original_model_fields[field_name] = OriginalModelField(annotation=annotation, field_info=field_info)
+
+            if field_info.default is not PydanticUndefined and is_optional(annotation):
+                annotation = annotation | None
+            fields[field_name] = (annotation, field_info)
+
+        # Add the output type to the model.
+        output_type_annotation = Literal[output_type]
+
+        # Field() returns an instance of FieldInfo, but thanks to a pydantic implementation detail, it is _typed_ as Any.
+        # This cast makes the type annotation match the class's true type.
+        output_type_field_info = cast(
+            FieldInfo,
+            Field(title="type", default=output_type, json_schema_extra={"field_kind": FieldKind.NodeAttribute}),
         )
 
+        fields["type"] = (output_type_annotation, output_type_field_info)
+
         docstring = cls.__doc__
-        cls = create_model(
-            cls.__qualname__,
-            __base__=cls,
-            __module__=cls.__module__,
-            type=(output_type_annotation, output_type_field),
-        )
-        cls.__doc__ = docstring
+        new_class = create_model(cls.__qualname__, __base__=cls, __module__=cls.__module__, **fields)
+        new_class.__doc__ = docstring
 
-        BaseInvocationOutput.register_output(cls)  # type: ignore # TODO: how to type this correctly?
+        InvocationRegistry.register_output(new_class)
 
-        return cls
+        return new_class
 
     return wrapper

invokeai/app/invocations/batch.py

@@ -64,7 +64,6 @@ class ImageBatchInvocation(BaseBatchInvocation):
     """Create a batched generation, where the workflow is executed once for each image in the batch."""
 
     images: list[ImageField] = InputField(
-        default=[],
         min_length=1,
         description="The images to batch over",
     )
@@ -120,7 +119,6 @@ class StringBatchInvocation(BaseBatchInvocation):
     """Create a batched generation, where the workflow is executed once for each string in the batch."""
 
     strings: list[str] = InputField(
-        default=[],
         min_length=1,
         description="The strings to batch over",
     )
@@ -176,7 +174,6 @@ class IntegerBatchInvocation(BaseBatchInvocation):
     """Create a batched generation, where the workflow is executed once for each integer in the batch."""
 
     integers: list[int] = InputField(
-        default=[],
         min_length=1,
         description="The integers to batch over",
     )
@@ -230,7 +227,6 @@ class FloatBatchInvocation(BaseBatchInvocation):
     """Create a batched generation, where the workflow is executed once for each float in the batch."""
 
     floats: list[float] = InputField(
-        default=[],
         min_length=1,
         description="The floats to batch over",
     )

invokeai/app/invocations/cogview4_denoise.py

@@ -0,0 +1,363 @@
+from typing import Callable, Optional
+
+import torch
+import torchvision.transforms as tv_transforms
+from diffusers.models.transformers.transformer_cogview4 import CogView4Transformer2DModel
+from torchvision.transforms.functional import resize as tv_resize
+from tqdm import tqdm
+
+from invokeai.app.invocations.baseinvocation import BaseInvocation, Classification, invocation
+from invokeai.app.invocations.constants import LATENT_SCALE_FACTOR
+from invokeai.app.invocations.fields import (
+    CogView4ConditioningField,
+    DenoiseMaskField,
+    FieldDescriptions,
+    Input,
+    InputField,
+    LatentsField,
+    WithBoard,
+    WithMetadata,
+)
+from invokeai.app.invocations.model import TransformerField
+from invokeai.app.invocations.primitives import LatentsOutput
+from invokeai.app.services.shared.invocation_context import InvocationContext
+from invokeai.backend.flux.sampling_utils import clip_timestep_schedule_fractional
+from invokeai.backend.model_manager.config import BaseModelType
+from invokeai.backend.rectified_flow.rectified_flow_inpaint_extension import RectifiedFlowInpaintExtension
+from invokeai.backend.stable_diffusion.diffusers_pipeline import PipelineIntermediateState
+from invokeai.backend.stable_diffusion.diffusion.conditioning_data import CogView4ConditioningInfo
+from invokeai.backend.util.devices import TorchDevice
+
+
+@invocation(
+    "cogview4_denoise",
+    title="Denoise - CogView4",
+    tags=["image", "cogview4"],
+    category="image",
+    version="1.0.0",
+    classification=Classification.Prototype,
+)
+class CogView4DenoiseInvocation(BaseInvocation, WithMetadata, WithBoard):
+    """Run the denoising process with a CogView4 model."""
+
+    # If latents is provided, this means we are doing image-to-image.
+    latents: Optional[LatentsField] = InputField(
+        default=None, description=FieldDescriptions.latents, input=Input.Connection
+    )
+    # denoise_mask is used for image-to-image inpainting. Only the masked region is modified.
+    denoise_mask: Optional[DenoiseMaskField] = InputField(
+        default=None, description=FieldDescriptions.denoise_mask, input=Input.Connection
+    )
+    denoising_start: float = InputField(default=0.0, ge=0, le=1, description=FieldDescriptions.denoising_start)
+    denoising_end: float = InputField(default=1.0, ge=0, le=1, description=FieldDescriptions.denoising_end)
+    transformer: TransformerField = InputField(
+        description=FieldDescriptions.cogview4_model, input=Input.Connection, title="Transformer"
+    )
+    positive_conditioning: CogView4ConditioningField = InputField(
+        description=FieldDescriptions.positive_cond, input=Input.Connection
+    )
+    negative_conditioning: CogView4ConditioningField = InputField(
+        description=FieldDescriptions.negative_cond, input=Input.Connection
+    )
+    cfg_scale: float | list[float] = InputField(default=3.5, description=FieldDescriptions.cfg_scale, title="CFG Scale")
+    width: int = InputField(default=1024, multiple_of=32, description="Width of the generated image.")
+    height: int = InputField(default=1024, multiple_of=32, description="Height of the generated image.")
+    steps: int = InputField(default=25, gt=0, description=FieldDescriptions.steps)
+    seed: int = InputField(default=0, description="Randomness seed for reproducibility.")
+
+    @torch.no_grad()
+    def invoke(self, context: InvocationContext) -> LatentsOutput:
+        latents = self._run_diffusion(context)
+        latents = latents.detach().to("cpu")
+
+        name = context.tensors.save(tensor=latents)
+        return LatentsOutput.build(latents_name=name, latents=latents, seed=None)
+
+    def _prep_inpaint_mask(self, context: InvocationContext, latents: torch.Tensor) -> torch.Tensor | None:
+        """Prepare the inpaint mask.
+        - Loads the mask
+        - Resizes if necessary
+        - Casts to same device/dtype as latents
+
+        Args:
+            context (InvocationContext): The invocation context, for loading the inpaint mask.
+            latents (torch.Tensor): A latent image tensor. Used to determine the target shape, device, and dtype for the
+                inpaint mask.
+
+        Returns:
+            torch.Tensor | None: Inpaint mask. Values of 0.0 represent the regions to be fully denoised, and 1.0
+                represent the regions to be preserved.
+        """
+        if self.denoise_mask is None:
+            return None
+        mask = context.tensors.load(self.denoise_mask.mask_name)
+
+        # The input denoise_mask contains values in [0, 1], where 0.0 represents the regions to be fully denoised, and
+        # 1.0 represents the regions to be preserved.
+        # We invert the mask so that the regions to be preserved are 0.0 and the regions to be denoised are 1.0.
+        mask = 1.0 - mask
+
+        _, _, latent_height, latent_width = latents.shape
+        mask = tv_resize(
+            img=mask,
+            size=[latent_height, latent_width],
+            interpolation=tv_transforms.InterpolationMode.BILINEAR,
+            antialias=False,
+        )
+
+        mask = mask.to(device=latents.device, dtype=latents.dtype)
+        return mask
+
+    def _load_text_conditioning(
+        self,
+        context: InvocationContext,
+        conditioning_name: str,
+        dtype: torch.dtype,
+        device: torch.device,
+    ) -> torch.Tensor:
+        # Load the conditioning data.
+        cond_data = context.conditioning.load(conditioning_name)
+        assert len(cond_data.conditionings) == 1
+        cogview4_conditioning = cond_data.conditionings[0]
+        assert isinstance(cogview4_conditioning, CogView4ConditioningInfo)
+        cogview4_conditioning = cogview4_conditioning.to(dtype=dtype, device=device)
+
+        return cogview4_conditioning.glm_embeds
+
+    def _get_noise(
+        self,
+        batch_size: int,
+        num_channels_latents: int,
+        height: int,
+        width: int,
+        dtype: torch.dtype,
+        device: torch.device,
+        seed: int,
+    ) -> torch.Tensor:
+        # We always generate noise on the same device and dtype then cast to ensure consistency across devices/dtypes.
+        rand_device = "cpu"
+        rand_dtype = torch.float16
+
+        return torch.randn(
+            batch_size,
+            num_channels_latents,
+            int(height) // LATENT_SCALE_FACTOR,
+            int(width) // LATENT_SCALE_FACTOR,
+            device=rand_device,
+            dtype=rand_dtype,
+            generator=torch.Generator(device=rand_device).manual_seed(seed),
+        ).to(device=device, dtype=dtype)
+
+    def _prepare_cfg_scale(self, num_timesteps: int) -> list[float]:
+        """Prepare the CFG scale list.
+
+        Args:
+            num_timesteps (int): The number of timesteps in the scheduler. Could be different from num_steps depending
+            on the scheduler used (e.g. higher order schedulers).
+
+        Returns:
+            list[float]: _description_
+        """
+        if isinstance(self.cfg_scale, float):
+            cfg_scale = [self.cfg_scale] * num_timesteps
+        elif isinstance(self.cfg_scale, list):
+            assert len(self.cfg_scale) == num_timesteps
+            cfg_scale = self.cfg_scale
+        else:
+            raise ValueError(f"Invalid CFG scale type: {type(self.cfg_scale)}")
+
+        return cfg_scale
+
+    def _convert_timesteps_to_sigmas(self, image_seq_len: int, timesteps: torch.Tensor) -> list[float]:
+        # The logic to prepare the timestep / sigma schedule is based on:
+        # https://github.com/huggingface/diffusers/blob/b38450d5d2e5b87d5ff7088ee5798c85587b9635/src/diffusers/pipelines/cogview4/pipeline_cogview4.py#L575-L595
+        # The default FlowMatchEulerDiscreteScheduler configs are based on:
+        # https://huggingface.co/THUDM/CogView4-6B/blob/fb6f57289c73ac6d139e8d81bd5a4602d1877847/scheduler/scheduler_config.json
+        # This implementation differs slightly from the original for the sake of simplicity (differs in terminal value
+        # handling, not quantizing timesteps to integers, etc.).
+
+        def calculate_timestep_shift(
+            image_seq_len: int, base_seq_len: int = 256, base_shift: float = 0.25, max_shift: float = 0.75
+        ) -> float:
+            m = (image_seq_len / base_seq_len) ** 0.5
+            mu = m * max_shift + base_shift
+            return mu
+
+        def time_shift_linear(mu: float, sigma: float, t: torch.Tensor) -> torch.Tensor:
+            return mu / (mu + (1 / t - 1) ** sigma)
+
+        mu = calculate_timestep_shift(image_seq_len)
+        sigmas = time_shift_linear(mu, 1.0, timesteps)
+        return sigmas.tolist()
+
+    def _run_diffusion(
+        self,
+        context: InvocationContext,
+    ):
+        inference_dtype = torch.bfloat16
+        device = TorchDevice.choose_torch_device()
+
+        transformer_info = context.models.load(self.transformer.transformer)
+        assert isinstance(transformer_info.model, CogView4Transformer2DModel)
+
+        # Load/process the conditioning data.
+        # TODO(ryand): Make CFG optional.
+        do_classifier_free_guidance = True
+        pos_prompt_embeds = self._load_text_conditioning(
+            context=context,
+            conditioning_name=self.positive_conditioning.conditioning_name,
+            dtype=inference_dtype,
+            device=device,
+        )
+        neg_prompt_embeds = self._load_text_conditioning(
+            context=context,
+            conditioning_name=self.negative_conditioning.conditioning_name,
+            dtype=inference_dtype,
+            device=device,
+        )
+
+        # Prepare misc. conditioning variables.
+        # TODO(ryand): We could expose these as params (like with SDXL). But, we should experiment to see if they are
+        # useful first.
+        original_size = torch.tensor([(self.height, self.width)], dtype=pos_prompt_embeds.dtype, device=device)
+        target_size = torch.tensor([(self.height, self.width)], dtype=pos_prompt_embeds.dtype, device=device)
+        crops_coords_top_left = torch.tensor([(0, 0)], dtype=pos_prompt_embeds.dtype, device=device)
+
+        # Prepare the timestep / sigma schedule.
+        patch_size = transformer_info.model.config.patch_size  # type: ignore
+        assert isinstance(patch_size, int)
+        image_seq_len = ((self.height // LATENT_SCALE_FACTOR) * (self.width // LATENT_SCALE_FACTOR)) // (patch_size**2)
+        # We add an extra step to the end to account for the final timestep of 0.0.
+        timesteps: list[float] = torch.linspace(1, 0, self.steps + 1).tolist()
+        # Clip the timesteps schedule based on denoising_start and denoising_end.
+        timesteps = clip_timestep_schedule_fractional(timesteps, self.denoising_start, self.denoising_end)
+        sigmas = self._convert_timesteps_to_sigmas(image_seq_len, torch.tensor(timesteps))
+        total_steps = len(timesteps) - 1
+
+        # Prepare the CFG scale list.
+        cfg_scale = self._prepare_cfg_scale(total_steps)
+
+        # Load the input latents, if provided.
+        init_latents = context.tensors.load(self.latents.latents_name) if self.latents else None
+        if init_latents is not None:
+            init_latents = init_latents.to(device=device, dtype=inference_dtype)
+
+        # Generate initial latent noise.
+        num_channels_latents = transformer_info.model.config.in_channels  # type: ignore
+        assert isinstance(num_channels_latents, int)
+        noise = self._get_noise(
+            batch_size=1,
+            num_channels_latents=num_channels_latents,
+            height=self.height,
+            width=self.width,
+            dtype=inference_dtype,
+            device=device,
+            seed=self.seed,
+        )
+
+        # Prepare input latent image.
+        if init_latents is not None:
+            # Noise the init_latents by the appropriate amount for the first timestep.
+            s_0 = sigmas[0]
+            latents = s_0 * noise + (1.0 - s_0) * init_latents
+        else:
+            # init_latents are not provided, so we are not doing image-to-image (i.e. we are starting from pure noise).
+            if self.denoising_start > 1e-5:
+                raise ValueError("denoising_start should be 0 when initial latents are not provided.")
+            latents = noise
+
+        # If len(timesteps) == 1, then short-circuit. We are just noising the input latents, but not taking any
+        # denoising steps.
+        if len(timesteps) <= 1:
+            return latents
+
+        # Prepare inpaint extension.
+        inpaint_mask = self._prep_inpaint_mask(context, latents)
+        inpaint_extension: RectifiedFlowInpaintExtension | None = None
+        if inpaint_mask is not None:
+            assert init_latents is not None
+            inpaint_extension = RectifiedFlowInpaintExtension(
+                init_latents=init_latents,
+                inpaint_mask=inpaint_mask,
+                noise=noise,
+            )
+
+        step_callback = self._build_step_callback(context)
+
+        step_callback(
+            PipelineIntermediateState(
+                step=0,
+                order=1,
+                total_steps=total_steps,
+                timestep=int(timesteps[0]),
+                latents=latents,
+            ),
+        )
+
+        with transformer_info.model_on_device() as (_, transformer):
+            assert isinstance(transformer, CogView4Transformer2DModel)
+
+            # Denoising loop
+            for step_idx in tqdm(range(total_steps)):
+                t_curr = timesteps[step_idx]
+                sigma_curr = sigmas[step_idx]
+                sigma_prev = sigmas[step_idx + 1]
+
+                # Expand the timestep to match the latent model input.
+                # Multiply by 1000 to match the default FlowMatchEulerDiscreteScheduler num_train_timesteps.
+                timestep = torch.tensor([t_curr * 1000], device=device).expand(latents.shape[0])
+
+                # TODO(ryand): Support both sequential and batched CFG inference.
+                noise_pred_cond = transformer(
+                    hidden_states=latents,
+                    encoder_hidden_states=pos_prompt_embeds,
+                    timestep=timestep,
+                    original_size=original_size,
+                    target_size=target_size,
+                    crop_coords=crops_coords_top_left,
+                    return_dict=False,
+                )[0]
+
+                # Apply CFG.
+                if do_classifier_free_guidance:
+                    noise_pred_uncond = transformer(
+                        hidden_states=latents,
+                        encoder_hidden_states=neg_prompt_embeds,
+                        timestep=timestep,
+                        original_size=original_size,
+                        target_size=target_size,
+                        crop_coords=crops_coords_top_left,
+                        return_dict=False,
+                    )[0]
+
+                    noise_pred = noise_pred_uncond + cfg_scale[step_idx] * (noise_pred_cond - noise_pred_uncond)
+                else:
+                    noise_pred = noise_pred_cond
+
+                # Compute the previous noisy sample x_t -> x_t-1.
+                latents_dtype = latents.dtype
+                # TODO(ryand): Is casting to float32 necessary for precision/stability? I copied this from SD3.
+                latents = latents.to(dtype=torch.float32)
+                latents = latents + (sigma_prev - sigma_curr) * noise_pred
+                latents = latents.to(dtype=latents_dtype)
+
+                if inpaint_extension is not None:
+                    latents = inpaint_extension.merge_intermediate_latents_with_init_latents(latents, sigma_prev)
+
+                step_callback(
+                    PipelineIntermediateState(
+                        step=step_idx + 1,
+                        order=1,
+                        total_steps=total_steps,
+                        timestep=int(t_curr),
+                        latents=latents,
+                    ),
+                )
+
+        return latents
+
+    def _build_step_callback(self, context: InvocationContext) -> Callable[[PipelineIntermediateState], None]:
+        def step_callback(state: PipelineIntermediateState) -> None:
+            context.util.sd_step_callback(state, BaseModelType.CogView4)
+
+        return step_callback

invokeai/app/invocations/cogview4_image_to_latents.py

@@ -0,0 +1,69 @@
+import einops
+import torch
+from diffusers.models.autoencoders.autoencoder_kl import AutoencoderKL
+
+from invokeai.app.invocations.baseinvocation import BaseInvocation, Classification, invocation
+from invokeai.app.invocations.fields import (
+    FieldDescriptions,
+    ImageField,
+    Input,
+    InputField,
+    WithBoard,
+    WithMetadata,
+)
+from invokeai.app.invocations.model import VAEField
+from invokeai.app.invocations.primitives import LatentsOutput
+from invokeai.app.services.shared.invocation_context import InvocationContext
+from invokeai.backend.model_manager.load.load_base import LoadedModel
+from invokeai.backend.stable_diffusion.diffusers_pipeline import image_resized_to_grid_as_tensor
+from invokeai.backend.util.devices import TorchDevice
+
+# TODO(ryand): This is effectively a copy of SD3ImageToLatentsInvocation and a subset of ImageToLatentsInvocation. We
+# should refactor to avoid this duplication.
+
+
+@invocation(
+    "cogview4_i2l",
+    title="Image to Latents - CogView4",
+    tags=["image", "latents", "vae", "i2l", "cogview4"],
+    category="image",
+    version="1.0.0",
+    classification=Classification.Prototype,
+)
+class CogView4ImageToLatentsInvocation(BaseInvocation, WithMetadata, WithBoard):
+    """Generates latents from an image."""
+
+    image: ImageField = InputField(description="The image to encode.")
+    vae: VAEField = InputField(description=FieldDescriptions.vae, input=Input.Connection)
+
+    @staticmethod
+    def vae_encode(vae_info: LoadedModel, image_tensor: torch.Tensor) -> torch.Tensor:
+        with vae_info as vae:
+            assert isinstance(vae, AutoencoderKL)
+
+            vae.disable_tiling()
+
+            image_tensor = image_tensor.to(device=TorchDevice.choose_torch_device(), dtype=vae.dtype)
+            with torch.inference_mode():
+                image_tensor_dist = vae.encode(image_tensor).latent_dist
+                # TODO: Use seed to make sampling reproducible.
+                latents: torch.Tensor = image_tensor_dist.sample().to(dtype=vae.dtype)
+
+            latents = vae.config.scaling_factor * latents
+
+        return latents
+
+    @torch.no_grad()
+    def invoke(self, context: InvocationContext) -> LatentsOutput:
+        image = context.images.get_pil(self.image.image_name)
+
+        image_tensor = image_resized_to_grid_as_tensor(image.convert("RGB"))
+        if image_tensor.dim() == 3:
+            image_tensor = einops.rearrange(image_tensor, "c h w -> 1 c h w")
+
+        vae_info = context.models.load(self.vae.vae)
+        latents = self.vae_encode(vae_info=vae_info, image_tensor=image_tensor)
+
+        latents = latents.to("cpu")
+        name = context.tensors.save(tensor=latents)
+        return LatentsOutput.build(latents_name=name, latents=latents, seed=None)

invokeai/app/invocations/cogview4_latents_to_image.py

@@ -0,0 +1,86 @@
+from contextlib import nullcontext
+
+import torch
+from diffusers.models.autoencoders.autoencoder_kl import AutoencoderKL
+from einops import rearrange
+from PIL import Image
+
+from invokeai.app.invocations.baseinvocation import BaseInvocation, Classification, invocation
+from invokeai.app.invocations.constants import LATENT_SCALE_FACTOR
+from invokeai.app.invocations.fields import (
+    FieldDescriptions,
+    Input,
+    InputField,
+    LatentsField,
+    WithBoard,
+    WithMetadata,
+)
+from invokeai.app.invocations.model import VAEField
+from invokeai.app.invocations.primitives import ImageOutput
+from invokeai.app.services.shared.invocation_context import InvocationContext
+from invokeai.backend.stable_diffusion.extensions.seamless import SeamlessExt
+from invokeai.backend.util.devices import TorchDevice
+
+# TODO(ryand): This is effectively a copy of SD3LatentsToImageInvocation and a subset of LatentsToImageInvocation. We
+# should refactor to avoid this duplication.
+
+
+@invocation(
+    "cogview4_l2i",
+    title="Latents to Image - CogView4",
+    tags=["latents", "image", "vae", "l2i", "cogview4"],
+    category="latents",
+    version="1.0.0",
+    classification=Classification.Prototype,
+)
+class CogView4LatentsToImageInvocation(BaseInvocation, WithMetadata, WithBoard):
+    """Generates an image from latents."""
+
+    latents: LatentsField = InputField(description=FieldDescriptions.latents, input=Input.Connection)
+    vae: VAEField = InputField(description=FieldDescriptions.vae, input=Input.Connection)
+
+    def _estimate_working_memory(self, latents: torch.Tensor, vae: AutoencoderKL) -> int:
+        """Estimate the working memory required by the invocation in bytes."""
+        out_h = LATENT_SCALE_FACTOR * latents.shape[-2]
+        out_w = LATENT_SCALE_FACTOR * latents.shape[-1]
+        element_size = next(vae.parameters()).element_size()
+        scaling_constant = 2200  # Determined experimentally.
+        working_memory = out_h * out_w * element_size * scaling_constant
+        return int(working_memory)
+
+    @torch.no_grad()
+    def invoke(self, context: InvocationContext) -> ImageOutput:
+        latents = context.tensors.load(self.latents.latents_name)
+
+        vae_info = context.models.load(self.vae.vae)
+        assert isinstance(vae_info.model, (AutoencoderKL))
+        estimated_working_memory = self._estimate_working_memory(latents, vae_info.model)
+        with (
+            SeamlessExt.static_patch_model(vae_info.model, self.vae.seamless_axes),
+            vae_info.model_on_device(working_mem_bytes=estimated_working_memory) as (_, vae),
+        ):
+            context.util.signal_progress("Running VAE")
+            assert isinstance(vae, (AutoencoderKL))
+            latents = latents.to(TorchDevice.choose_torch_device())
+
+            vae.disable_tiling()
+
+            tiling_context = nullcontext()
+
+            # clear memory as vae decode can request a lot
+            TorchDevice.empty_cache()
+
+            with torch.inference_mode(), tiling_context:
+                # copied from diffusers pipeline
+                latents = latents / vae.config.scaling_factor
+                img = vae.decode(latents, return_dict=False)[0]
+
+            img = img.clamp(-1, 1)
+            img = rearrange(img[0], "c h w -> h w c")  # noqa: F821
+            img_pil = Image.fromarray((127.5 * (img + 1.0)).byte().cpu().numpy())
+
+        TorchDevice.empty_cache()
+
+        image_dto = context.images.save(image=img_pil)
+
+        return ImageOutput.build(image_dto)

invokeai/app/invocations/cogview4_model_loader.py

@@ -0,0 +1,55 @@
+from invokeai.app.invocations.baseinvocation import (
+    BaseInvocation,
+    BaseInvocationOutput,
+    Classification,
+    invocation,
+    invocation_output,
+)
+from invokeai.app.invocations.fields import FieldDescriptions, Input, InputField, OutputField, UIType
+from invokeai.app.invocations.model import (
+    GlmEncoderField,
+    ModelIdentifierField,
+    TransformerField,
+    VAEField,
+)
+from invokeai.app.services.shared.invocation_context import InvocationContext
+from invokeai.backend.model_manager.config import SubModelType
+
+
+@invocation_output("cogview4_model_loader_output")
+class CogView4ModelLoaderOutput(BaseInvocationOutput):
+    """CogView4 base model loader output."""
+
+    transformer: TransformerField = OutputField(description=FieldDescriptions.transformer, title="Transformer")
+    glm_encoder: GlmEncoderField = OutputField(description=FieldDescriptions.glm_encoder, title="GLM Encoder")
+    vae: VAEField = OutputField(description=FieldDescriptions.vae, title="VAE")
+
+
+@invocation(
+    "cogview4_model_loader",
+    title="Main Model - CogView4",
+    tags=["model", "cogview4"],
+    category="model",
+    version="1.0.0",
+    classification=Classification.Prototype,
+)
+class CogView4ModelLoaderInvocation(BaseInvocation):
+    """Loads a CogView4 base model, outputting its submodels."""
+
+    model: ModelIdentifierField = InputField(
+        description=FieldDescriptions.cogview4_model,
+        ui_type=UIType.CogView4MainModel,
+        input=Input.Direct,
+    )
+
+    def invoke(self, context: InvocationContext) -> CogView4ModelLoaderOutput:
+        transformer = self.model.model_copy(update={"submodel_type": SubModelType.Transformer})
+        vae = self.model.model_copy(update={"submodel_type": SubModelType.VAE})
+        glm_tokenizer = self.model.model_copy(update={"submodel_type": SubModelType.Tokenizer})
+        glm_encoder = self.model.model_copy(update={"submodel_type": SubModelType.TextEncoder})
+
+        return CogView4ModelLoaderOutput(
+            transformer=TransformerField(transformer=transformer, loras=[]),
+            glm_encoder=GlmEncoderField(tokenizer=glm_tokenizer, text_encoder=glm_encoder),
+            vae=VAEField(vae=vae),
+        )

invokeai/app/invocations/cogview4_text_encoder.py

@@ -0,0 +1,92 @@
+import torch
+from transformers import GlmModel, PreTrainedTokenizerFast
+
+from invokeai.app.invocations.baseinvocation import BaseInvocation, Classification, invocation
+from invokeai.app.invocations.fields import FieldDescriptions, Input, InputField, UIComponent
+from invokeai.app.invocations.model import GlmEncoderField
+from invokeai.app.invocations.primitives import CogView4ConditioningOutput
+from invokeai.app.services.shared.invocation_context import InvocationContext
+from invokeai.backend.stable_diffusion.diffusion.conditioning_data import (
+    CogView4ConditioningInfo,
+    ConditioningFieldData,
+)
+from invokeai.backend.util.devices import TorchDevice
+
+# The CogView4 GLM Text Encoder max sequence length set based on the default in diffusers.
+COGVIEW4_GLM_MAX_SEQ_LEN = 1024
+
+
+@invocation(
+    "cogview4_text_encoder",
+    title="Prompt - CogView4",
+    tags=["prompt", "conditioning", "cogview4"],
+    category="conditioning",
+    version="1.0.0",
+    classification=Classification.Prototype,
+)
+class CogView4TextEncoderInvocation(BaseInvocation):
+    """Encodes and preps a prompt for a cogview4 image."""
+
+    prompt: str = InputField(description="Text prompt to encode.", ui_component=UIComponent.Textarea)
+    glm_encoder: GlmEncoderField = InputField(
+        title="GLM Encoder",
+        description=FieldDescriptions.glm_encoder,
+        input=Input.Connection,
+    )
+
+    @torch.no_grad()
+    def invoke(self, context: InvocationContext) -> CogView4ConditioningOutput:
+        glm_embeds = self._glm_encode(context, max_seq_len=COGVIEW4_GLM_MAX_SEQ_LEN)
+        conditioning_data = ConditioningFieldData(conditionings=[CogView4ConditioningInfo(glm_embeds=glm_embeds)])
+        conditioning_name = context.conditioning.save(conditioning_data)
+        return CogView4ConditioningOutput.build(conditioning_name)
+
+    def _glm_encode(self, context: InvocationContext, max_seq_len: int) -> torch.Tensor:
+        prompt = [self.prompt]
+
+        # TODO(ryand): Add model inputs to the invocation rather than hard-coding.
+        with (
+            context.models.load(self.glm_encoder.text_encoder).model_on_device() as (_, glm_text_encoder),
+            context.models.load(self.glm_encoder.tokenizer).model_on_device() as (_, glm_tokenizer),
+        ):
+            context.util.signal_progress("Running GLM text encoder")
+            assert isinstance(glm_text_encoder, GlmModel)
+            assert isinstance(glm_tokenizer, PreTrainedTokenizerFast)
+
+            text_inputs = glm_tokenizer(
+                prompt,
+                padding="longest",
+                max_length=max_seq_len,
+                truncation=True,
+                add_special_tokens=True,
+                return_tensors="pt",
+            )
+            text_input_ids = text_inputs.input_ids
+            untruncated_ids = glm_tokenizer(prompt, padding="longest", return_tensors="pt").input_ids
+            assert isinstance(text_input_ids, torch.Tensor)
+            assert isinstance(untruncated_ids, torch.Tensor)
+            if untruncated_ids.shape[-1] >= text_input_ids.shape[-1] and not torch.equal(
+                text_input_ids, untruncated_ids
+            ):
+                removed_text = glm_tokenizer.batch_decode(untruncated_ids[:, max_seq_len - 1 : -1])
+                context.logger.warning(
+                    "The following part of your input was truncated because `max_sequence_length` is set to "
+                    f" {max_seq_len} tokens: {removed_text}"
+                )
+
+            current_length = text_input_ids.shape[1]
+            pad_length = (16 - (current_length % 16)) % 16
+            if pad_length > 0:
+                pad_ids = torch.full(
+                    (text_input_ids.shape[0], pad_length),
+                    fill_value=glm_tokenizer.pad_token_id,
+                    dtype=text_input_ids.dtype,
+                    device=text_input_ids.device,
+                )
+                text_input_ids = torch.cat([pad_ids, text_input_ids], dim=1)
+            prompt_embeds = glm_text_encoder(
+                text_input_ids.to(TorchDevice.choose_torch_device()), output_hidden_states=True
+            ).hidden_states[-2]
+
+        assert isinstance(prompt_embeds, torch.Tensor)
+        return prompt_embeds

invokeai/app/invocations/compel.py

@@ -1,7 +1,7 @@
 from typing import Iterator, List, Optional, Tuple, Union, cast
 
 import torch
-from compel import Compel, ReturnedEmbeddingsType
+from compel import Compel, ReturnedEmbeddingsType, SplitLongTextMode
 from compel.prompt_parser import Blend, Conjunction, CrossAttentionControlSubstitute, FlattenedPrompt, Fragment
 from transformers import CLIPTextModel, CLIPTextModelWithProjection, CLIPTokenizer
 
@@ -104,6 +104,7 @@ class CompelInvocation(BaseInvocation):
                 dtype_for_device_getter=TorchDevice.choose_torch_dtype,
                 truncate_long_prompts=False,
                 device=TorchDevice.choose_torch_device(),
+                split_long_text_mode=SplitLongTextMode.SENTENCES,
             )
 
             conjunction = Compel.parse_prompt_string(self.prompt)
@@ -113,6 +114,13 @@ class CompelInvocation(BaseInvocation):
 
             c, _options = compel.build_conditioning_tensor_for_conjunction(conjunction)
 
+        del compel
+        del patched_tokenizer
+        del tokenizer
+        del ti_manager
+        del text_encoder
+        del text_encoder_info
+
         c = c.detach().to("cpu")
 
         conditioning_data = ConditioningFieldData(conditionings=[BasicConditioningInfo(embeds=c)])
@@ -205,6 +213,7 @@ class SDXLPromptInvocationBase:
                 returned_embeddings_type=ReturnedEmbeddingsType.PENULTIMATE_HIDDEN_STATES_NON_NORMALIZED,  # TODO: clip skip
                 requires_pooled=get_pooled,
                 device=TorchDevice.choose_torch_device(),
+                split_long_text_mode=SplitLongTextMode.SENTENCES,
             )
 
             conjunction = Compel.parse_prompt_string(prompt)
@@ -220,7 +229,10 @@ class SDXLPromptInvocationBase:
             else:
                 c_pooled = None
 
+        del compel
+        del patched_tokenizer
         del tokenizer
+        del ti_manager
         del text_encoder
         del text_encoder_info
 

invokeai/app/invocations/composition-nodes.py

@@ -274,12 +274,12 @@ class InvokeAdjustImageHuePlusInvocation(BaseInvocation, WithMetadata, WithBoard
     title="Enhance Image",
     tags=["enhance", "image"],
     category="image",
-    version="1.2.0",
+    version="1.2.1",
 )
 class InvokeImageEnhanceInvocation(BaseInvocation, WithMetadata, WithBoard):
     """Applies processing from PIL's ImageEnhance module. Originally created by @dwringer"""
 
-    image: ImageField = InputField(default=None, description="The image for which to apply processing")
+    image: ImageField = InputField(description="The image for which to apply processing")
     invert: bool = InputField(default=False, description="Whether to invert the image colors")
     color: float = InputField(ge=0, default=1.0, description="Color enhancement factor")
     contrast: float = InputField(ge=0, default=1.0, description="Contrast enhancement factor")

invokeai/app/invocations/controlnet.py

@@ -0,0 +1,132 @@
+# Invocations for ControlNet image preprocessors
+# initial implementation by Gregg Helt, 2023
+from typing import List, Union
+
+from pydantic import BaseModel, Field, field_validator, model_validator
+
+from invokeai.app.invocations.baseinvocation import (
+    BaseInvocation,
+    BaseInvocationOutput,
+    Classification,
+    invocation,
+    invocation_output,
+)
+from invokeai.app.invocations.fields import (
+    FieldDescriptions,
+    ImageField,
+    InputField,
+    OutputField,
+    UIType,
+)
+from invokeai.app.invocations.model import ModelIdentifierField
+from invokeai.app.invocations.primitives import ImageOutput
+from invokeai.app.invocations.util import validate_begin_end_step, validate_weights
+from invokeai.app.services.shared.invocation_context import InvocationContext
+from invokeai.app.util.controlnet_utils import (
+    CONTROLNET_MODE_VALUES,
+    CONTROLNET_RESIZE_VALUES,
+    heuristic_resize_fast,
+)
+from invokeai.backend.image_util.util import np_to_pil, pil_to_np
+
+
+class ControlField(BaseModel):
+    image: ImageField = Field(description="The control image")
+    control_model: ModelIdentifierField = Field(description="The ControlNet model 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")
+
+    @field_validator("control_weight")
+    @classmethod
+    def validate_control_weight(cls, v):
+        validate_weights(v)
+        return v
+
+    @model_validator(mode="after")
+    def validate_begin_end_step_percent(self):
+        validate_begin_end_step(self.begin_step_percent, self.end_step_percent)
+        return self
+
+
+@invocation_output("control_output")
+class ControlOutput(BaseInvocationOutput):
+    """node output for ControlNet info"""
+
+    # Outputs
+    control: ControlField = OutputField(description=FieldDescriptions.control)
+
+
+@invocation("controlnet", title="ControlNet - SD1.5, SDXL", tags=["controlnet"], category="controlnet", version="1.1.3")
+class ControlNetInvocation(BaseInvocation):
+    """Collects ControlNet info to pass to other nodes"""
+
+    image: ImageField = InputField(description="The control image")
+    control_model: ModelIdentifierField = InputField(
+        description=FieldDescriptions.controlnet_model, ui_type=UIType.ControlNetModel
+    )
+    control_weight: Union[float, List[float]] = InputField(
+        default=1.0, ge=-1, le=2, description="The weight given to the ControlNet"
+    )
+    begin_step_percent: float = InputField(
+        default=0, ge=0, le=1, description="When the ControlNet is first applied (% of total steps)"
+    )
+    end_step_percent: float = InputField(
+        default=1, ge=0, le=1, description="When the ControlNet is last applied (% of total steps)"
+    )
+    control_mode: CONTROLNET_MODE_VALUES = InputField(default="balanced", description="The control mode used")
+    resize_mode: CONTROLNET_RESIZE_VALUES = InputField(default="just_resize", description="The resize mode used")
+
+    @field_validator("control_weight")
+    @classmethod
+    def validate_control_weight(cls, v):
+        validate_weights(v)
+        return v
+
+    @model_validator(mode="after")
+    def validate_begin_end_step_percent(self) -> "ControlNetInvocation":
+        validate_begin_end_step(self.begin_step_percent, self.end_step_percent)
+        return self
+
+    def invoke(self, context: InvocationContext) -> ControlOutput:
+        return ControlOutput(
+            control=ControlField(
+                image=self.image,
+                control_model=self.control_model,
+                control_weight=self.control_weight,
+                begin_step_percent=self.begin_step_percent,
+                end_step_percent=self.end_step_percent,
+                control_mode=self.control_mode,
+                resize_mode=self.resize_mode,
+            ),
+        )
+
+
+@invocation(
+    "heuristic_resize",
+    title="Heuristic Resize",
+    tags=["image, controlnet"],
+    category="image",
+    version="1.1.1",
+    classification=Classification.Prototype,
+)
+class HeuristicResizeInvocation(BaseInvocation):
+    """Resize an image using a heuristic method. Preserves edge maps."""
+
+    image: ImageField = InputField(description="The image to resize")
+    width: int = InputField(default=512, ge=1, description="The width to resize to (px)")
+    height: int = InputField(default=512, ge=1, description="The height to resize to (px)")
+
+    def invoke(self, context: InvocationContext) -> ImageOutput:
+        image = context.images.get_pil(self.image.image_name, "RGB")
+        np_img = pil_to_np(image)
+        np_resized = heuristic_resize_fast(np_img, (self.width, self.height))
+        resized = np_to_pil(np_resized)
+        image_dto = context.images.save(image=resized)
+        return ImageOutput.build(image_dto)

invokeai/app/invocations/controlnet_image_processors.py

@@ -1,716 +0,0 @@
-# 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 bool, float
-from pathlib import Path
-from typing import Dict, List, Literal, Union
-
-import cv2
-import numpy as np
-from controlnet_aux import (
-    ContentShuffleDetector,
-    LeresDetector,
-    MediapipeFaceDetector,
-    MidasDetector,
-    MLSDdetector,
-    NormalBaeDetector,
-    PidiNetDetector,
-    SamDetector,
-    ZoeDetector,
-)
-from controlnet_aux.util import HWC3, ade_palette
-from PIL import Image
-from pydantic import BaseModel, Field, field_validator, model_validator
-from transformers import pipeline
-from transformers.pipelines import DepthEstimationPipeline
-
-from invokeai.app.invocations.baseinvocation import (
-    BaseInvocation,
-    BaseInvocationOutput,
-    Classification,
-    invocation,
-    invocation_output,
-)
-from invokeai.app.invocations.fields import (
-    FieldDescriptions,
-    ImageField,
-    InputField,
-    OutputField,
-    UIType,
-    WithBoard,
-    WithMetadata,
-)
-from invokeai.app.invocations.model import ModelIdentifierField
-from invokeai.app.invocations.primitives import ImageOutput
-from invokeai.app.invocations.util import validate_begin_end_step, validate_weights
-from invokeai.app.services.shared.invocation_context import InvocationContext
-from invokeai.app.util.controlnet_utils import CONTROLNET_MODE_VALUES, CONTROLNET_RESIZE_VALUES, heuristic_resize
-from invokeai.backend.image_util.canny import get_canny_edges
-from invokeai.backend.image_util.depth_anything.depth_anything_pipeline import DepthAnythingPipeline
-from invokeai.backend.image_util.dw_openpose import DWPOSE_MODELS, DWOpenposeDetector
-from invokeai.backend.image_util.hed import HEDProcessor
-from invokeai.backend.image_util.lineart import LineartProcessor
-from invokeai.backend.image_util.lineart_anime import LineartAnimeProcessor
-from invokeai.backend.image_util.util import np_to_pil, pil_to_np
-
-
-class ControlField(BaseModel):
-    image: ImageField = Field(description="The control image")
-    control_model: ModelIdentifierField = Field(description="The ControlNet model 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")
-
-    @field_validator("control_weight")
-    @classmethod
-    def validate_control_weight(cls, v):
-        validate_weights(v)
-        return v
-
-    @model_validator(mode="after")
-    def validate_begin_end_step_percent(self):
-        validate_begin_end_step(self.begin_step_percent, self.end_step_percent)
-        return self
-
-
-@invocation_output("control_output")
-class ControlOutput(BaseInvocationOutput):
-    """node output for ControlNet info"""
-
-    # Outputs
-    control: ControlField = OutputField(description=FieldDescriptions.control)
-
-
-@invocation("controlnet", title="ControlNet - SD1.5, SDXL", tags=["controlnet"], category="controlnet", version="1.1.3")
-class ControlNetInvocation(BaseInvocation):
-    """Collects ControlNet info to pass to other nodes"""
-
-    image: ImageField = InputField(description="The control image")
-    control_model: ModelIdentifierField = InputField(
-        description=FieldDescriptions.controlnet_model, ui_type=UIType.ControlNetModel
-    )
-    control_weight: Union[float, List[float]] = InputField(
-        default=1.0, ge=-1, le=2, description="The weight given to the ControlNet"
-    )
-    begin_step_percent: float = InputField(
-        default=0, ge=0, le=1, description="When the ControlNet is first applied (% of total steps)"
-    )
-    end_step_percent: float = InputField(
-        default=1, ge=0, le=1, description="When the ControlNet is last applied (% of total steps)"
-    )
-    control_mode: CONTROLNET_MODE_VALUES = InputField(default="balanced", description="The control mode used")
-    resize_mode: CONTROLNET_RESIZE_VALUES = InputField(default="just_resize", description="The resize mode used")
-
-    @field_validator("control_weight")
-    @classmethod
-    def validate_control_weight(cls, v):
-        validate_weights(v)
-        return v
-
-    @model_validator(mode="after")
-    def validate_begin_end_step_percent(self) -> "ControlNetInvocation":
-        validate_begin_end_step(self.begin_step_percent, self.end_step_percent)
-        return self
-
-    def invoke(self, context: InvocationContext) -> ControlOutput:
-        return ControlOutput(
-            control=ControlField(
-                image=self.image,
-                control_model=self.control_model,
-                control_weight=self.control_weight,
-                begin_step_percent=self.begin_step_percent,
-                end_step_percent=self.end_step_percent,
-                control_mode=self.control_mode,
-                resize_mode=self.resize_mode,
-            ),
-        )
-
-
-# This invocation exists for other invocations to subclass it - do not register with @invocation!
-class ImageProcessorInvocation(BaseInvocation, WithMetadata, WithBoard):
-    """Base class for invocations that preprocess images for ControlNet"""
-
-    image: ImageField = InputField(description="The image to process")
-
-    def run_processor(self, image: Image.Image) -> Image.Image:
-        # superclass just passes through image without processing
-        return image
-
-    def load_image(self, context: InvocationContext) -> Image.Image:
-        # allows override for any special formatting specific to the preprocessor
-        return context.images.get_pil(self.image.image_name, "RGB")
-
-    def invoke(self, context: InvocationContext) -> ImageOutput:
-        self._context = context
-        raw_image = self.load_image(context)
-        # image type should be PIL.PngImagePlugin.PngImageFile ?
-        processed_image = self.run_processor(raw_image)
-
-        # currently can't see processed image in node UI without a showImage node,
-        #    so for now setting image_type to RESULT instead of INTERMEDIATE so will get saved in gallery
-        image_dto = context.images.save(image=processed_image)
-
-        """Builds an ImageOutput and its ImageField"""
-        processed_image_field = ImageField(image_name=image_dto.image_name)
-        return ImageOutput(
-            image=processed_image_field,
-            # width=processed_image.width,
-            width=image_dto.width,
-            # height=processed_image.height,
-            height=image_dto.height,
-            # mode=processed_image.mode,
-        )
-
-
-@invocation(
-    "canny_image_processor",
-    title="Canny Processor",
-    tags=["controlnet", "canny"],
-    category="controlnet",
-    version="1.3.3",
-    classification=Classification.Deprecated,
-)
-class CannyImageProcessorInvocation(ImageProcessorInvocation):
-    """Canny edge detection for ControlNet"""
-
-    detect_resolution: int = InputField(default=512, ge=1, description=FieldDescriptions.detect_res)
-    image_resolution: int = InputField(default=512, ge=1, description=FieldDescriptions.image_res)
-    low_threshold: int = InputField(
-        default=100, ge=0, le=255, description="The low threshold of the Canny pixel gradient (0-255)"
-    )
-    high_threshold: int = InputField(
-        default=200, ge=0, le=255, description="The high threshold of the Canny pixel gradient (0-255)"
-    )
-
-    def load_image(self, context: InvocationContext) -> Image.Image:
-        # Keep alpha channel for Canny processing to detect edges of transparent areas
-        return context.images.get_pil(self.image.image_name, "RGBA")
-
-    def run_processor(self, image: Image.Image) -> Image.Image:
-        processed_image = get_canny_edges(
-            image,
-            self.low_threshold,
-            self.high_threshold,
-            detect_resolution=self.detect_resolution,
-            image_resolution=self.image_resolution,
-        )
-        return processed_image
-
-
-@invocation(
-    "hed_image_processor",
-    title="HED (softedge) Processor",
-    tags=["controlnet", "hed", "softedge"],
-    category="controlnet",
-    version="1.2.3",
-    classification=Classification.Deprecated,
-)
-class HedImageProcessorInvocation(ImageProcessorInvocation):
-    """Applies HED edge detection to image"""
-
-    detect_resolution: int = InputField(default=512, ge=1, description=FieldDescriptions.detect_res)
-    image_resolution: int = InputField(default=512, ge=1, description=FieldDescriptions.image_res)
-    # safe not supported in controlnet_aux v0.0.3
-    # safe: bool = InputField(default=False, description=FieldDescriptions.safe_mode)
-    scribble: bool = InputField(default=False, description=FieldDescriptions.scribble_mode)
-
-    def run_processor(self, image: Image.Image) -> Image.Image:
-        hed_processor = HEDProcessor()
-        processed_image = hed_processor.run(
-            image,
-            detect_resolution=self.detect_resolution,
-            image_resolution=self.image_resolution,
-            # safe not supported in controlnet_aux v0.0.3
-            # safe=self.safe,
-            scribble=self.scribble,
-        )
-        return processed_image
-
-
-@invocation(
-    "lineart_image_processor",
-    title="Lineart Processor",
-    tags=["controlnet", "lineart"],
-    category="controlnet",
-    version="1.2.3",
-    classification=Classification.Deprecated,
-)
-class LineartImageProcessorInvocation(ImageProcessorInvocation):
-    """Applies line art processing to image"""
-
-    detect_resolution: int = InputField(default=512, ge=1, description=FieldDescriptions.detect_res)
-    image_resolution: int = InputField(default=512, ge=1, description=FieldDescriptions.image_res)
-    coarse: bool = InputField(default=False, description="Whether to use coarse mode")
-
-    def run_processor(self, image: Image.Image) -> Image.Image:
-        lineart_processor = LineartProcessor()
-        processed_image = lineart_processor.run(
-            image, detect_resolution=self.detect_resolution, image_resolution=self.image_resolution, coarse=self.coarse
-        )
-        return processed_image
-
-
-@invocation(
-    "lineart_anime_image_processor",
-    title="Lineart Anime Processor",
-    tags=["controlnet", "lineart", "anime"],
-    category="controlnet",
-    version="1.2.3",
-    classification=Classification.Deprecated,
-)
-class LineartAnimeImageProcessorInvocation(ImageProcessorInvocation):
-    """Applies line art anime processing to image"""
-
-    detect_resolution: int = InputField(default=512, ge=1, description=FieldDescriptions.detect_res)
-    image_resolution: int = InputField(default=512, ge=1, description=FieldDescriptions.image_res)
-
-    def run_processor(self, image: Image.Image) -> Image.Image:
-        processor = LineartAnimeProcessor()
-        processed_image = processor.run(
-            image,
-            detect_resolution=self.detect_resolution,
-            image_resolution=self.image_resolution,
-        )
-        return processed_image
-
-
-@invocation(
-    "midas_depth_image_processor",
-    title="Midas Depth Processor",
-    tags=["controlnet", "midas"],
-    category="controlnet",
-    version="1.2.4",
-    classification=Classification.Deprecated,
-)
-class MidasDepthImageProcessorInvocation(ImageProcessorInvocation):
-    """Applies Midas depth processing to image"""
-
-    a_mult: float = InputField(default=2.0, ge=0, description="Midas parameter `a_mult` (a = a_mult * PI)")
-    bg_th: float = InputField(default=0.1, ge=0, description="Midas parameter `bg_th`")
-    detect_resolution: int = InputField(default=512, ge=1, description=FieldDescriptions.detect_res)
-    image_resolution: int = InputField(default=512, ge=1, description=FieldDescriptions.image_res)
-    # depth_and_normal not supported in controlnet_aux v0.0.3
-    # depth_and_normal: bool = InputField(default=False, description="whether to use depth and normal mode")
-
-    def run_processor(self, image: Image.Image) -> Image.Image:
-        # TODO: replace from_pretrained() calls with context.models.download_and_cache() (or similar)
-        midas_processor = MidasDetector.from_pretrained("lllyasviel/Annotators")
-        processed_image = midas_processor(
-            image,
-            a=np.pi * self.a_mult,
-            bg_th=self.bg_th,
-            image_resolution=self.image_resolution,
-            detect_resolution=self.detect_resolution,
-            # dept_and_normal not supported in controlnet_aux v0.0.3
-            # depth_and_normal=self.depth_and_normal,
-        )
-        return processed_image
-
-
-@invocation(
-    "normalbae_image_processor",
-    title="Normal BAE Processor",
-    tags=["controlnet"],
-    category="controlnet",
-    version="1.2.3",
-    classification=Classification.Deprecated,
-)
-class NormalbaeImageProcessorInvocation(ImageProcessorInvocation):
-    """Applies NormalBae processing to image"""
-
-    detect_resolution: int = InputField(default=512, ge=1, description=FieldDescriptions.detect_res)
-    image_resolution: int = InputField(default=512, ge=1, description=FieldDescriptions.image_res)
-
-    def run_processor(self, image: Image.Image) -> Image.Image:
-        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
-
-
-@invocation(
-    "mlsd_image_processor",
-    title="MLSD Processor",
-    tags=["controlnet", "mlsd"],
-    category="controlnet",
-    version="1.2.3",
-    classification=Classification.Deprecated,
-)
-class MlsdImageProcessorInvocation(ImageProcessorInvocation):
-    """Applies MLSD processing to image"""
-
-    detect_resolution: int = InputField(default=512, ge=1, description=FieldDescriptions.detect_res)
-    image_resolution: int = InputField(default=512, ge=1, description=FieldDescriptions.image_res)
-    thr_v: float = InputField(default=0.1, ge=0, description="MLSD parameter `thr_v`")
-    thr_d: float = InputField(default=0.1, ge=0, description="MLSD parameter `thr_d`")
-
-    def run_processor(self, image: Image.Image) -> Image.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,
-        )
-        return processed_image
-
-
-@invocation(
-    "pidi_image_processor",
-    title="PIDI Processor",
-    tags=["controlnet", "pidi"],
-    category="controlnet",
-    version="1.2.3",
-    classification=Classification.Deprecated,
-)
-class PidiImageProcessorInvocation(ImageProcessorInvocation):
-    """Applies PIDI processing to image"""
-
-    detect_resolution: int = InputField(default=512, ge=1, description=FieldDescriptions.detect_res)
-    image_resolution: int = InputField(default=512, ge=1, description=FieldDescriptions.image_res)
-    safe: bool = InputField(default=False, description=FieldDescriptions.safe_mode)
-    scribble: bool = InputField(default=False, description=FieldDescriptions.scribble_mode)
-
-    def run_processor(self, image: Image.Image) -> Image.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,
-        )
-        return processed_image
-
-
-@invocation(
-    "content_shuffle_image_processor",
-    title="Content Shuffle Processor",
-    tags=["controlnet", "contentshuffle"],
-    category="controlnet",
-    version="1.2.3",
-    classification=Classification.Deprecated,
-)
-class ContentShuffleImageProcessorInvocation(ImageProcessorInvocation):
-    """Applies content shuffle processing to image"""
-
-    detect_resolution: int = InputField(default=512, ge=1, description=FieldDescriptions.detect_res)
-    image_resolution: int = InputField(default=512, ge=1, description=FieldDescriptions.image_res)
-    h: int = InputField(default=512, ge=0, description="Content shuffle `h` parameter")
-    w: int = InputField(default=512, ge=0, description="Content shuffle `w` parameter")
-    f: int = InputField(default=256, ge=0, description="Content shuffle `f` parameter")
-
-    def run_processor(self, image: Image.Image) -> Image.Image:
-        content_shuffle_processor = ContentShuffleDetector()
-        processed_image = content_shuffle_processor(
-            image,
-            detect_resolution=self.detect_resolution,
-            image_resolution=self.image_resolution,
-            h=self.h,
-            w=self.w,
-            f=self.f,
-        )
-        return processed_image
-
-
-# should work with controlnet_aux >= 0.0.4 and timm <= 0.6.13
-@invocation(
-    "zoe_depth_image_processor",
-    title="Zoe (Depth) Processor",
-    tags=["controlnet", "zoe", "depth"],
-    category="controlnet",
-    version="1.2.3",
-    classification=Classification.Deprecated,
-)
-class ZoeDepthImageProcessorInvocation(ImageProcessorInvocation):
-    """Applies Zoe depth processing to image"""
-
-    def run_processor(self, image: Image.Image) -> Image.Image:
-        zoe_depth_processor = ZoeDetector.from_pretrained("lllyasviel/Annotators")
-        processed_image = zoe_depth_processor(image)
-        return processed_image
-
-
-@invocation(
-    "mediapipe_face_processor",
-    title="Mediapipe Face Processor",
-    tags=["controlnet", "mediapipe", "face"],
-    category="controlnet",
-    version="1.2.4",
-    classification=Classification.Deprecated,
-)
-class MediapipeFaceProcessorInvocation(ImageProcessorInvocation):
-    """Applies mediapipe face processing to image"""
-
-    max_faces: int = InputField(default=1, ge=1, description="Maximum number of faces to detect")
-    min_confidence: float = InputField(default=0.5, ge=0, le=1, description="Minimum confidence for face detection")
-    detect_resolution: int = InputField(default=512, ge=1, description=FieldDescriptions.detect_res)
-    image_resolution: int = InputField(default=512, ge=1, description=FieldDescriptions.image_res)
-
-    def run_processor(self, image: Image.Image) -> Image.Image:
-        mediapipe_face_processor = MediapipeFaceDetector()
-        processed_image = mediapipe_face_processor(
-            image,
-            max_faces=self.max_faces,
-            min_confidence=self.min_confidence,
-            image_resolution=self.image_resolution,
-            detect_resolution=self.detect_resolution,
-        )
-        return processed_image
-
-
-@invocation(
-    "leres_image_processor",
-    title="Leres (Depth) Processor",
-    tags=["controlnet", "leres", "depth"],
-    category="controlnet",
-    version="1.2.3",
-    classification=Classification.Deprecated,
-)
-class LeresImageProcessorInvocation(ImageProcessorInvocation):
-    """Applies leres processing to image"""
-
-    thr_a: float = InputField(default=0, description="Leres parameter `thr_a`")
-    thr_b: float = InputField(default=0, description="Leres parameter `thr_b`")
-    boost: bool = InputField(default=False, description="Whether to use boost mode")
-    detect_resolution: int = InputField(default=512, ge=1, description=FieldDescriptions.detect_res)
-    image_resolution: int = InputField(default=512, ge=1, description=FieldDescriptions.image_res)
-
-    def run_processor(self, image: Image.Image) -> Image.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,
-        )
-        return processed_image
-
-
-@invocation(
-    "tile_image_processor",
-    title="Tile Resample Processor",
-    tags=["controlnet", "tile"],
-    category="controlnet",
-    version="1.2.3",
-    classification=Classification.Deprecated,
-)
-class TileResamplerProcessorInvocation(ImageProcessorInvocation):
-    """Tile resampler processor"""
-
-    # res: int = InputField(default=512, ge=0, le=1024, description="The pixel resolution for each tile")
-    down_sampling_rate: float = InputField(default=1.0, ge=1.0, le=8.0, description="Down sampling rate")
-
-    # tile_resample copied from sd-webui-controlnet/scripts/processor.py
-    def tile_resample(
-        self,
-        np_img: np.ndarray,
-        res=512,  # never used?
-        down_sampling_rate=1.0,
-    ):
-        np_img = HWC3(np_img)
-        if down_sampling_rate < 1.1:
-            return np_img
-        H, W, C = np_img.shape
-        H = int(float(H) / float(down_sampling_rate))
-        W = int(float(W) / float(down_sampling_rate))
-        np_img = cv2.resize(np_img, (W, H), interpolation=cv2.INTER_AREA)
-        return np_img
-
-    def run_processor(self, image: Image.Image) -> Image.Image:
-        np_img = np.array(image, dtype=np.uint8)
-        processed_np_image = self.tile_resample(
-            np_img,
-            # res=self.tile_size,
-            down_sampling_rate=self.down_sampling_rate,
-        )
-        processed_image = Image.fromarray(processed_np_image)
-        return processed_image
-
-
-@invocation(
-    "segment_anything_processor",
-    title="Segment Anything Processor",
-    tags=["controlnet", "segmentanything"],
-    category="controlnet",
-    version="1.2.4",
-    classification=Classification.Deprecated,
-)
-class SegmentAnythingProcessorInvocation(ImageProcessorInvocation):
-    """Applies segment anything processing to image"""
-
-    detect_resolution: int = InputField(default=512, ge=1, description=FieldDescriptions.detect_res)
-    image_resolution: int = InputField(default=512, ge=1, description=FieldDescriptions.image_res)
-
-    def run_processor(self, image: Image.Image) -> Image.Image:
-        # segment_anything_processor = SamDetector.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, image_resolution=self.image_resolution, detect_resolution=self.detect_resolution
-        )
-        return processed_image
-
-
-class SamDetectorReproducibleColors(SamDetector):
-    # overriding SamDetector.show_anns() method to use reproducible colors for segmentation image
-    #     base class show_anns() method randomizes colors,
-    #     which seems to also lead to non-reproducible image generation
-    # so using ADE20k color palette instead
-    def show_anns(self, anns: List[Dict]):
-        if len(anns) == 0:
-            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")
-        palette = ade_palette()
-        for i, ann in enumerate(sorted_anns):
-            m = ann["segmentation"]
-            img = np.empty((m.shape[0], m.shape[1], 3), dtype=np.uint8)
-            # 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)))
-        return np.array(final_img, dtype=np.uint8)
-
-
-@invocation(
-    "color_map_image_processor",
-    title="Color Map Processor",
-    tags=["controlnet"],
-    category="controlnet",
-    version="1.2.3",
-    classification=Classification.Deprecated,
-)
-class ColorMapImageProcessorInvocation(ImageProcessorInvocation):
-    """Generates a color map from the provided image"""
-
-    color_map_tile_size: int = InputField(default=64, ge=1, description=FieldDescriptions.tile_size)
-
-    def run_processor(self, image: Image.Image) -> Image.Image:
-        np_image = np.array(image, dtype=np.uint8)
-        height, width = np_image.shape[:2]
-
-        width_tile_size = min(self.color_map_tile_size, width)
-        height_tile_size = min(self.color_map_tile_size, height)
-
-        color_map = cv2.resize(
-            np_image,
-            (width // width_tile_size, height // height_tile_size),
-            interpolation=cv2.INTER_CUBIC,
-        )
-        color_map = cv2.resize(color_map, (width, height), interpolation=cv2.INTER_NEAREST)
-        color_map = Image.fromarray(color_map)
-        return color_map
-
-
-DEPTH_ANYTHING_MODEL_SIZES = Literal["large", "base", "small", "small_v2"]
-# DepthAnything V2 Small model is licensed under Apache 2.0 but not the base and large models.
-DEPTH_ANYTHING_MODELS = {
-    "large": "LiheYoung/depth-anything-large-hf",
-    "base": "LiheYoung/depth-anything-base-hf",
-    "small": "LiheYoung/depth-anything-small-hf",
-    "small_v2": "depth-anything/Depth-Anything-V2-Small-hf",
-}
-
-
-@invocation(
-    "depth_anything_image_processor",
-    title="Depth Anything Processor",
-    tags=["controlnet", "depth", "depth anything"],
-    category="controlnet",
-    version="1.1.3",
-    classification=Classification.Deprecated,
-)
-class DepthAnythingImageProcessorInvocation(ImageProcessorInvocation):
-    """Generates a depth map based on the Depth Anything algorithm"""
-
-    model_size: DEPTH_ANYTHING_MODEL_SIZES = InputField(
-        default="small_v2", description="The size of the depth model to use"
-    )
-    resolution: int = InputField(default=512, ge=1, description=FieldDescriptions.image_res)
-
-    def run_processor(self, image: Image.Image) -> Image.Image:
-        def load_depth_anything(model_path: Path):
-            depth_anything_pipeline = pipeline(model=str(model_path), task="depth-estimation", local_files_only=True)
-            assert isinstance(depth_anything_pipeline, DepthEstimationPipeline)
-            return DepthAnythingPipeline(depth_anything_pipeline)
-
-        with self._context.models.load_remote_model(
-            source=DEPTH_ANYTHING_MODELS[self.model_size], loader=load_depth_anything
-        ) as depth_anything_detector:
-            assert isinstance(depth_anything_detector, DepthAnythingPipeline)
-            depth_map = depth_anything_detector.generate_depth(image)
-
-            # Resizing to user target specified size
-            new_height = int(image.size[1] * (self.resolution / image.size[0]))
-            depth_map = depth_map.resize((self.resolution, new_height))
-
-            return depth_map
-
-
-@invocation(
-    "dw_openpose_image_processor",
-    title="DW Openpose Image Processor",
-    tags=["controlnet", "dwpose", "openpose"],
-    category="controlnet",
-    version="1.1.1",
-    classification=Classification.Deprecated,
-)
-class DWOpenposeImageProcessorInvocation(ImageProcessorInvocation):
-    """Generates an openpose pose from an image using DWPose"""
-
-    draw_body: bool = InputField(default=True)
-    draw_face: bool = InputField(default=False)
-    draw_hands: bool = InputField(default=False)
-    image_resolution: int = InputField(default=512, ge=1, description=FieldDescriptions.image_res)
-
-    def run_processor(self, image: Image.Image) -> Image.Image:
-        onnx_det = self._context.models.download_and_cache_model(DWPOSE_MODELS["yolox_l.onnx"])
-        onnx_pose = self._context.models.download_and_cache_model(DWPOSE_MODELS["dw-ll_ucoco_384.onnx"])
-
-        dw_openpose = DWOpenposeDetector(onnx_det=onnx_det, onnx_pose=onnx_pose)
-        processed_image = dw_openpose(
-            image,
-            draw_face=self.draw_face,
-            draw_hands=self.draw_hands,
-            draw_body=self.draw_body,
-            resolution=self.image_resolution,
-        )
-        return processed_image
-
-
-@invocation(
-    "heuristic_resize",
-    title="Heuristic Resize",
-    tags=["image, controlnet"],
-    category="image",
-    version="1.0.1",
-    classification=Classification.Prototype,
-)
-class HeuristicResizeInvocation(BaseInvocation):
-    """Resize an image using a heuristic method. Preserves edge maps."""
-
-    image: ImageField = InputField(description="The image to resize")
-    width: int = InputField(default=512, ge=1, description="The width to resize to (px)")
-    height: int = InputField(default=512, ge=1, description="The height to resize to (px)")
-
-    def invoke(self, context: InvocationContext) -> ImageOutput:
-        image = context.images.get_pil(self.image.image_name, "RGB")
-        np_img = pil_to_np(image)
-        np_resized = heuristic_resize(np_img, (self.width, self.height))
-        resized = np_to_pil(np_resized)
-        image_dto = context.images.save(image=resized)
-        return ImageOutput.build(image_dto)

invokeai/app/invocations/create_gradient_mask.py

@@ -1,12 +1,14 @@
 from typing import Literal, Optional
 
+import cv2
 import numpy as np
 import torch
 import torchvision.transforms as T
-from PIL import Image, ImageFilter
+from PIL import Image
 from torchvision.transforms.functional import resize as tv_resize
 
 from invokeai.app.invocations.baseinvocation import BaseInvocation, BaseInvocationOutput, invocation, invocation_output
+from invokeai.app.invocations.constants import LATENT_SCALE_FACTOR
 from invokeai.app.invocations.fields import (
     DenoiseMaskField,
     FieldDescriptions,
@@ -19,7 +21,8 @@ from invokeai.app.invocations.image_to_latents import ImageToLatentsInvocation
 from invokeai.app.invocations.model import UNetField, VAEField
 from invokeai.app.services.shared.invocation_context import InvocationContext
 from invokeai.backend.model_manager import LoadedModel
-from invokeai.backend.model_manager.config import MainConfigBase, ModelVariantType
+from invokeai.backend.model_manager.config import MainConfigBase
+from invokeai.backend.model_manager.taxonomy import ModelVariantType
 from invokeai.backend.stable_diffusion.diffusers_pipeline import image_resized_to_grid_as_tensor
 
 
@@ -41,15 +44,13 @@ class GradientMaskOutput(BaseInvocationOutput):
     title="Create Gradient Mask",
     tags=["mask", "denoise"],
     category="latents",
-    version="1.2.0",
+    version="1.3.0",
 )
 class CreateGradientMaskInvocation(BaseInvocation):
-    """Creates mask for denoising model run."""
+    """Creates mask for denoising."""
 
-    mask: ImageField = InputField(default=None, description="Image which will be masked", ui_order=1)
-    edge_radius: int = InputField(
-        default=16, ge=0, description="How far to blur/expand the edges of the mask", ui_order=2
-    )
+    mask: ImageField = InputField(description="Image which will be masked", ui_order=1)
+    edge_radius: int = InputField(default=16, ge=0, description="How far to expand the edges of the mask", ui_order=2)
     coherence_mode: Literal["Gaussian Blur", "Box Blur", "Staged"] = InputField(default="Gaussian Blur", ui_order=3)
     minimum_denoise: float = InputField(
         default=0.0, ge=0, le=1, description="Minimum denoise level for the coherence region", ui_order=4
@@ -80,45 +81,110 @@ class CreateGradientMaskInvocation(BaseInvocation):
     @torch.no_grad()
     def invoke(self, context: InvocationContext) -> GradientMaskOutput:
         mask_image = context.images.get_pil(self.mask.image_name, mode="L")
+
+        # Resize the mask_image. Makes the filter 64x faster and doesn't hurt quality in latent scale anyway
+        mask_image = mask_image.resize(
+            (
+                mask_image.width // LATENT_SCALE_FACTOR,
+                mask_image.height // LATENT_SCALE_FACTOR,
+            ),
+            resample=Image.Resampling.BILINEAR,
+        )
+
+        mask_np_orig = np.array(mask_image, dtype=np.float32)
+
+        self.edge_radius = self.edge_radius // LATENT_SCALE_FACTOR  # scale the edge radius to match the mask size
+
         if self.edge_radius > 0:
+            mask_np = 255 - mask_np_orig  # invert so 0 is unmasked (higher values = higher denoise strength)
+            dilated_mask = mask_np.copy()
+
+            # Create kernel based on coherence mode
             if self.coherence_mode == "Box Blur":
-                blur_mask = mask_image.filter(ImageFilter.BoxBlur(self.edge_radius))
-            else:  # Gaussian Blur OR Staged
-                # Gaussian Blur uses standard deviation. 1/2 radius is a good approximation
-                blur_mask = mask_image.filter(ImageFilter.GaussianBlur(self.edge_radius / 2))
+                # Create a circular distance kernel that fades from center outward
+                kernel_size = self.edge_radius * 2 + 1
+                center = self.edge_radius
+                kernel = np.zeros((kernel_size, kernel_size), dtype=np.float32)
+                for i in range(kernel_size):
+                    for j in range(kernel_size):
+                        dist = np.sqrt((i - center) ** 2 + (j - center) ** 2)
+                        if dist <= self.edge_radius:
+                            kernel[i, j] = 1.0 - (dist / self.edge_radius)
+            else:  # Gaussian Blur or Staged
+                # Create a Gaussian kernel
+                kernel_size = self.edge_radius * 2 + 1
+                kernel = cv2.getGaussianKernel(
+                    kernel_size, self.edge_radius / 2.5
+                )  # 2.5 is a magic number (standard deviation capturing)
+                kernel = kernel * kernel.T  # Make 2D gaussian kernel
+                kernel = kernel / np.max(kernel)  # Normalize center to 1.0
 
-            blur_tensor: torch.Tensor = image_resized_to_grid_as_tensor(blur_mask, normalize=False)
+                # Ensure values outside radius are 0
+                center = self.edge_radius
+                for i in range(kernel_size):
+                    for j in range(kernel_size):
+                        dist = np.sqrt((i - center) ** 2 + (j - center) ** 2)
+                        if dist > self.edge_radius:
+                            kernel[i, j] = 0
 
-            # redistribute blur so that the original edges are 0 and blur outwards to 1
-            blur_tensor = (blur_tensor - 0.5) * 2
-            blur_tensor[blur_tensor < 0] = 0.0
+            # 2D max filter
+            mask_tensor = torch.tensor(mask_np)
+            kernel_tensor = torch.tensor(kernel)
+            dilated_mask = 255 - self.max_filter2D_torch(mask_tensor, kernel_tensor).cpu()
+            dilated_mask = dilated_mask.numpy()
 
-            threshold = 1 - self.minimum_denoise
+            threshold = (1 - self.minimum_denoise) * 255
 
             if self.coherence_mode == "Staged":
-                # wherever the blur_tensor is less than fully masked, convert it to threshold
-                blur_tensor = torch.where((blur_tensor < 1) & (blur_tensor > 0), threshold, blur_tensor)
-            else:
-                # wherever the blur_tensor is above threshold but less than 1, drop it to threshold
-                blur_tensor = torch.where((blur_tensor > threshold) & (blur_tensor < 1), threshold, blur_tensor)
+                # wherever expanded mask is darker than the original mask but original was above threshhold, set it to the threshold
+                # makes any expansion areas drop to threshhold. Raising minimum across the image happen outside of this if
+                threshold_mask = (dilated_mask < mask_np_orig) & (mask_np_orig > threshold)
+                dilated_mask = np.where(threshold_mask, threshold, mask_np_orig)
+
+            # wherever expanded mask is less than 255 but greater than threshold, drop it to threshold (minimum denoise)
+            threshold_mask = (dilated_mask > threshold) & (dilated_mask < 255)
+            dilated_mask = np.where(threshold_mask, threshold, dilated_mask)
 
         else:
-            blur_tensor: torch.Tensor = image_resized_to_grid_as_tensor(mask_image, normalize=False)
+            dilated_mask = mask_np_orig.copy()
 
-        mask_name = context.tensors.save(tensor=blur_tensor.unsqueeze(1))
+        # convert to tensor
+        dilated_mask = np.clip(dilated_mask, 0, 255).astype(np.uint8)
+        mask_tensor = torch.tensor(dilated_mask, device=torch.device("cpu"))
 
-        # compute a [0, 1] mask from the blur_tensor
-        expanded_mask = torch.where((blur_tensor < 1), 0, 1)
-        expanded_mask_image = Image.fromarray((expanded_mask.squeeze(0).numpy() * 255).astype(np.uint8), mode="L")
+        # binary mask for compositing
+        expanded_mask = np.where((dilated_mask < 255), 0, 255)
+        expanded_mask_image = Image.fromarray(expanded_mask.astype(np.uint8), mode="L")
+        expanded_mask_image = expanded_mask_image.resize(
+            (
+                mask_image.width * LATENT_SCALE_FACTOR,
+                mask_image.height * LATENT_SCALE_FACTOR,
+            ),
+            resample=Image.Resampling.NEAREST,
+        )
         expanded_image_dto = context.images.save(expanded_mask_image)
 
+        # restore the original mask size
+        dilated_mask = Image.fromarray(dilated_mask.astype(np.uint8))
+        dilated_mask = dilated_mask.resize(
+            (
+                mask_image.width * LATENT_SCALE_FACTOR,
+                mask_image.height * LATENT_SCALE_FACTOR,
+            ),
+            resample=Image.Resampling.NEAREST,
+        )
+
+        # stack the mask as a tensor, repeating 4 times on dimmension 1
+        dilated_mask_tensor = image_resized_to_grid_as_tensor(dilated_mask, normalize=False)
+        mask_name = context.tensors.save(tensor=dilated_mask_tensor.unsqueeze(0))
+
         masked_latents_name = None
         if self.unet is not None and self.vae is not None and self.image is not None:
             # all three fields must be present at the same time
             main_model_config = context.models.get_config(self.unet.unet.key)
             assert isinstance(main_model_config, MainConfigBase)
             if main_model_config.variant is ModelVariantType.Inpaint:
-                mask = blur_tensor
+                mask = dilated_mask_tensor
                 vae_info: LoadedModel = context.models.load(self.vae.vae)
                 image = context.images.get_pil(self.image.image_name)
                 image_tensor = image_resized_to_grid_as_tensor(image.convert("RGB"))
@@ -136,3 +202,29 @@ class CreateGradientMaskInvocation(BaseInvocation):
             denoise_mask=DenoiseMaskField(mask_name=mask_name, masked_latents_name=masked_latents_name, gradient=True),
             expanded_mask_area=ImageField(image_name=expanded_image_dto.image_name),
         )
+
+    def max_filter2D_torch(self, image: torch.Tensor, kernel: torch.Tensor) -> torch.Tensor:
+        """
+        This morphological operation is much faster in torch than numpy or opencv
+        For reasonable kernel sizes, the overhead of copying the data to the GPU is not worth it.
+        """
+        h, w = kernel.shape
+        pad_h, pad_w = h // 2, w // 2
+
+        padded = torch.nn.functional.pad(image, (pad_w, pad_w, pad_h, pad_h), mode="constant", value=0)
+        result = torch.zeros_like(image)
+
+        # This looks like it's inside out, but it does the same thing and is more efficient
+        for i in range(h):
+            for j in range(w):
+                weight = kernel[i, j]
+                if weight <= 0:
+                    continue
+
+                # Extract the region from padded tensor
+                region = padded[i : i + image.shape[0], j : j + image.shape[1]]
+
+                # Apply weight and update max
+                result = torch.maximum(result, region * weight)
+
+        return result

invokeai/app/invocations/denoise_latents.py

@@ -22,7 +22,7 @@ from transformers import CLIPVisionModelWithProjection
 
 from invokeai.app.invocations.baseinvocation import BaseInvocation, invocation
 from invokeai.app.invocations.constants import LATENT_SCALE_FACTOR
-from invokeai.app.invocations.controlnet_image_processors import ControlField
+from invokeai.app.invocations.controlnet import ControlField
 from invokeai.app.invocations.fields import (
     ConditioningField,
     DenoiseMaskField,
@@ -39,8 +39,8 @@ from invokeai.app.invocations.t2i_adapter import T2IAdapterField
 from invokeai.app.services.shared.invocation_context import InvocationContext
 from invokeai.app.util.controlnet_utils import prepare_control_image
 from invokeai.backend.ip_adapter.ip_adapter import IPAdapter
-from invokeai.backend.model_manager import BaseModelType, ModelVariantType
 from invokeai.backend.model_manager.config import AnyModelConfig
+from invokeai.backend.model_manager.taxonomy import BaseModelType, ModelVariantType
 from invokeai.backend.model_patcher import ModelPatcher
 from invokeai.backend.patches.layer_patcher import LayerPatcher
 from invokeai.backend.patches.model_patch_raw import ModelPatchRaw
@@ -608,6 +608,7 @@ class DenoiseLatentsInvocation(BaseInvocation):
                     end_step_percent=single_ip_adapter.end_step_percent,
                     ip_adapter_conditioning=IPAdapterConditioningInfo(image_prompt_embeds, uncond_image_prompt_embeds),
                     mask=mask,
+                    method=single_ip_adapter.method,
                 )
             )
 

invokeai/app/invocations/dw_openpose.py

@@ -4,7 +4,7 @@ from invokeai.app.invocations.baseinvocation import BaseInvocation, invocation
 from invokeai.app.invocations.fields import ImageField, InputField, WithBoard, WithMetadata
 from invokeai.app.invocations.primitives import ImageOutput
 from invokeai.app.services.shared.invocation_context import InvocationContext
-from invokeai.backend.image_util.dw_openpose import DWOpenposeDetector2
+from invokeai.backend.image_util.dw_openpose import DWOpenposeDetector
 
 
 @invocation(
@@ -25,20 +25,20 @@ class DWOpenposeDetectionInvocation(BaseInvocation, WithMetadata, WithBoard):
     def invoke(self, context: InvocationContext) -> ImageOutput:
         image = context.images.get_pil(self.image.image_name, "RGB")
 
-        onnx_det_path = context.models.download_and_cache_model(DWOpenposeDetector2.get_model_url_det())
-        onnx_pose_path = context.models.download_and_cache_model(DWOpenposeDetector2.get_model_url_pose())
+        onnx_det_path = context.models.download_and_cache_model(DWOpenposeDetector.get_model_url_det())
+        onnx_pose_path = context.models.download_and_cache_model(DWOpenposeDetector.get_model_url_pose())
 
         loaded_session_det = context.models.load_local_model(
-            onnx_det_path, DWOpenposeDetector2.create_onnx_inference_session
+            onnx_det_path, DWOpenposeDetector.create_onnx_inference_session
         )
         loaded_session_pose = context.models.load_local_model(
-            onnx_pose_path, DWOpenposeDetector2.create_onnx_inference_session
+            onnx_pose_path, DWOpenposeDetector.create_onnx_inference_session
         )
 
         with loaded_session_det as session_det, loaded_session_pose as session_pose:
             assert isinstance(session_det, ort.InferenceSession)
             assert isinstance(session_pose, ort.InferenceSession)
-            detector = DWOpenposeDetector2(session_det=session_det, session_pose=session_pose)
+            detector = DWOpenposeDetector(session_det=session_det, session_pose=session_pose)
             detected_image = detector.run(
                 image,
                 draw_face=self.draw_face,

invokeai/app/invocations/fields.py

@@ -40,6 +40,7 @@ class UIType(str, Enum, metaclass=MetaEnum):
 
     # region Model Field Types
     MainModel = "MainModelField"
+    CogView4MainModel = "CogView4MainModelField"
     FluxMainModel = "FluxMainModelField"
     SD3MainModel = "SD3MainModelField"
     SDXLMainModel = "SDXLMainModelField"
@@ -59,6 +60,10 @@ class UIType(str, Enum, metaclass=MetaEnum):
     ControlLoRAModel = "ControlLoRAModelField"
     SigLipModel = "SigLipModelField"
     FluxReduxModel = "FluxReduxModelField"
+    LlavaOnevisionModel = "LLaVAModelField"
+    Imagen3Model = "Imagen3ModelField"
+    Imagen4Model = "Imagen4ModelField"
+    ChatGPT4oModel = "ChatGPT4oModelField"
     # endregion
 
     # region Misc Field Types
@@ -136,6 +141,7 @@ class FieldDescriptions:
     noise = "Noise tensor"
     clip = "CLIP (tokenizer, text encoder, LoRAs) and skipped layer count"
     t5_encoder = "T5 tokenizer and text encoder"
+    glm_encoder = "GLM (THUDM) tokenizer and text encoder"
     clip_embed_model = "CLIP Embed loader"
     clip_g_model = "CLIP-G Embed loader"
     unet = "UNet (scheduler, LoRAs)"
@@ -150,6 +156,7 @@ class FieldDescriptions:
     main_model = "Main model (UNet, VAE, CLIP) to load"
     flux_model = "Flux model (Transformer) to load"
     sd3_model = "SD3 model (MMDiTX) to load"
+    cogview4_model = "CogView4 model (Transformer) to load"
     sdxl_main_model = "SDXL Main model (UNet, VAE, CLIP1, CLIP2) to load"
     sdxl_refiner_model = "SDXL Refiner Main Modde (UNet, VAE, CLIP2) to load"
     onnx_main_model = "ONNX Main model (UNet, VAE, CLIP) to load"
@@ -205,6 +212,8 @@ class FieldDescriptions:
     freeu_b2 = "Scaling factor for stage 2 to amplify the contributions of backbone features."
     instantx_control_mode = "The control mode for InstantX ControlNet union models. Ignored for other ControlNet models. The standard mapping is: canny (0), tile (1), depth (2), blur (3), pose (4), gray (5), low quality (6). Negative values will be treated as 'None'."
     flux_redux_conditioning = "FLUX Redux conditioning tensor"
+    vllm_model = "The VLLM model to use"
+    flux_fill_conditioning = "FLUX Fill conditioning tensor"
 
 
 class ImageField(BaseModel):
@@ -274,12 +283,25 @@ class FluxReduxConditioningField(BaseModel):
     )
 
 
+class FluxFillConditioningField(BaseModel):
+    """A FLUX Fill conditioning field."""
+
+    image: ImageField = Field(description="The FLUX Fill reference image.")
+    mask: TensorField = Field(description="The FLUX Fill inpaint mask.")
+
+
 class SD3ConditioningField(BaseModel):
     """A conditioning tensor primitive value"""
 
     conditioning_name: str = Field(description="The name of conditioning tensor")
 
 
+class CogView4ConditioningField(BaseModel):
+    """A conditioning tensor primitive value"""
+
+    conditioning_name: str = Field(description="The name of conditioning tensor")
+
+
 class ConditioningField(BaseModel):
     """A conditioning tensor primitive value"""
 
@@ -379,8 +401,8 @@ class InputFieldJSONSchemaExtra(BaseModel):
     """
 
     input: Input
-    orig_required: bool
     field_kind: FieldKind
+    orig_required: bool = True
     default: Optional[Any] = None
     orig_default: Optional[Any] = None
     ui_hidden: bool = False
@@ -415,7 +437,7 @@ class WithWorkflow:
     workflow = None
 
     def __init_subclass__(cls) -> None:
-        logger.warn(
+        logger.warning(
             f"{cls.__module__.split('.')[0]}.{cls.__name__}: WithWorkflow is deprecated. Use `context.workflow` to access the workflow."
         )
         super().__init_subclass__()
@@ -477,7 +499,7 @@ def InputField(
     input: Input = Input.Any,
     ui_type: Optional[UIType] = None,
     ui_component: Optional[UIComponent] = None,
-    ui_hidden: bool = False,
+    ui_hidden: Optional[bool] = None,
     ui_order: Optional[int] = None,
     ui_choice_labels: Optional[dict[str, str]] = None,
 ) -> Any:
@@ -513,15 +535,20 @@ def InputField(
 
     json_schema_extra_ = InputFieldJSONSchemaExtra(
         input=input,
-        ui_type=ui_type,
-        ui_component=ui_component,
-        ui_hidden=ui_hidden,
-        ui_order=ui_order,
-        ui_choice_labels=ui_choice_labels,
         field_kind=FieldKind.Input,
-        orig_required=True,
     )
 
+    if ui_type is not None:
+        json_schema_extra_.ui_type = ui_type
+    if ui_component is not None:
+        json_schema_extra_.ui_component = ui_component
+    if ui_hidden is not None:
+        json_schema_extra_.ui_hidden = ui_hidden
+    if ui_order is not None:
+        json_schema_extra_.ui_order = ui_order
+    if ui_choice_labels is not None:
+        json_schema_extra_.ui_choice_labels = ui_choice_labels
+
     """
     There is a conflict between the typing of invocation definitions and the typing of an invocation's
     `invoke()` function.
@@ -551,7 +578,7 @@ def InputField(
 
     if default_factory is not _Unset and default_factory is not None:
         default = default_factory()
-        logger.warn('"default_factory" is not supported, calling it now to set "default"')
+        logger.warning('"default_factory" is not supported, calling it now to set "default"')
 
     # These are the args we may wish pass to the pydantic `Field()` function
     field_args = {
@@ -593,7 +620,7 @@ def InputField(
 
     return Field(
         **provided_args,
-        json_schema_extra=json_schema_extra_.model_dump(exclude_none=True),
+        json_schema_extra=json_schema_extra_.model_dump(exclude_unset=True),
     )
 
 

invokeai/app/invocations/flux_control_lora_loader.py

@@ -1,7 +1,6 @@
 from invokeai.app.invocations.baseinvocation import (
     BaseInvocation,
     BaseInvocationOutput,
-    Classification,
     invocation,
     invocation_output,
 )
@@ -25,7 +24,6 @@ class FluxControlLoRALoaderOutput(BaseInvocationOutput):
     tags=["lora", "model", "flux"],
     category="model",
     version="1.1.1",
-    classification=Classification.Prototype,
 )
 class FluxControlLoRALoaderInvocation(BaseInvocation):
     """LoRA model and Image to use with FLUX transformer generation."""

invokeai/app/invocations/flux_controlnet.py

@@ -3,7 +3,6 @@ from pydantic import BaseModel, Field, field_validator, model_validator
 from invokeai.app.invocations.baseinvocation import (
     BaseInvocation,
     BaseInvocationOutput,
-    Classification,
     invocation,
     invocation_output,
 )
@@ -52,7 +51,6 @@ class FluxControlNetOutput(BaseInvocationOutput):
     tags=["controlnet", "flux"],
     category="controlnet",
     version="1.0.0",
-    classification=Classification.Prototype,
 )
 class FluxControlNetInvocation(BaseInvocation):
     """Collect FLUX ControlNet info to pass to other nodes."""

invokeai/app/invocations/flux_denoise.py

@@ -10,11 +10,12 @@ from PIL import Image
 from torchvision.transforms.functional import resize as tv_resize
 from transformers import CLIPImageProcessor, CLIPVisionModelWithProjection
 
-from invokeai.app.invocations.baseinvocation import BaseInvocation, Classification, invocation
+from invokeai.app.invocations.baseinvocation import BaseInvocation, invocation
 from invokeai.app.invocations.fields import (
     DenoiseMaskField,
     FieldDescriptions,
     FluxConditioningField,
+    FluxFillConditioningField,
     FluxReduxConditioningField,
     ImageField,
     Input,
@@ -32,7 +33,6 @@ from invokeai.app.services.shared.invocation_context import InvocationContext
 from invokeai.backend.flux.controlnet.instantx_controlnet_flux import InstantXControlNetFlux
 from invokeai.backend.flux.controlnet.xlabs_controlnet_flux import XLabsControlNetFlux
 from invokeai.backend.flux.denoise import denoise
-from invokeai.backend.flux.extensions.inpaint_extension import InpaintExtension
 from invokeai.backend.flux.extensions.instantx_controlnet_extension import InstantXControlNetExtension
 from invokeai.backend.flux.extensions.regional_prompting_extension import RegionalPromptingExtension
 from invokeai.backend.flux.extensions.xlabs_controlnet_extension import XLabsControlNetExtension
@@ -48,10 +48,11 @@ from invokeai.backend.flux.sampling_utils import (
     unpack,
 )
 from invokeai.backend.flux.text_conditioning import FluxReduxConditioning, FluxTextConditioning
-from invokeai.backend.model_manager.config import ModelFormat
+from invokeai.backend.model_manager.taxonomy import ModelFormat, ModelVariantType
 from invokeai.backend.patches.layer_patcher import LayerPatcher
 from invokeai.backend.patches.lora_conversions.flux_lora_constants import FLUX_LORA_TRANSFORMER_PREFIX
 from invokeai.backend.patches.model_patch_raw import ModelPatchRaw
+from invokeai.backend.rectified_flow.rectified_flow_inpaint_extension import RectifiedFlowInpaintExtension
 from invokeai.backend.stable_diffusion.diffusers_pipeline import PipelineIntermediateState
 from invokeai.backend.stable_diffusion.diffusion.conditioning_data import FLUXConditioningInfo
 from invokeai.backend.util.devices import TorchDevice
@@ -62,8 +63,7 @@ from invokeai.backend.util.devices import TorchDevice
     title="FLUX Denoise",
     tags=["image", "flux"],
     category="image",
-    version="3.2.3",
-    classification=Classification.Prototype,
+    version="3.3.0",
 )
 class FluxDenoiseInvocation(BaseInvocation, WithMetadata, WithBoard):
     """Run denoising process with a FLUX transformer model."""
@@ -109,6 +109,11 @@ class FluxDenoiseInvocation(BaseInvocation, WithMetadata, WithBoard):
         description="FLUX Redux conditioning tensor.",
         input=Input.Connection,
     )
+    fill_conditioning: FluxFillConditioningField | None = InputField(
+        default=None,
+        description="FLUX Fill conditioning.",
+        input=Input.Connection,
+    )
     cfg_scale: float | list[float] = InputField(default=1.0, description=FieldDescriptions.cfg_scale, title="CFG Scale")
     cfg_scale_start_step: int = InputField(
         default=0,
@@ -261,8 +266,19 @@ class FluxDenoiseInvocation(BaseInvocation, WithMetadata, WithBoard):
         if is_schnell and self.control_lora:
             raise ValueError("Control LoRAs cannot be used with FLUX Schnell")
 
-        # Prepare the extra image conditioning tensor if a FLUX structural control image is provided.
-        img_cond = self._prep_structural_control_img_cond(context)
+        # Prepare the extra image conditioning tensor (img_cond) for either FLUX structural control or FLUX Fill.
+        img_cond: torch.Tensor | None = None
+        is_flux_fill = transformer_config.variant == ModelVariantType.Inpaint  # type: ignore
+        if is_flux_fill:
+            img_cond = self._prep_flux_fill_img_cond(
+                context, device=TorchDevice.choose_torch_device(), dtype=inference_dtype
+            )
+        else:
+            if self.fill_conditioning is not None:
+                raise ValueError("fill_conditioning was provided, but the model is not a FLUX Fill model.")
+
+            if self.control_lora is not None:
+                img_cond = self._prep_structural_control_img_cond(context)
 
         inpaint_mask = self._prep_inpaint_mask(context, x)
 
@@ -271,7 +287,6 @@ class FluxDenoiseInvocation(BaseInvocation, WithMetadata, WithBoard):
         # Pack all latent tensors.
         init_latents = pack(init_latents) if init_latents is not None else None
         inpaint_mask = pack(inpaint_mask) if inpaint_mask is not None else None
-        img_cond = pack(img_cond) if img_cond is not None else None
         noise = pack(noise)
         x = pack(x)
 
@@ -280,10 +295,10 @@ class FluxDenoiseInvocation(BaseInvocation, WithMetadata, WithBoard):
         assert packed_h * packed_w == x.shape[1]
 
         # Prepare inpaint extension.
-        inpaint_extension: InpaintExtension | None = None
+        inpaint_extension: RectifiedFlowInpaintExtension | None = None
         if inpaint_mask is not None:
             assert init_latents is not None
-            inpaint_extension = InpaintExtension(
+            inpaint_extension = RectifiedFlowInpaintExtension(
                 init_latents=init_latents,
                 inpaint_mask=inpaint_mask,
                 noise=noise,
@@ -664,7 +679,70 @@ class FluxDenoiseInvocation(BaseInvocation, WithMetadata, WithBoard):
         img_cond = einops.rearrange(img_cond, "h w c -> 1 c h w")
 
         vae_info = context.models.load(self.controlnet_vae.vae)
-        return FluxVaeEncodeInvocation.vae_encode(vae_info=vae_info, image_tensor=img_cond)
+        img_cond = FluxVaeEncodeInvocation.vae_encode(vae_info=vae_info, image_tensor=img_cond)
+
+        return pack(img_cond)
+
+    def _prep_flux_fill_img_cond(
+        self, context: InvocationContext, device: torch.device, dtype: torch.dtype
+    ) -> torch.Tensor:
+        """Prepare the FLUX Fill conditioning. This method should be called iff the model is a FLUX Fill model.
+
+        This logic is based on:
+        https://github.com/black-forest-labs/flux/blob/716724eb276d94397be99710a0a54d352664e23b/src/flux/sampling.py#L107-L157
+        """
+        # Validate inputs.
+        if self.fill_conditioning is None:
+            raise ValueError("A FLUX Fill model is being used without fill_conditioning.")
+        # TODO(ryand): We should probable rename controlnet_vae. It's used for more than just ControlNets.
+        if self.controlnet_vae is None:
+            raise ValueError("A FLUX Fill model is being used without controlnet_vae.")
+        if self.control_lora is not None:
+            raise ValueError(
+                "A FLUX Fill model is being used, but a control_lora was provided. Control LoRAs are not compatible with FLUX Fill models."
+            )
+
+        # Log input warnings related to FLUX Fill usage.
+        if self.denoise_mask is not None:
+            context.logger.warning(
+                "Both fill_conditioning and a denoise_mask were provided. You probably meant to use one or the other."
+            )
+        if self.guidance < 25.0:
+            context.logger.warning("A guidance value of ~30.0 is recommended for FLUX Fill models.")
+
+        # Load the conditioning image and resize it to the target image size.
+        cond_img = context.images.get_pil(self.fill_conditioning.image.image_name, mode="RGB")
+        cond_img = cond_img.resize((self.width, self.height), Image.Resampling.BICUBIC)
+        cond_img = np.array(cond_img)
+        cond_img = torch.from_numpy(cond_img).float() / 127.5 - 1.0
+        cond_img = einops.rearrange(cond_img, "h w c -> 1 c h w")
+        cond_img = cond_img.to(device=device, dtype=dtype)
+
+        # Load the mask and resize it to the target image size.
+        mask = context.tensors.load(self.fill_conditioning.mask.tensor_name)
+        # We expect mask to be a bool tensor with shape [1, H, W].
+        assert mask.dtype == torch.bool
+        assert mask.dim() == 3
+        assert mask.shape[0] == 1
+        mask = tv_resize(mask, size=[self.height, self.width], interpolation=tv_transforms.InterpolationMode.NEAREST)
+        mask = mask.to(device=device, dtype=dtype)
+        mask = einops.rearrange(mask, "1 h w -> 1 1 h w")
+
+        # Prepare image conditioning.
+        cond_img = cond_img * (1 - mask)
+        vae_info = context.models.load(self.controlnet_vae.vae)
+        cond_img = FluxVaeEncodeInvocation.vae_encode(vae_info=vae_info, image_tensor=cond_img)
+        cond_img = pack(cond_img)
+
+        # Prepare mask conditioning.
+        mask = mask[:, 0, :, :]
+        # Rearrange mask to a 16-channel representation that matches the shape of the VAE-encoded latent space.
+        mask = einops.rearrange(mask, "b (h ph) (w pw) -> b (ph pw) h w", ph=8, pw=8)
+        mask = pack(mask)
+
+        # Merge image and mask conditioning.
+        img_cond = torch.cat((cond_img, mask), dim=-1)
+        return img_cond
 
     def _normalize_ip_adapter_fields(self) -> list[IPAdapterField]:
         if self.ip_adapter is None:

invokeai/app/invocations/flux_fill.py

@@ -0,0 +1,46 @@
+from invokeai.app.invocations.baseinvocation import (
+    BaseInvocation,
+    BaseInvocationOutput,
+    Classification,
+    invocation,
+    invocation_output,
+)
+from invokeai.app.invocations.fields import (
+    FieldDescriptions,
+    FluxFillConditioningField,
+    InputField,
+    OutputField,
+    TensorField,
+)
+from invokeai.app.invocations.primitives import ImageField
+from invokeai.app.services.shared.invocation_context import InvocationContext
+
+
+@invocation_output("flux_fill_output")
+class FluxFillOutput(BaseInvocationOutput):
+    """The conditioning output of a FLUX Fill invocation."""
+
+    fill_cond: FluxFillConditioningField = OutputField(
+        description=FieldDescriptions.flux_redux_conditioning, title="Conditioning"
+    )
+
+
+@invocation(
+    "flux_fill",
+    title="FLUX Fill Conditioning",
+    tags=["inpaint"],
+    category="inpaint",
+    version="1.0.0",
+    classification=Classification.Beta,
+)
+class FluxFillInvocation(BaseInvocation):
+    """Prepare the FLUX Fill conditioning data."""
+
+    image: ImageField = InputField(description="The FLUX Fill reference image.")
+    mask: TensorField = InputField(
+        description="The bool inpainting mask. Excluded regions should be set to "
+        "False, included regions should be set to True.",
+    )
+
+    def invoke(self, context: InvocationContext) -> FluxFillOutput:
+        return FluxFillOutput(fill_cond=FluxFillConditioningField(image=self.image, mask=self.mask))

invokeai/app/invocations/flux_ip_adapter.py

@@ -4,7 +4,7 @@ from typing import List, Literal, Union
 from pydantic import field_validator, model_validator
 from typing_extensions import Self
 
-from invokeai.app.invocations.baseinvocation import BaseInvocation, Classification, invocation
+from invokeai.app.invocations.baseinvocation import BaseInvocation, invocation
 from invokeai.app.invocations.fields import InputField, UIType
 from invokeai.app.invocations.ip_adapter import (
     CLIP_VISION_MODEL_MAP,
@@ -28,7 +28,6 @@ from invokeai.backend.model_manager.config import (
     tags=["ip_adapter", "control"],
     category="ip_adapter",
     version="1.0.0",
-    classification=Classification.Prototype,
 )
 class FluxIPAdapterInvocation(BaseInvocation):
     """Collects FLUX IP-Adapter info to pass to other nodes."""

invokeai/app/invocations/flux_lora_loader.py

@@ -3,14 +3,13 @@ from typing import Optional
 from invokeai.app.invocations.baseinvocation import (
     BaseInvocation,
     BaseInvocationOutput,
-    Classification,
     invocation,
     invocation_output,
 )
 from invokeai.app.invocations.fields import FieldDescriptions, Input, InputField, OutputField, UIType
 from invokeai.app.invocations.model import CLIPField, LoRAField, ModelIdentifierField, T5EncoderField, TransformerField
 from invokeai.app.services.shared.invocation_context import InvocationContext
-from invokeai.backend.model_manager.config import BaseModelType
+from invokeai.backend.model_manager.taxonomy import BaseModelType
 
 
 @invocation_output("flux_lora_loader_output")
@@ -28,11 +27,10 @@ class FluxLoRALoaderOutput(BaseInvocationOutput):
 
 @invocation(
     "flux_lora_loader",
-    title="FLUX LoRA",
+    title="Apply LoRA - FLUX",
     tags=["lora", "model", "flux"],
     category="model",
-    version="1.2.0",
-    classification=Classification.Prototype,
+    version="1.2.1",
 )
 class FluxLoRALoaderInvocation(BaseInvocation):
     """Apply a LoRA model to a FLUX transformer and/or text encoder."""
@@ -107,11 +105,10 @@ class FluxLoRALoaderInvocation(BaseInvocation):
 
 @invocation(
     "flux_lora_collection_loader",
-    title="FLUX LoRA Collection Loader",
+    title="Apply LoRA Collection - FLUX",
     tags=["lora", "model", "flux"],
     category="model",
-    version="1.3.0",
-    classification=Classification.Prototype,
+    version="1.3.1",
 )
 class FLUXLoRACollectionLoader(BaseInvocation):
     """Applies a collection of LoRAs to a FLUX transformer."""

invokeai/app/invocations/flux_model_loader.py

@@ -3,7 +3,6 @@ from typing import Literal
 from invokeai.app.invocations.baseinvocation import (
     BaseInvocation,
     BaseInvocationOutput,
-    Classification,
     invocation,
     invocation_output,
 )
@@ -17,8 +16,8 @@ from invokeai.app.util.t5_model_identifier import (
 from invokeai.backend.flux.util import max_seq_lengths
 from invokeai.backend.model_manager.config import (
     CheckpointConfigBase,
-    SubModelType,
 )
+from invokeai.backend.model_manager.taxonomy import SubModelType
 
 
 @invocation_output("flux_model_loader_output")
@@ -41,7 +40,6 @@ class FluxModelLoaderOutput(BaseInvocationOutput):
     tags=["model", "flux"],
     category="model",
     version="1.0.6",
-    classification=Classification.Prototype,
 )
 class FluxModelLoaderInvocation(BaseInvocation):
     """Loads a flux base model, outputting its submodels."""

invokeai/app/invocations/flux_redux.py

@@ -1,7 +1,9 @@
-from typing import Optional
+import math
+from typing import Literal, Optional
 
 import torch
 from PIL import Image
+from transformers import SiglipImageProcessor, SiglipVisionModel
 
 from invokeai.app.invocations.baseinvocation import (
     BaseInvocation,
@@ -23,7 +25,8 @@ from invokeai.app.invocations.primitives import ImageField
 from invokeai.app.services.model_records.model_records_base import ModelRecordChanges
 from invokeai.app.services.shared.invocation_context import InvocationContext
 from invokeai.backend.flux.redux.flux_redux_model import FluxReduxModel
-from invokeai.backend.model_manager.config import AnyModelConfig, BaseModelType, ModelType
+from invokeai.backend.model_manager import BaseModelType, ModelType
+from invokeai.backend.model_manager.config import AnyModelConfig
 from invokeai.backend.model_manager.starter_models import siglip
 from invokeai.backend.sig_lip.sig_lip_pipeline import SigLipPipeline
 from invokeai.backend.util.devices import TorchDevice
@@ -38,13 +41,16 @@ class FluxReduxOutput(BaseInvocationOutput):
     )
 
 
+DOWNSAMPLING_FUNCTIONS = Literal["nearest", "bilinear", "bicubic", "area", "nearest-exact"]
+
+
 @invocation(
     "flux_redux",
     title="FLUX Redux",
     tags=["ip_adapter", "control"],
     category="ip_adapter",
-    version="2.0.0",
-    classification=Classification.Prototype,
+    version="2.1.0",
+    classification=Classification.Beta,
 )
 class FluxReduxInvocation(BaseInvocation):
     """Runs a FLUX Redux model to generate a conditioning tensor."""
@@ -60,23 +66,64 @@ class FluxReduxInvocation(BaseInvocation):
         title="FLUX Redux Model",
         ui_type=UIType.FluxReduxModel,
     )
+    downsampling_factor: int = InputField(
+        ge=1,
+        le=9,
+        default=1,
+        description="Redux Downsampling Factor (1-9)",
+    )
+    downsampling_function: DOWNSAMPLING_FUNCTIONS = InputField(
+        default="area",
+        description="Redux Downsampling Function",
+    )
+    weight: float = InputField(
+        ge=0,
+        le=1,
+        default=1.0,
+        description="Redux weight (0.0-1.0)",
+    )
 
     def invoke(self, context: InvocationContext) -> FluxReduxOutput:
         image = context.images.get_pil(self.image.image_name, "RGB")
 
         encoded_x = self._siglip_encode(context, image)
         redux_conditioning = self._flux_redux_encode(context, encoded_x)
+        if self.downsampling_factor > 1 or self.weight != 1.0:
+            redux_conditioning = self._downsample_weight(context, redux_conditioning)
 
         tensor_name = context.tensors.save(redux_conditioning)
         return FluxReduxOutput(
             redux_cond=FluxReduxConditioningField(conditioning=TensorField(tensor_name=tensor_name), mask=self.mask)
         )
 
+    @torch.no_grad()
+    def _downsample_weight(self, context: InvocationContext, redux_conditioning: torch.Tensor) -> torch.Tensor:
+        # Downsampling derived from https://github.com/kaibioinfo/ComfyUI_AdvancedRefluxControl
+        (b, t, h) = redux_conditioning.shape
+        m = int(math.sqrt(t))
+        if self.downsampling_factor > 1:
+            redux_conditioning = redux_conditioning.view(b, m, m, h)
+            redux_conditioning = torch.nn.functional.interpolate(
+                redux_conditioning.transpose(1, -1),
+                size=(m // self.downsampling_factor, m // self.downsampling_factor),
+                mode=self.downsampling_function,
+            )
+            redux_conditioning = redux_conditioning.transpose(1, -1).reshape(b, -1, h)
+        if self.weight != 1.0:
+            redux_conditioning = redux_conditioning * self.weight * self.weight
+        return redux_conditioning
+
     @torch.no_grad()
     def _siglip_encode(self, context: InvocationContext, image: Image.Image) -> torch.Tensor:
         siglip_model_config = self._get_siglip_model(context)
-        with context.models.load(siglip_model_config.key).model_on_device() as (_, siglip_pipeline):
-            assert isinstance(siglip_pipeline, SigLipPipeline)
+        with context.models.load(siglip_model_config.key).model_on_device() as (_, model):
+            assert isinstance(model, SiglipVisionModel)
+
+            model_abs_path = context.models.get_absolute_path(siglip_model_config)
+            processor = SiglipImageProcessor.from_pretrained(model_abs_path, local_files_only=True)
+            assert isinstance(processor, SiglipImageProcessor)
+
+            siglip_pipeline = SigLipPipeline(processor, model)
             return siglip_pipeline.encode_image(
                 x=image, device=TorchDevice.choose_torch_device(), dtype=TorchDevice.choose_torch_dtype()
             )

invokeai/app/invocations/flux_text_encoder.py

@@ -4,7 +4,7 @@ from typing import Iterator, Literal, Optional, Tuple
 import torch
 from transformers import CLIPTextModel, CLIPTokenizer, T5EncoderModel, T5Tokenizer, T5TokenizerFast
 
-from invokeai.app.invocations.baseinvocation import BaseInvocation, Classification, invocation
+from invokeai.app.invocations.baseinvocation import BaseInvocation, invocation
 from invokeai.app.invocations.fields import (
     FieldDescriptions,
     FluxConditioningField,
@@ -17,7 +17,7 @@ from invokeai.app.invocations.model import CLIPField, T5EncoderField
 from invokeai.app.invocations.primitives import FluxConditioningOutput
 from invokeai.app.services.shared.invocation_context import InvocationContext
 from invokeai.backend.flux.modules.conditioner import HFEncoder
-from invokeai.backend.model_manager.config import ModelFormat
+from invokeai.backend.model_manager import ModelFormat
 from invokeai.backend.patches.layer_patcher import LayerPatcher
 from invokeai.backend.patches.lora_conversions.flux_lora_constants import FLUX_LORA_CLIP_PREFIX, FLUX_LORA_T5_PREFIX
 from invokeai.backend.patches.model_patch_raw import ModelPatchRaw
@@ -30,7 +30,6 @@ from invokeai.backend.stable_diffusion.diffusion.conditioning_data import Condit
     tags=["prompt", "conditioning", "flux"],
     category="conditioning",
     version="1.1.2",
-    classification=Classification.Prototype,
 )
 class FluxTextEncoderInvocation(BaseInvocation):
     """Encodes and preps a prompt for a flux image."""

invokeai/app/invocations/ideal_size.py

@@ -6,7 +6,7 @@ from invokeai.app.invocations.constants import LATENT_SCALE_FACTOR
 from invokeai.app.invocations.fields import FieldDescriptions, InputField, OutputField
 from invokeai.app.invocations.model import UNetField
 from invokeai.app.services.shared.invocation_context import InvocationContext
-from invokeai.backend.model_manager.config import BaseModelType
+from invokeai.backend.model_manager.taxonomy import BaseModelType
 
 
 @invocation_output("ideal_size_output")
@@ -21,14 +21,14 @@ class IdealSizeOutput(BaseInvocationOutput):
     "ideal_size",
     title="Ideal Size - SD1.5, SDXL",
     tags=["latents", "math", "ideal_size"],
-    version="1.0.5",
+    version="1.0.6",
 )
 class IdealSizeInvocation(BaseInvocation):
     """Calculates the ideal size for generation to avoid duplication"""
 
     width: int = InputField(default=1024, description="Final image width")
     height: int = InputField(default=576, description="Final image height")
-    unet: UNetField = InputField(default=None, description=FieldDescriptions.unet)
+    unet: UNetField = InputField(description=FieldDescriptions.unet)
     multiplier: float = InputField(
         default=1.0,
         description="Amount to multiply the model's dimensions by when calculating the ideal size (may result in "

invokeai/app/invocations/image.py

@@ -355,7 +355,6 @@ class ImageBlurInvocation(BaseInvocation, WithMetadata, WithBoard):
     tags=["image", "unsharp_mask"],
     category="image",
     version="1.2.2",
-    classification=Classification.Beta,
 )
 class UnsharpMaskInvocation(BaseInvocation, WithMetadata, WithBoard):
     """Applies an unsharp mask filter to an image"""
@@ -976,13 +975,13 @@ class SaveImageInvocation(BaseInvocation, WithMetadata, WithBoard):
     title="Canvas Paste Back",
     tags=["image", "combine"],
     category="image",
-    version="1.0.0",
+    version="1.0.1",
 )
 class CanvasPasteBackInvocation(BaseInvocation, WithMetadata, WithBoard):
     """Combines two images by using the mask provided. Intended for use on the Unified Canvas."""
 
     source_image: ImageField = InputField(description="The source image")
-    target_image: ImageField = InputField(default=None, description="The target image")
+    target_image: ImageField = InputField(description="The target image")
     mask: ImageField = InputField(
         description="The mask to use when pasting",
     )
@@ -1051,7 +1050,7 @@ class MaskFromIDInvocation(BaseInvocation, WithMetadata, WithBoard):
     tags=["image", "mask", "id"],
     category="image",
     version="1.0.0",
-    classification=Classification.Internal,
+    classification=Classification.Deprecated,
 )
 class CanvasV2MaskAndCropInvocation(BaseInvocation, WithMetadata, WithBoard):
     """Handles Canvas V2 image output masking and cropping"""
@@ -1089,17 +1088,145 @@ class CanvasV2MaskAndCropInvocation(BaseInvocation, WithMetadata, WithBoard):
         return ImageOutput.build(image_dto)
 
 
+@invocation(
+    "expand_mask_with_fade", title="Expand Mask with Fade", tags=["image", "mask"], category="image", version="1.0.1"
+)
+class ExpandMaskWithFadeInvocation(BaseInvocation, WithMetadata, WithBoard):
+    """Expands a mask with a fade effect. The mask uses black to indicate areas to keep from the generated image and white for areas to discard.
+    The mask is thresholded to create a binary mask, and then a distance transform is applied to create a fade effect.
+    The fade size is specified in pixels, and the mask is expanded by that amount. The result is a mask with a smooth transition from black to white.
+    If the fade size is 0, the mask is returned as-is.
+    """
+
+    mask: ImageField = InputField(description="The mask to expand")
+    threshold: int = InputField(default=0, ge=0, le=255, description="The threshold for the binary mask (0-255)")
+    fade_size_px: int = InputField(default=32, ge=0, description="The size of the fade in pixels")
+
+    def invoke(self, context: InvocationContext) -> ImageOutput:
+        pil_mask = context.images.get_pil(self.mask.image_name, mode="L")
+
+        if self.fade_size_px == 0:
+            # If the fade size is 0, just return the mask as-is.
+            image_dto = context.images.save(image=pil_mask, image_category=ImageCategory.MASK)
+            return ImageOutput.build(image_dto)
+
+        np_mask = numpy.array(pil_mask)
+
+        # Threshold the mask to create a binary mask - 0 for black, 255 for white
+        # If we don't threshold we can get some weird artifacts
+        np_mask = numpy.where(np_mask > self.threshold, 255, 0).astype(numpy.uint8)
+
+        # Create a mask for the black region (1 where black, 0 otherwise)
+        black_mask = (np_mask == 0).astype(numpy.uint8)
+
+        # Invert the black region
+        bg_mask = 1 - black_mask
+
+        # Create a distance transform of the inverted mask
+        dist = cv2.distanceTransform(bg_mask, cv2.DIST_L2, 5)
+
+        # Normalize distances so that pixels <fade_size_px become a linear gradient (0 to 1)
+        d_norm = numpy.clip(dist / self.fade_size_px, 0, 1)
+
+        # Control points: x values (normalized distance) and corresponding fade pct y values.
+
+        # There are some magic numbers here that are used to create a smooth transition:
+        # - The first point is at 0% of fade size from edge of mask (meaning the edge of the mask), and is 0% fade (black)
+        # - The second point is 1px from the edge of the mask and also has 0% fade, effectively expanding the mask
+        #   by 1px. This fixes an issue where artifacts can occur at the edge of the mask
+        # - The third point is at 20% of the fade size from the edge of the mask and has 20% fade
+        # - The fourth point is at 80% of the fade size from the edge of the mask and has 90% fade
+        # - The last point is at 100% of the fade size from the edge of the mask and has 100% fade (white)
+
+        # x values: 0 = mask edge, 1 = fade_size_px from edge
+        x_control = numpy.array([0.0, 1.0 / self.fade_size_px, 0.2, 0.8, 1.0])
+        # y values: 0 = black, 1 = white
+        y_control = numpy.array([0.0, 0.0, 0.2, 0.9, 1.0])
+
+        # Fit a cubic polynomial that smoothly passes through the control points
+        coeffs = numpy.polyfit(x_control, y_control, 3)
+        poly = numpy.poly1d(coeffs)
+
+        # Evaluate the polynomial
+        feather = poly(d_norm)
+
+        # The polynomial fit isn't perfect. Points beyond the fade distance are likely to be slightly less than 1.0,
+        # even though the control points indicate that they should be exactly 1.0. This is due to the nature of the
+        # polynomial fit, which is a best approximation of the control points but not an exact match.
+
+        # When this occurs, the area outside the mask and fade-out will not be 100% transparent. For example, it may
+        # have an alpha value of 1 instead of 0. So we must force pixels at or beyond the fade distance to exactly 1.0.
+
+        # Force pixels at or beyond the fade distance to exactly 1.0
+        feather = numpy.where(d_norm >= 1.0, 1.0, feather)
+
+        # Clip any other values to ensure they're in the valid range [0,1]
+        feather = numpy.clip(feather, 0, 1)
+
+        # Build final image.
+        np_result = numpy.where(black_mask == 1, 0, (feather * 255).astype(numpy.uint8))
+
+        # Convert back to PIL, grayscale
+        pil_result = Image.fromarray(np_result.astype(numpy.uint8), mode="L")
+
+        image_dto = context.images.save(image=pil_result, image_category=ImageCategory.MASK)
+
+        return ImageOutput.build(image_dto)
+
+
+@invocation(
+    "apply_mask_to_image",
+    title="Apply Mask to Image",
+    tags=["image", "mask", "blend"],
+    category="image",
+    version="1.0.0",
+)
+class ApplyMaskToImageInvocation(BaseInvocation, WithMetadata, WithBoard):
+    """
+    Extracts a region from a generated image using a mask and blends it seamlessly onto a source image.
+    The mask uses black to indicate areas to keep from the generated image and white for areas to discard.
+    """
+
+    image: ImageField = InputField(description="The image from which to extract the masked region")
+    mask: ImageField = InputField(description="The mask defining the region (black=keep, white=discard)")
+    invert_mask: bool = InputField(
+        default=False,
+        description="Whether to invert the mask before applying it",
+    )
+
+    def invoke(self, context: InvocationContext) -> ImageOutput:
+        # Load images
+        image = context.images.get_pil(self.image.image_name, mode="RGBA")
+        mask = context.images.get_pil(self.mask.image_name, mode="L")
+
+        if self.invert_mask:
+            # Invert the mask if requested
+            mask = ImageOps.invert(mask.copy())
+
+        # Combine the mask as the alpha channel of the image
+        r, g, b, _ = image.split()  # Split the image into RGB and alpha channels
+        result_image = Image.merge("RGBA", (r, g, b, mask))  # Use the mask as the new alpha channel
+
+        # Save the resulting image
+        image_dto = context.images.save(image=result_image)
+
+        return ImageOutput.build(image_dto)
+
+
 @invocation(
     "img_noise",
     title="Add Image Noise",
     tags=["image", "noise"],
     category="image",
-    version="1.0.1",
+    version="1.1.0",
 )
 class ImageNoiseInvocation(BaseInvocation, WithMetadata, WithBoard):
     """Add noise to an image"""
 
     image: ImageField = InputField(description="The image to add noise to")
+    mask: Optional[ImageField] = InputField(
+        default=None, description="Optional mask determining where to apply noise (black=noise, white=no noise)"
+    )
     seed: int = InputField(
         default=0,
         ge=0,
@@ -1143,12 +1270,27 @@ class ImageNoiseInvocation(BaseInvocation, WithMetadata, WithBoard):
         noise = Image.fromarray(noise.astype(numpy.uint8), mode="RGB").resize(
             (image.width, image.height), Image.Resampling.NEAREST
         )
+
+        # Create a noisy version of the input image
         noisy_image = Image.blend(image.convert("RGB"), noise, self.amount).convert("RGBA")
 
-        # Paste back the alpha channel
-        noisy_image.putalpha(alpha)
+        # Apply mask if provided
+        if self.mask is not None:
+            mask_image = context.images.get_pil(self.mask.image_name, mode="L")
 
-        image_dto = context.images.save(image=noisy_image)
+            if mask_image.size != image.size:
+                mask_image = mask_image.resize(image.size, Image.Resampling.LANCZOS)
+
+            result_image = image.copy()
+            mask_image = ImageOps.invert(mask_image)
+            result_image.paste(noisy_image, (0, 0), mask=mask_image)
+        else:
+            result_image = noisy_image
+
+        # Paste back the alpha channel from the original image
+        result_image.putalpha(alpha)
+
+        image_dto = context.images.save(image=result_image)
 
         return ImageOutput.build(image_dto)
 
@@ -1159,7 +1301,6 @@ class ImageNoiseInvocation(BaseInvocation, WithMetadata, WithBoard):
     category="image",
     version="1.0.0",
     tags=["image", "crop"],
-    classification=Classification.Beta,
 )
 class CropImageToBoundingBoxInvocation(BaseInvocation, WithMetadata, WithBoard):
     """Crop an image to the given bounding box. If the bounding box is omitted, the image is cropped to the non-transparent pixels."""
@@ -1186,7 +1327,6 @@ class CropImageToBoundingBoxInvocation(BaseInvocation, WithMetadata, WithBoard):
     category="image",
     version="1.0.0",
     tags=["image", "crop"],
-    classification=Classification.Beta,
 )
 class PasteImageIntoBoundingBoxInvocation(BaseInvocation, WithMetadata, WithBoard):
     """Paste the source image into the target image at the given bounding box.

invokeai/app/invocations/infill.py

@@ -127,13 +127,16 @@ class InfillPatchMatchInvocation(InfillImageProcessorInvocation):
         return infilled
 
 
+LAMA_MODEL_URL = "https://github.com/Sanster/models/releases/download/add_big_lama/big-lama.pt"
+
+
 @invocation("infill_lama", title="LaMa Infill", tags=["image", "inpaint"], category="inpaint", version="1.2.2")
 class LaMaInfillInvocation(InfillImageProcessorInvocation):
     """Infills transparent areas of an image using the LaMa model"""
 
     def infill(self, image: Image.Image):
         with self._context.models.load_remote_model(
-            source="https://github.com/Sanster/models/releases/download/add_big_lama/big-lama.pt",
+            source=LAMA_MODEL_URL,
             loader=LaMA.load_jit_model,
         ) as model:
             lama = LaMA(model)

invokeai/app/invocations/ip_adapter.py

@@ -13,10 +13,8 @@ from invokeai.app.services.model_records.model_records_base import ModelRecordCh
 from invokeai.app.services.shared.invocation_context import InvocationContext
 from invokeai.backend.model_manager.config import (
     AnyModelConfig,
-    BaseModelType,
     IPAdapterCheckpointConfig,
     IPAdapterInvokeAIConfig,
-    ModelType,
 )
 from invokeai.backend.model_manager.starter_models import (
     StarterModel,
@@ -24,6 +22,7 @@ from invokeai.backend.model_manager.starter_models import (
     ip_adapter_sd_image_encoder,
     ip_adapter_sdxl_image_encoder,
 )
+from invokeai.backend.model_manager.taxonomy import BaseModelType, ModelType
 
 
 class IPAdapterField(BaseModel):
@@ -32,6 +31,7 @@ class IPAdapterField(BaseModel):
     image_encoder_model: ModelIdentifierField = Field(description="The name of the CLIP image encoder model.")
     weight: Union[float, List[float]] = Field(default=1, description="The weight given to the IP-Adapter.")
     target_blocks: List[str] = Field(default=[], description="The IP Adapter blocks to apply")
+    method: str = Field(default="full", description="Weight apply method")
     begin_step_percent: float = Field(
         default=0, ge=0, le=1, description="When the IP-Adapter is first applied (% of total steps)"
     )
@@ -95,7 +95,7 @@ class IPAdapterInvocation(BaseInvocation):
     weight: Union[float, List[float]] = InputField(
         default=1, description="The weight given to the IP-Adapter", title="Weight"
     )
-    method: Literal["full", "style", "composition"] = InputField(
+    method: Literal["full", "style", "composition", "style_strong", "style_precise"] = InputField(
         default="full", description="The method to apply the IP-Adapter"
     )
     begin_step_percent: float = InputField(
@@ -148,6 +148,38 @@ class IPAdapterInvocation(BaseInvocation):
                 target_blocks = ["down_blocks.2.attentions.1"]
             else:
                 raise ValueError(f"Unsupported IP-Adapter base type: '{ip_adapter_info.base}'.")
+        elif self.method == "style_precise":
+            if ip_adapter_info.base == "sd-1":
+                target_blocks = ["up_blocks.1", "down_blocks.2", "mid_block"]
+            elif ip_adapter_info.base == "sdxl":
+                target_blocks = ["up_blocks.0.attentions.1", "down_blocks.2.attentions.1"]
+            else:
+                raise ValueError(f"Unsupported IP-Adapter base type: '{ip_adapter_info.base}'.")
+        elif self.method == "style_strong":
+            if ip_adapter_info.base == "sd-1":
+                target_blocks = ["up_blocks.0", "up_blocks.1", "up_blocks.2", "down_blocks.0", "down_blocks.1"]
+            elif ip_adapter_info.base == "sdxl":
+                target_blocks = [
+                    "up_blocks.0.attentions.1",
+                    "up_blocks.1.attentions.1",
+                    "up_blocks.2.attentions.1",
+                    "up_blocks.0.attentions.2",
+                    "up_blocks.1.attentions.2",
+                    "up_blocks.2.attentions.2",
+                    "up_blocks.0.attentions.0",
+                    "up_blocks.1.attentions.0",
+                    "up_blocks.2.attentions.0",
+                    "down_blocks.0.attentions.0",
+                    "down_blocks.0.attentions.1",
+                    "down_blocks.0.attentions.2",
+                    "down_blocks.1.attentions.0",
+                    "down_blocks.1.attentions.1",
+                    "down_blocks.1.attentions.2",
+                    "down_blocks.2.attentions.0",
+                    "down_blocks.2.attentions.2",
+                ]
+            else:
+                raise ValueError(f"Unsupported IP-Adapter base type: '{ip_adapter_info.base}'.")
         elif self.method == "full":
             target_blocks = ["block"]
         else:
@@ -163,6 +195,7 @@ class IPAdapterInvocation(BaseInvocation):
                 begin_step_percent=self.begin_step_percent,
                 end_step_percent=self.end_step_percent,
                 mask=self.mask,
+                method=self.method,
             ),
         )
 

invokeai/app/invocations/llava_onevision_vllm.py

@@ -0,0 +1,75 @@
+from typing import Any
+
+import torch
+from PIL.Image import Image
+from pydantic import field_validator
+from transformers import AutoProcessor, LlavaOnevisionForConditionalGeneration, LlavaOnevisionProcessor
+
+from invokeai.app.invocations.baseinvocation import BaseInvocation, Classification, invocation
+from invokeai.app.invocations.fields import FieldDescriptions, ImageField, InputField, UIComponent, UIType
+from invokeai.app.invocations.model import ModelIdentifierField
+from invokeai.app.invocations.primitives import StringOutput
+from invokeai.app.services.shared.invocation_context import InvocationContext
+from invokeai.backend.llava_onevision_pipeline import LlavaOnevisionPipeline
+from invokeai.backend.util.devices import TorchDevice
+
+
+@invocation(
+    "llava_onevision_vllm",
+    title="LLaVA OneVision VLLM",
+    tags=["vllm"],
+    category="vllm",
+    version="1.0.0",
+    classification=Classification.Beta,
+)
+class LlavaOnevisionVllmInvocation(BaseInvocation):
+    """Run a LLaVA OneVision VLLM model."""
+
+    images: list[ImageField] | ImageField | None = InputField(default=None, max_length=3, description="Input image.")
+    prompt: str = InputField(
+        default="",
+        description="Input text prompt.",
+        ui_component=UIComponent.Textarea,
+    )
+    vllm_model: ModelIdentifierField = InputField(
+        title="LLaVA Model Type",
+        description=FieldDescriptions.vllm_model,
+        ui_type=UIType.LlavaOnevisionModel,
+    )
+
+    @field_validator("images", mode="before")
+    def listify_images(cls, v: Any) -> list:
+        if v is None:
+            return v
+        if not isinstance(v, list):
+            return [v]
+        return v
+
+    def _get_images(self, context: InvocationContext) -> list[Image]:
+        if self.images is None:
+            return []
+
+        image_fields = self.images if isinstance(self.images, list) else [self.images]
+        return [context.images.get_pil(image_field.image_name, "RGB") for image_field in image_fields]
+
+    @torch.no_grad()
+    def invoke(self, context: InvocationContext) -> StringOutput:
+        images = self._get_images(context)
+        model_config = context.models.get_config(self.vllm_model)
+
+        with context.models.load(self.vllm_model).model_on_device() as (_, model):
+            assert isinstance(model, LlavaOnevisionForConditionalGeneration)
+
+            model_abs_path = context.models.get_absolute_path(model_config)
+            processor = AutoProcessor.from_pretrained(model_abs_path, local_files_only=True)
+            assert isinstance(processor, LlavaOnevisionProcessor)
+
+            model = LlavaOnevisionPipeline(model, processor)
+            output = model.run(
+                prompt=self.prompt,
+                images=images,
+                device=TorchDevice.choose_torch_device(),
+                dtype=TorchDevice.choose_torch_dtype(),
+            )
+
+        return StringOutput(value=output)

invokeai/app/invocations/mask.py

@@ -4,7 +4,6 @@ from PIL import Image
 
 from invokeai.app.invocations.baseinvocation import (
     BaseInvocation,
-    Classification,
     InvocationContext,
     invocation,
 )
@@ -58,7 +57,6 @@ class RectangleMaskInvocation(BaseInvocation, WithMetadata):
     tags=["conditioning"],
     category="conditioning",
     version="1.0.0",
-    classification=Classification.Beta,
 )
 class AlphaMaskToTensorInvocation(BaseInvocation):
     """Convert a mask image to a tensor. Opaque regions are 1 and transparent regions are 0."""
@@ -67,7 +65,7 @@ class AlphaMaskToTensorInvocation(BaseInvocation):
     invert: bool = InputField(default=False, description="Whether to invert the mask.")
 
     def invoke(self, context: InvocationContext) -> MaskOutput:
-        image = context.images.get_pil(self.image.image_name)
+        image = context.images.get_pil(self.image.image_name, mode="RGBA")
         mask = torch.zeros((1, image.height, image.width), dtype=torch.bool)
         if self.invert:
             mask[0] = torch.tensor(np.array(image)[:, :, 3] == 0, dtype=torch.bool)
@@ -87,7 +85,6 @@ class AlphaMaskToTensorInvocation(BaseInvocation):
     tags=["conditioning"],
     category="conditioning",
     version="1.1.0",
-    classification=Classification.Beta,
 )
 class InvertTensorMaskInvocation(BaseInvocation):
     """Inverts a tensor mask."""
@@ -234,7 +231,6 @@ WHITE = ColorField(r=255, g=255, b=255, a=255)
     tags=["mask"],
     category="mask",
     version="1.0.0",
-    classification=Classification.Beta,
 )
 class GetMaskBoundingBoxInvocation(BaseInvocation):
     """Gets the bounding box of the given mask image."""

invokeai/app/invocations/metadata.py

@@ -42,7 +42,9 @@ class IPAdapterMetadataField(BaseModel):
     image: ImageField = Field(description="The IP-Adapter image prompt.")
     ip_adapter_model: ModelIdentifierField = Field(description="The IP-Adapter model.")
     clip_vision_model: Literal["ViT-L", "ViT-H", "ViT-G"] = Field(description="The CLIP Vision model")
-    method: Literal["full", "style", "composition"] = Field(description="Method to apply IP Weights with")
+    method: Literal["full", "style", "composition", "style_strong", "style_precise"] = Field(
+        description="Method to apply IP Weights with"
+    )
     weight: Union[float, list[float]] = Field(description="The weight given to the IP-Adapter")
     begin_step_percent: float = Field(description="When the IP-Adapter is first applied (% of total steps)")
     end_step_percent: float = Field(description="When the IP-Adapter is last applied (% of total steps)")
@@ -152,6 +154,10 @@ GENERATION_MODES = Literal[
     "sd3_img2img",
     "sd3_inpaint",
     "sd3_outpaint",
+    "cogview4_txt2img",
+    "cogview4_img2img",
+    "cogview4_inpaint",
+    "cogview4_outpaint",
 ]
 
 

invokeai/app/invocations/metadata_linked.py

@@ -14,7 +14,7 @@ from invokeai.app.invocations.baseinvocation import (
     invocation,
     invocation_output,
 )
-from invokeai.app.invocations.controlnet_image_processors import ControlField, ControlNetInvocation
+from invokeai.app.invocations.controlnet import ControlField, ControlNetInvocation
 from invokeai.app.invocations.denoise_latents import DenoiseLatentsInvocation
 from invokeai.app.invocations.fields import (
     FieldDescriptions,
@@ -39,11 +39,21 @@ from invokeai.app.invocations.model import (
     VAEField,
     VAEOutput,
 )
-from invokeai.app.invocations.primitives import BooleanOutput, FloatOutput, IntegerOutput, LatentsOutput, StringOutput
+from invokeai.app.invocations.primitives import (
+    BooleanCollectionOutput,
+    BooleanOutput,
+    FloatCollectionOutput,
+    FloatOutput,
+    IntegerCollectionOutput,
+    IntegerOutput,
+    LatentsOutput,
+    StringCollectionOutput,
+    StringOutput,
+)
 from invokeai.app.invocations.scheduler import SchedulerOutput
 from invokeai.app.invocations.t2i_adapter import T2IAdapterField, T2IAdapterInvocation
 from invokeai.app.services.shared.invocation_context import InvocationContext
-from invokeai.backend.model_manager.config import ModelType, SubModelType
+from invokeai.backend.model_manager.taxonomy import ModelType, SubModelType
 from invokeai.backend.stable_diffusion.schedulers.schedulers import SCHEDULER_NAME_VALUES
 from invokeai.version import __version__
 
@@ -1162,3 +1172,133 @@ class MetadataToT2IAdaptersInvocation(BaseInvocation, WithMetadata):
             adapters = append_list(T2IAdapterField, i.t2i_adapter, adapters)
 
         return MDT2IAdapterListOutput(t2i_adapter_list=adapters)
+
+
+@invocation(
+    "metadata_to_string_collection",
+    title="Metadata To String Collection",
+    tags=["metadata"],
+    category="metadata",
+    version="1.0.0",
+    classification=Classification.Beta,
+)
+class MetadataToStringCollectionInvocation(BaseInvocation, WithMetadata):
+    """Extracts a string collection value of a label from metadata"""
+
+    label: CORE_LABELS_STRING = InputField(
+        default=CUSTOM_LABEL,
+        description=FieldDescriptions.metadata_item_label,
+        input=Input.Direct,
+    )
+    custom_label: Optional[str] = InputField(
+        default=None,
+        description=FieldDescriptions.metadata_item_label,
+        input=Input.Direct,
+    )
+    default_value: list[str] = InputField(
+        description="The default string collection to use if not found in the metadata"
+    )
+
+    _validate_custom_label = model_validator(mode="after")(validate_custom_label)
+
+    def invoke(self, context: InvocationContext) -> StringCollectionOutput:
+        data: Dict[str, Any] = {} if self.metadata is None else self.metadata.root
+        output = data.get(str(self.custom_label if self.label == CUSTOM_LABEL else self.label), self.default_value)
+
+        return StringCollectionOutput(collection=output)
+
+
+@invocation(
+    "metadata_to_integer_collection",
+    title="Metadata To Integer Collection",
+    tags=["metadata"],
+    category="metadata",
+    version="1.0.0",
+    classification=Classification.Beta,
+)
+class MetadataToIntegerCollectionInvocation(BaseInvocation, WithMetadata):
+    """Extracts an integer value Collection of a label from metadata"""
+
+    label: CORE_LABELS_INTEGER = InputField(
+        default=CUSTOM_LABEL,
+        description=FieldDescriptions.metadata_item_label,
+        input=Input.Direct,
+    )
+    custom_label: Optional[str] = InputField(
+        default=None,
+        description=FieldDescriptions.metadata_item_label,
+        input=Input.Direct,
+    )
+    default_value: list[int] = InputField(description="The default integer to use if not found in the metadata")
+
+    _validate_custom_label = model_validator(mode="after")(validate_custom_label)
+
+    def invoke(self, context: InvocationContext) -> IntegerCollectionOutput:
+        data: Dict[str, Any] = {} if self.metadata is None else self.metadata.root
+        output = data.get(str(self.custom_label if self.label == CUSTOM_LABEL else self.label), self.default_value)
+
+        return IntegerCollectionOutput(collection=output)
+
+
+@invocation(
+    "metadata_to_float_collection",
+    title="Metadata To Float Collection",
+    tags=["metadata"],
+    category="metadata",
+    version="1.0.0",
+    classification=Classification.Beta,
+)
+class MetadataToFloatCollectionInvocation(BaseInvocation, WithMetadata):
+    """Extracts a Float value Collection of a label from metadata"""
+
+    label: CORE_LABELS_FLOAT = InputField(
+        default=CUSTOM_LABEL,
+        description=FieldDescriptions.metadata_item_label,
+        input=Input.Direct,
+    )
+    custom_label: Optional[str] = InputField(
+        default=None,
+        description=FieldDescriptions.metadata_item_label,
+        input=Input.Direct,
+    )
+    default_value: list[float] = InputField(description="The default float to use if not found in the metadata")
+
+    _validate_custom_label = model_validator(mode="after")(validate_custom_label)
+
+    def invoke(self, context: InvocationContext) -> FloatCollectionOutput:
+        data: Dict[str, Any] = {} if self.metadata is None else self.metadata.root
+        output = data.get(str(self.custom_label if self.label == CUSTOM_LABEL else self.label), self.default_value)
+
+        return FloatCollectionOutput(collection=output)
+
+
+@invocation(
+    "metadata_to_bool_collection",
+    title="Metadata To Bool Collection",
+    tags=["metadata"],
+    category="metadata",
+    version="1.0.0",
+    classification=Classification.Beta,
+)
+class MetadataToBoolCollectionInvocation(BaseInvocation, WithMetadata):
+    """Extracts a Boolean value Collection of a label from metadata"""
+
+    label: CORE_LABELS_BOOL = InputField(
+        default=CUSTOM_LABEL,
+        description=FieldDescriptions.metadata_item_label,
+        input=Input.Direct,
+    )
+    custom_label: Optional[str] = InputField(
+        default=None,
+        description=FieldDescriptions.metadata_item_label,
+        input=Input.Direct,
+    )
+    default_value: list[bool] = InputField(description="The default bool to use if not found in the metadata")
+
+    _validate_custom_label = model_validator(mode="after")(validate_custom_label)
+
+    def invoke(self, context: InvocationContext) -> BooleanCollectionOutput:
+        data: Dict[str, Any] = {} if self.metadata is None else self.metadata.root
+        output = data.get(str(self.custom_label if self.label == CUSTOM_LABEL else self.label), self.default_value)
+
+        return BooleanCollectionOutput(collection=output)

invokeai/app/invocations/model.py

@@ -6,7 +6,6 @@ from pydantic import BaseModel, Field
 from invokeai.app.invocations.baseinvocation import (
     BaseInvocation,
     BaseInvocationOutput,
-    Classification,
     invocation,
     invocation_output,
 )
@@ -15,10 +14,8 @@ from invokeai.app.services.shared.invocation_context import InvocationContext
 from invokeai.app.shared.models import FreeUConfig
 from invokeai.backend.model_manager.config import (
     AnyModelConfig,
-    BaseModelType,
-    ModelType,
-    SubModelType,
 )
+from invokeai.backend.model_manager.taxonomy import BaseModelType, ModelType, SubModelType
 
 
 class ModelIdentifierField(BaseModel):
@@ -71,6 +68,11 @@ class T5EncoderField(BaseModel):
     loras: List[LoRAField] = Field(description="LoRAs to apply on model loading")
 
 
+class GlmEncoderField(BaseModel):
+    tokenizer: ModelIdentifierField = Field(description="Info to load tokenizer submodel")
+    text_encoder: ModelIdentifierField = Field(description="Info to load text_encoder submodel")
+
+
 class VAEField(BaseModel):
     vae: ModelIdentifierField = Field(description="Info to load vae submodel")
     seamless_axes: List[str] = Field(default_factory=list, description='Axes("x" and "y") to which apply seamless')
@@ -126,7 +128,6 @@ class ModelIdentifierOutput(BaseInvocationOutput):
     tags=["model"],
     category="model",
     version="1.0.1",
-    classification=Classification.Prototype,
 )
 class ModelIdentifierInvocation(BaseInvocation):
     """Selects any model, outputting it its identifier. Be careful with this one! The identifier will be accepted as
@@ -181,7 +182,7 @@ class LoRALoaderOutput(BaseInvocationOutput):
     clip: Optional[CLIPField] = OutputField(default=None, description=FieldDescriptions.clip, title="CLIP")
 
 
-@invocation("lora_loader", title="LoRA", tags=["model"], category="model", version="1.0.3")
+@invocation("lora_loader", title="Apply LoRA - SD1.5", tags=["model"], category="model", version="1.0.4")
 class LoRALoaderInvocation(BaseInvocation):
     """Apply selected lora to unet and text_encoder."""
 
@@ -244,7 +245,7 @@ class LoRASelectorOutput(BaseInvocationOutput):
     lora: LoRAField = OutputField(description="LoRA model and weight", title="LoRA")
 
 
-@invocation("lora_selector", title="LoRA Model - SD1.5", tags=["model"], category="model", version="1.0.2")
+@invocation("lora_selector", title="Select LoRA", tags=["model"], category="model", version="1.0.3")
 class LoRASelectorInvocation(BaseInvocation):
     """Selects a LoRA model and weight."""
 
@@ -258,7 +259,7 @@ class LoRASelectorInvocation(BaseInvocation):
 
 
 @invocation(
-    "lora_collection_loader", title="LoRA Collection - SD1.5", tags=["model"], category="model", version="1.1.1"
+    "lora_collection_loader", title="Apply LoRA Collection - SD1.5", tags=["model"], category="model", version="1.1.2"
 )
 class LoRACollectionLoader(BaseInvocation):
     """Applies a collection of LoRAs to the provided UNet and CLIP models."""
@@ -322,10 +323,10 @@ class SDXLLoRALoaderOutput(BaseInvocationOutput):
 
 @invocation(
     "sdxl_lora_loader",
-    title="LoRA Model - SDXL",
+    title="Apply LoRA - SDXL",
     tags=["lora", "model"],
     category="model",
-    version="1.0.4",
+    version="1.0.5",
 )
 class SDXLLoRALoaderInvocation(BaseInvocation):
     """Apply selected lora to unet and text_encoder."""
@@ -402,10 +403,10 @@ class SDXLLoRALoaderInvocation(BaseInvocation):
 
 @invocation(
     "sdxl_lora_collection_loader",
-    title="LoRA Collection - SDXL",
+    title="Apply LoRA Collection - SDXL",
     tags=["model"],
     category="model",
-    version="1.1.1",
+    version="1.1.2",
 )
 class SDXLLoRACollectionLoader(BaseInvocation):
     """Applies a collection of SDXL LoRAs to the provided UNet and CLIP models."""

invokeai/app/invocations/primitives.py

@@ -13,6 +13,7 @@ from invokeai.app.invocations.baseinvocation import (
 from invokeai.app.invocations.constants import LATENT_SCALE_FACTOR
 from invokeai.app.invocations.fields import (
     BoundingBoxField,
+    CogView4ConditioningField,
     ColorField,
     ConditioningField,
     DenoiseMaskField,
@@ -440,6 +441,17 @@ class SD3ConditioningOutput(BaseInvocationOutput):
         return cls(conditioning=SD3ConditioningField(conditioning_name=conditioning_name))
 
 
+@invocation_output("cogview4_conditioning_output")
+class CogView4ConditioningOutput(BaseInvocationOutput):
+    """Base class for nodes that output a CogView text conditioning tensor."""
+
+    conditioning: CogView4ConditioningField = OutputField(description=FieldDescriptions.cond)
+
+    @classmethod
+    def build(cls, conditioning_name: str) -> "CogView4ConditioningOutput":
+        return cls(conditioning=CogView4ConditioningField(conditioning_name=conditioning_name))
+
+
 @invocation_output("conditioning_output")
 class ConditioningOutput(BaseInvocationOutput):
     """Base class for nodes that output a single conditioning tensor"""

invokeai/app/invocations/sd3_denoise.py

@@ -6,7 +6,7 @@ from diffusers.models.transformers.transformer_sd3 import SD3Transformer2DModel
 from torchvision.transforms.functional import resize as tv_resize
 from tqdm import tqdm
 
-from invokeai.app.invocations.baseinvocation import BaseInvocation, Classification, invocation
+from invokeai.app.invocations.baseinvocation import BaseInvocation, invocation
 from invokeai.app.invocations.constants import LATENT_SCALE_FACTOR
 from invokeai.app.invocations.fields import (
     DenoiseMaskField,
@@ -23,8 +23,8 @@ from invokeai.app.invocations.primitives import LatentsOutput
 from invokeai.app.invocations.sd3_text_encoder import SD3_T5_MAX_SEQ_LEN
 from invokeai.app.services.shared.invocation_context import InvocationContext
 from invokeai.backend.flux.sampling_utils import clip_timestep_schedule_fractional
-from invokeai.backend.model_manager.config import BaseModelType
-from invokeai.backend.sd3.extensions.inpaint_extension import InpaintExtension
+from invokeai.backend.model_manager import BaseModelType
+from invokeai.backend.rectified_flow.rectified_flow_inpaint_extension import RectifiedFlowInpaintExtension
 from invokeai.backend.stable_diffusion.diffusers_pipeline import PipelineIntermediateState
 from invokeai.backend.stable_diffusion.diffusion.conditioning_data import SD3ConditioningInfo
 from invokeai.backend.util.devices import TorchDevice
@@ -36,7 +36,6 @@ from invokeai.backend.util.devices import TorchDevice
     tags=["image", "sd3"],
     category="image",
     version="1.1.1",
-    classification=Classification.Prototype,
 )
 class SD3DenoiseInvocation(BaseInvocation, WithMetadata, WithBoard):
     """Run denoising process with a SD3 model."""
@@ -264,10 +263,10 @@ class SD3DenoiseInvocation(BaseInvocation, WithMetadata, WithBoard):
 
         # Prepare inpaint extension.
         inpaint_mask = self._prep_inpaint_mask(context, latents)
-        inpaint_extension: InpaintExtension | None = None
+        inpaint_extension: RectifiedFlowInpaintExtension | None = None
         if inpaint_mask is not None:
             assert init_latents is not None
-            inpaint_extension = InpaintExtension(
+            inpaint_extension = RectifiedFlowInpaintExtension(
                 init_latents=init_latents,
                 inpaint_mask=inpaint_mask,
                 noise=noise,

invokeai/app/invocations/sd3_image_to_latents.py

@@ -2,7 +2,7 @@ import einops
 import torch
 from diffusers.models.autoencoders.autoencoder_kl import AutoencoderKL
 
-from invokeai.app.invocations.baseinvocation import BaseInvocation, Classification, invocation
+from invokeai.app.invocations.baseinvocation import BaseInvocation, invocation
 from invokeai.app.invocations.fields import (
     FieldDescriptions,
     ImageField,
@@ -25,7 +25,6 @@ from invokeai.backend.util.devices import TorchDevice
     tags=["image", "latents", "vae", "i2l", "sd3"],
     category="image",
     version="1.0.1",
-    classification=Classification.Prototype,
 )
 class SD3ImageToLatentsInvocation(BaseInvocation, WithMetadata, WithBoard):
     """Generates latents from an image."""

invokeai/app/invocations/sd3_model_loader.py

@@ -3,7 +3,6 @@ from typing import Optional
 from invokeai.app.invocations.baseinvocation import (
     BaseInvocation,
     BaseInvocationOutput,
-    Classification,
     invocation,
     invocation_output,
 )
@@ -14,7 +13,7 @@ from invokeai.app.util.t5_model_identifier import (
     preprocess_t5_encoder_model_identifier,
     preprocess_t5_tokenizer_model_identifier,
 )
-from invokeai.backend.model_manager.config import SubModelType
+from invokeai.backend.model_manager.taxonomy import SubModelType
 
 
 @invocation_output("sd3_model_loader_output")
@@ -34,7 +33,6 @@ class Sd3ModelLoaderOutput(BaseInvocationOutput):
     tags=["model", "sd3"],
     category="model",
     version="1.0.1",
-    classification=Classification.Prototype,
 )
 class Sd3ModelLoaderInvocation(BaseInvocation):
     """Loads a SD3 base model, outputting its submodels."""

invokeai/app/invocations/sd3_text_encoder.py

@@ -11,12 +11,12 @@ from transformers import (
     T5TokenizerFast,
 )
 
-from invokeai.app.invocations.baseinvocation import BaseInvocation, Classification, invocation
+from invokeai.app.invocations.baseinvocation import BaseInvocation, invocation
 from invokeai.app.invocations.fields import FieldDescriptions, Input, InputField
 from invokeai.app.invocations.model import CLIPField, T5EncoderField
 from invokeai.app.invocations.primitives import SD3ConditioningOutput
 from invokeai.app.services.shared.invocation_context import InvocationContext
-from invokeai.backend.model_manager.config import ModelFormat
+from invokeai.backend.model_manager.taxonomy import ModelFormat
 from invokeai.backend.patches.layer_patcher import LayerPatcher
 from invokeai.backend.patches.lora_conversions.flux_lora_constants import FLUX_LORA_CLIP_PREFIX
 from invokeai.backend.patches.model_patch_raw import ModelPatchRaw
@@ -33,7 +33,6 @@ SD3_T5_MAX_SEQ_LEN = 256
     tags=["prompt", "conditioning", "sd3"],
     category="conditioning",
     version="1.0.1",
-    classification=Classification.Prototype,
 )
 class Sd3TextEncoderInvocation(BaseInvocation):
     """Encodes and preps a prompt for a SD3 image."""

invokeai/app/invocations/sdxl.py

@@ -2,7 +2,7 @@ from invokeai.app.invocations.baseinvocation import BaseInvocation, BaseInvocati
 from invokeai.app.invocations.fields import FieldDescriptions, InputField, OutputField, UIType
 from invokeai.app.invocations.model import CLIPField, ModelIdentifierField, UNetField, VAEField
 from invokeai.app.services.shared.invocation_context import InvocationContext
-from invokeai.backend.model_manager import SubModelType
+from invokeai.backend.model_manager.taxonomy import SubModelType
 
 
 @invocation_output("sdxl_model_loader_output")

invokeai/app/invocations/segment_anything.py

@@ -6,7 +6,7 @@ import numpy as np
 import torch
 from PIL import Image
 from pydantic import BaseModel, Field
-from transformers import AutoModelForMaskGeneration, AutoProcessor
+from transformers import AutoProcessor
 from transformers.models.sam import SamModel
 from transformers.models.sam.processing_sam import SamProcessor
 
@@ -104,14 +104,13 @@ class SegmentAnythingInvocation(BaseInvocation):
 
     @staticmethod
     def _load_sam_model(model_path: Path):
-        sam_model = AutoModelForMaskGeneration.from_pretrained(
+        sam_model = SamModel.from_pretrained(
             model_path,
             local_files_only=True,
             # TODO(ryand): Setting the torch_dtype here doesn't work. Investigate whether fp16 is supported by the
             # model, and figure out how to make it work in the pipeline.
             # torch_dtype=TorchDevice.choose_torch_dtype(),
         )
-        assert isinstance(sam_model, SamModel)
 
         sam_processor = AutoProcessor.from_pretrained(model_path, local_files_only=True)
         assert isinstance(sam_processor, SamProcessor)

invokeai/app/invocations/tiled_multi_diffusion_denoise_latents.py

@@ -7,9 +7,9 @@ from diffusers.models.unets.unet_2d_condition import UNet2DConditionModel
 from diffusers.schedulers.scheduling_utils import SchedulerMixin
 from pydantic import field_validator
 
-from invokeai.app.invocations.baseinvocation import BaseInvocation, Classification, invocation
+from invokeai.app.invocations.baseinvocation import BaseInvocation, invocation
 from invokeai.app.invocations.constants import LATENT_SCALE_FACTOR
-from invokeai.app.invocations.controlnet_image_processors import ControlField
+from invokeai.app.invocations.controlnet import ControlField
 from invokeai.app.invocations.denoise_latents import DenoiseLatentsInvocation, get_scheduler
 from invokeai.app.invocations.fields import (
     ConditioningField,
@@ -56,7 +56,6 @@ def crop_controlnet_data(control_data: ControlNetData, latent_region: TBLR) -> C
     title="Tiled Multi-Diffusion Denoise - SD1.5, SDXL",
     tags=["upscale", "denoise"],
     category="latents",
-    classification=Classification.Beta,
     version="1.0.1",
 )
 class TiledMultiDiffusionDenoiseLatents(BaseInvocation):

invokeai/app/invocations/tiles.py

@@ -7,7 +7,6 @@ from pydantic import BaseModel
 from invokeai.app.invocations.baseinvocation import (
     BaseInvocation,
     BaseInvocationOutput,
-    Classification,
     invocation,
     invocation_output,
 )
@@ -40,7 +39,6 @@ class CalculateImageTilesOutput(BaseInvocationOutput):
     tags=["tiles"],
     category="tiles",
     version="1.0.1",
-    classification=Classification.Beta,
 )
 class CalculateImageTilesInvocation(BaseInvocation):
     """Calculate the coordinates and overlaps of tiles that cover a target image shape."""
@@ -74,7 +72,6 @@ class CalculateImageTilesInvocation(BaseInvocation):
     tags=["tiles"],
     category="tiles",
     version="1.1.1",
-    classification=Classification.Beta,
 )
 class CalculateImageTilesEvenSplitInvocation(BaseInvocation):
     """Calculate the coordinates and overlaps of tiles that cover a target image shape."""
@@ -117,7 +114,6 @@ class CalculateImageTilesEvenSplitInvocation(BaseInvocation):
     tags=["tiles"],
     category="tiles",
     version="1.0.1",
-    classification=Classification.Beta,
 )
 class CalculateImageTilesMinimumOverlapInvocation(BaseInvocation):
     """Calculate the coordinates and overlaps of tiles that cover a target image shape."""
@@ -168,7 +164,6 @@ class TileToPropertiesOutput(BaseInvocationOutput):
     tags=["tiles"],
     category="tiles",
     version="1.0.1",
-    classification=Classification.Beta,
 )
 class TileToPropertiesInvocation(BaseInvocation):
     """Split a Tile into its individual properties."""
@@ -201,7 +196,6 @@ class PairTileImageOutput(BaseInvocationOutput):
     tags=["tiles"],
     category="tiles",
     version="1.0.1",
-    classification=Classification.Beta,
 )
 class PairTileImageInvocation(BaseInvocation):
     """Pair an image with its tile properties."""
@@ -230,7 +224,6 @@ BLEND_MODES = Literal["Linear", "Seam"]
     tags=["tiles"],
     category="tiles",
     version="1.1.1",
-    classification=Classification.Beta,
 )
 class MergeTilesToImageInvocation(BaseInvocation, WithMetadata, WithBoard):
     """Merge multiple tile images into a single image."""

invokeai/app/run_app.py

@@ -1,12 +1,3 @@
-import uvicorn
-
-from invokeai.app.invocations.load_custom_nodes import load_custom_nodes
-from invokeai.app.services.config.config_default import get_config
-from invokeai.app.util.torch_cuda_allocator import configure_torch_cuda_allocator
-from invokeai.backend.util.logging import InvokeAILogger
-from invokeai.frontend.cli.arg_parser import InvokeAIArgs
-
-
 def get_app():
     """Import the app and event loop. We wrap this in a function to more explicitly control when it happens, because
     importing from api_app does a bunch of stuff - it's more like calling a function than importing a module.
@@ -18,9 +9,18 @@ def get_app():
 
 def run_app() -> None:
     """The main entrypoint for the app."""
-    # Parse the CLI arguments.
+    from invokeai.frontend.cli.arg_parser import InvokeAIArgs
+
+    # Parse the CLI arguments before doing anything else, which ensures CLI args correctly override settings from other
+    # sources like `invokeai.yaml` or env vars.
     InvokeAIArgs.parse_args()
 
+    import uvicorn
+
+    from invokeai.app.services.config.config_default import get_config
+    from invokeai.app.util.torch_cuda_allocator import configure_torch_cuda_allocator
+    from invokeai.backend.util.logging import InvokeAILogger
+
     # Load config.
     app_config = get_config()
 
@@ -31,6 +31,14 @@ def run_app() -> None:
     if app_config.pytorch_cuda_alloc_conf:
         configure_torch_cuda_allocator(app_config.pytorch_cuda_alloc_conf, logger)
 
+    # This import must happen after configure_torch_cuda_allocator() is called, because the module imports torch.
+    from invokeai.app.invocations.baseinvocation import InvocationRegistry
+    from invokeai.app.invocations.load_custom_nodes import load_custom_nodes
+    from invokeai.backend.util.devices import TorchDevice
+
+    torch_device_name = TorchDevice.get_torch_device_name()
+    logger.info(f"Using torch device: {torch_device_name}")
+
     # Import from startup_utils here to avoid importing torch before configure_torch_cuda_allocator() is called.
     from invokeai.app.util.startup_utils import (
         apply_monkeypatches,
@@ -41,16 +49,15 @@ def run_app() -> None:
     )
 
     # Find an open port, and modify the config accordingly.
-    orig_config_port = app_config.port
-    app_config.port = find_open_port(app_config.port)
-    if orig_config_port != app_config.port:
+    first_open_port = find_open_port(app_config.port)
+    if app_config.port != first_open_port:
+        orig_config_port = app_config.port
+        app_config.port = first_open_port
         logger.warning(f"Port {orig_config_port} is already in use. Using port {app_config.port}.")
 
     # Miscellaneous startup tasks.
     apply_monkeypatches()
     register_mime_types()
-    if app_config.dev_reload:
-        enable_dev_reload()
     check_cudnn(logger)
 
     # Initialize the app and event loop.
@@ -61,6 +68,20 @@ def run_app() -> None:
     # core nodes have been imported so that we can catch when a custom node clobbers a core node.
     load_custom_nodes(custom_nodes_path=app_config.custom_nodes_path, logger=logger)
 
+    # Check all invocations and ensure their outputs are registered.
+    for invocation in InvocationRegistry.get_invocation_classes():
+        invocation_type = invocation.get_type()
+        output_annotation = invocation.get_output_annotation()
+        if output_annotation not in InvocationRegistry.get_output_classes():
+            logger.warning(
+                f'Invocation "{invocation_type}" has unregistered output class "{output_annotation.__name__}"'
+            )
+
+    if app_config.dev_reload:
+        # load_custom_nodes seems to bypass jurrigged's import sniffer, so be sure to call it *after* they're already
+        # imported.
+        enable_dev_reload(custom_nodes_path=app_config.custom_nodes_path)
+
     # Start the server.
     config = uvicorn.Config(
         app=app,

invokeai/app/services/board_image_records/board_image_records_sqlite.py

@@ -98,9 +98,18 @@ class SqliteBoardImageRecordStorage(BoardImageRecordStorageBase):
                 FROM images
                 LEFT JOIN board_images ON board_images.image_name = images.image_name
                 WHERE 1=1
+                """
+
+        # Handle board_id filter
+        if board_id == "none":
+            stmt += """--sql
+                AND board_images.board_id IS NULL
+                """
+        else:
+            stmt += """--sql
                 AND board_images.board_id = ?
                 """
-        params.append(board_id)
+            params.append(board_id)
 
         # Add the category filter
         if categories is not None:

invokeai/app/services/config/config_default.py

@@ -24,7 +24,6 @@ from invokeai.frontend.cli.arg_parser import InvokeAIArgs
 INIT_FILE = Path("invokeai.yaml")
 DB_FILE = Path("invokeai.db")
 LEGACY_INIT_FILE = Path("invokeai.init")
-DEVICE = Literal["auto", "cpu", "cuda", "cuda:1", "mps"]
 PRECISION = Literal["auto", "float16", "bfloat16", "float32"]
 ATTENTION_TYPE = Literal["auto", "normal", "xformers", "sliced", "torch-sdp"]
 ATTENTION_SLICE_SIZE = Literal["auto", "balanced", "max", 1, 2, 3, 4, 5, 6, 7, 8]
@@ -93,7 +92,7 @@ class InvokeAIAppConfig(BaseSettings):
         vram: DEPRECATED: This setting is no longer used. It has been replaced by `max_cache_vram_gb`, but most users will not need to use this config since automatic cache size limits should work well in most cases. This config setting will be removed once the new model cache behavior is stable.
         lazy_offload: DEPRECATED: This setting is no longer used. Lazy-offloading is enabled by default. This config setting will be removed once the new model cache behavior is stable.
         pytorch_cuda_alloc_conf: Configure the Torch CUDA memory allocator. This will impact peak reserved VRAM usage and performance. Setting to "backend:cudaMallocAsync" works well on many systems. The optimal configuration is highly dependent on the system configuration (device type, VRAM, CUDA driver version, etc.), so must be tuned experimentally.
-        device: Preferred execution device. `auto` will choose the device depending on the hardware platform and the installed torch capabilities.<br>Valid values: `auto`, `cpu`, `cuda`, `cuda:1`, `mps`
+        device: Preferred execution device. `auto` will choose the device depending on the hardware platform and the installed torch capabilities.<br>Valid values: `auto`, `cpu`, `cuda`, `mps`, `cuda:N` (where N is a device number)
         precision: Floating point precision. `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.<br>Valid values: `auto`, `float16`, `bfloat16`, `float32`
         sequential_guidance: Whether to calculate guidance in serial instead of in parallel, lowering memory requirements.
         attention_type: Attention type.<br>Valid values: `auto`, `normal`, `xformers`, `sliced`, `torch-sdp`
@@ -176,7 +175,7 @@ class InvokeAIAppConfig(BaseSettings):
     pytorch_cuda_alloc_conf: Optional[str] = Field(default=None,            description="Configure the Torch CUDA memory allocator. This will impact peak reserved VRAM usage and performance. Setting to \"backend:cudaMallocAsync\" works well on many systems. The optimal configuration is highly dependent on the system configuration (device type, VRAM, CUDA driver version, etc.), so must be tuned experimentally.")
 
     # DEVICE
-    device:                      DEVICE = Field(default="auto",             description="Preferred execution device. `auto` will choose the device depending on the hardware platform and the installed torch capabilities.")
+    device:                      str = Field(default="auto",                description="Preferred execution device. `auto` will choose the device depending on the hardware platform and the installed torch capabilities.<br>Valid values: `auto`, `cpu`, `cuda`, `mps`, `cuda:N` (where N is a device number)", pattern=r"^(auto|cpu|mps|cuda(:\d+)?)$")
     precision:                PRECISION = Field(default="auto",             description="Floating point precision. `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.")
 
     # GENERATION

invokeai/app/services/events/events_base.py

@@ -44,7 +44,8 @@ if TYPE_CHECKING:
         SessionQueueItem,
         SessionQueueStatus,
     )
-    from invokeai.backend.model_manager.config import AnyModelConfig, SubModelType
+    from invokeai.backend.model_manager import SubModelType
+    from invokeai.backend.model_manager.config import AnyModelConfig
 
 
 class EventServiceBase:

invokeai/app/services/events/events_common.py

@@ -16,7 +16,8 @@ from invokeai.app.services.session_queue.session_queue_common import (
 )
 from invokeai.app.services.shared.graph import AnyInvocation, AnyInvocationOutput
 from invokeai.app.util.misc import get_timestamp
-from invokeai.backend.model_manager.config import AnyModelConfig, SubModelType
+from invokeai.backend.model_manager import SubModelType
+from invokeai.backend.model_manager.config import AnyModelConfig
 
 if TYPE_CHECKING:
     from invokeai.app.services.download.download_base import DownloadJob
@@ -240,6 +241,7 @@ class QueueItemStatusChangedEvent(QueueItemEventBase):
     batch_status: BatchStatus = Field(description="The status of the batch")
     queue_status: SessionQueueStatus = Field(description="The status of the queue")
     session_id: str = Field(description="The ID of the session (aka graph execution state)")
+    credits: Optional[float] = Field(default=None, description="The total credits used for this queue item")
 
     @classmethod
     def build(
@@ -262,6 +264,7 @@ class QueueItemStatusChangedEvent(QueueItemEventBase):
             completed_at=str(queue_item.completed_at) if queue_item.completed_at else None,
             batch_status=batch_status,
             queue_status=queue_status,
+            credits=queue_item.credits,
         )
 
 

invokeai/app/services/image_records/image_records_sqlite.py

@@ -196,9 +196,13 @@ class SqliteImageRecordStorage(ImageRecordStorageBase):
         # Search term condition
         if search_term:
             query_conditions += """--sql
-            AND images.metadata LIKE ?
+            AND (
+                images.metadata LIKE ?
+                OR images.created_at LIKE ?
+            )
             """
             query_params.append(f"%{search_term.lower()}%")
+            query_params.append(f"%{search_term.lower()}%")
 
         if starred_first:
             query_pagination = f"""--sql

invokeai/app/services/images/images_default.py

@@ -78,7 +78,7 @@ class ImageService(ImageServiceABC):
                         board_id=board_id, image_name=image_name
                     )
                 except Exception as e:
-                    self.__invoker.services.logger.warn(f"Failed to add image to board {board_id}: {str(e)}")
+                    self.__invoker.services.logger.warning(f"Failed to add image to board {board_id}: {str(e)}")
             self.__invoker.services.image_files.save(
                 image_name=image_name, image=image, metadata=metadata, workflow=workflow, graph=graph
             )

invokeai/app/services/invocation_services.py

@@ -27,6 +27,10 @@ if TYPE_CHECKING:
     from invokeai.app.services.invocation_stats.invocation_stats_base import InvocationStatsServiceBase
     from invokeai.app.services.model_images.model_images_base import ModelImageFileStorageBase
     from invokeai.app.services.model_manager.model_manager_base import ModelManagerServiceBase
+    from invokeai.app.services.model_relationship_records.model_relationship_records_base import (
+        ModelRelationshipRecordStorageBase,
+    )
+    from invokeai.app.services.model_relationships.model_relationships_base import ModelRelationshipsServiceABC
     from invokeai.app.services.names.names_base import NameServiceBase
     from invokeai.app.services.session_processor.session_processor_base import SessionProcessorBase
     from invokeai.app.services.session_queue.session_queue_base import SessionQueueBase
@@ -54,6 +58,8 @@ class InvocationServices:
         logger: "Logger",
         model_images: "ModelImageFileStorageBase",
         model_manager: "ModelManagerServiceBase",
+        model_relationships: "ModelRelationshipsServiceABC",
+        model_relationship_records: "ModelRelationshipRecordStorageBase",
         download_queue: "DownloadQueueServiceBase",
         performance_statistics: "InvocationStatsServiceBase",
         session_queue: "SessionQueueBase",
@@ -81,6 +87,8 @@ class InvocationServices:
         self.logger = logger
         self.model_images = model_images
         self.model_manager = model_manager
+        self.model_relationships = model_relationships
+        self.model_relationship_records = model_relationship_records
         self.download_queue = download_queue
         self.performance_statistics = performance_statistics
         self.session_queue = session_queue

invokeai/app/services/invocation_stats/invocation_stats_base.py

@@ -60,7 +60,7 @@ class InvocationStatsServiceBase(ABC):
         pass
 
     @abstractmethod
-    def reset_stats(self):
+    def reset_stats(self, graph_execution_state_id: str) -> None:
         """Reset all stored statistics."""
         pass
 

invokeai/app/services/invocation_stats/invocation_stats_default.py

@@ -73,9 +73,9 @@ class InvocationStatsService(InvocationStatsServiceBase):
             )
             self._stats[graph_execution_state_id].add_node_execution_stats(node_stats)
 
-    def reset_stats(self):
-        self._stats = {}
-        self._cache_stats = {}
+    def reset_stats(self, graph_execution_state_id: str) -> None:
+        self._stats.pop(graph_execution_state_id, None)
+        self._cache_stats.pop(graph_execution_state_id, None)
 
     def get_stats(self, graph_execution_state_id: str) -> InvocationStatsSummary:
         graph_stats_summary = self._get_graph_summary(graph_execution_state_id)

invokeai/app/services/model_install/model_install_common.py

@@ -10,9 +10,9 @@ from typing_extensions import Annotated
 
 from invokeai.app.services.download import DownloadJob, MultiFileDownloadJob
 from invokeai.app.services.model_records import ModelRecordChanges
-from invokeai.backend.model_manager import AnyModelConfig, ModelRepoVariant
-from invokeai.backend.model_manager.config import ModelSourceType
+from invokeai.backend.model_manager.config import AnyModelConfig
 from invokeai.backend.model_manager.metadata import AnyModelRepoMetadata
+from invokeai.backend.model_manager.taxonomy import ModelRepoVariant, ModelSourceType
 
 
 class InstallStatus(str, Enum):

invokeai/app/services/model_install/model_install_default.py

@@ -38,9 +38,9 @@ from invokeai.backend.model_manager.config import (
     AnyModelConfig,
     CheckpointConfigBase,
     InvalidModelConfigException,
-    ModelRepoVariant,
-    ModelSourceType,
+    ModelConfigBase,
 )
+from invokeai.backend.model_manager.legacy_probe import ModelProbe
 from invokeai.backend.model_manager.metadata import (
     AnyModelRepoMetadata,
     HuggingFaceMetadataFetch,
@@ -49,8 +49,8 @@ from invokeai.backend.model_manager.metadata import (
     RemoteModelFile,
 )
 from invokeai.backend.model_manager.metadata.metadata_base import HuggingFaceMetadata
-from invokeai.backend.model_manager.probe import ModelProbe
 from invokeai.backend.model_manager.search import ModelSearch
+from invokeai.backend.model_manager.taxonomy import ModelRepoVariant, ModelSourceType
 from invokeai.backend.util import InvokeAILogger
 from invokeai.backend.util.catch_sigint import catch_sigint
 from invokeai.backend.util.devices import TorchDevice
@@ -148,7 +148,7 @@ class ModelInstallService(ModelInstallServiceBase):
     def _clear_pending_jobs(self) -> None:
         for job in self.list_jobs():
             if not job.in_terminal_state:
-                self._logger.warning("Cancelling job {job.id}")
+                self._logger.warning(f"Cancelling job {job.id}")
                 self.cancel_job(job)
         while True:
             try:
@@ -182,9 +182,7 @@ class ModelInstallService(ModelInstallServiceBase):
     ) -> str:  # noqa D102
         model_path = Path(model_path)
         config = config or ModelRecordChanges()
-        info: AnyModelConfig = ModelProbe.probe(
-            Path(model_path), config.model_dump(), hash_algo=self._app_config.hashing_algorithm
-        )  # type: ignore
+        info: AnyModelConfig = self._probe(Path(model_path), config)  # type: ignore
 
         if preferred_name := config.name:
             preferred_name = Path(preferred_name).with_suffix(model_path.suffix)
@@ -644,12 +642,30 @@ class ModelInstallService(ModelInstallServiceBase):
         move(old_path, new_path)
         return new_path
 
+    def _probe(self, model_path: Path, config: Optional[ModelRecordChanges] = None):
+        config = config or ModelRecordChanges()
+        hash_algo = self._app_config.hashing_algorithm
+        fields = config.model_dump()
+
+        # WARNING!
+        # The legacy probe relies on the implicit order of tests to determine model classification.
+        # This can lead to regressions between the legacy and new probes.
+        # Do NOT change the order of `probe` and `classify` without implementing one of the following fixes:
+        # Short-term fix: `classify` tests `matches` in the same order as the legacy probe.
+        # Long-term fix: Improve `matches` to be more specific so that only one config matches
+        #   any given model - eliminating ambiguity and removing reliance on order.
+        # After implementing either of these fixes, remove @pytest.mark.xfail from `test_regression_against_model_probe`
+        try:
+            return ModelProbe.probe(model_path=model_path, fields=fields, hash_algo=hash_algo)  # type: ignore
+        except InvalidModelConfigException:
+            return ModelConfigBase.classify(model_path, hash_algo, **fields)
+
     def _register(
         self, model_path: Path, config: Optional[ModelRecordChanges] = None, info: Optional[AnyModelConfig] = None
     ) -> str:
         config = config or ModelRecordChanges()
 
-        info = info or ModelProbe.probe(model_path, config.model_dump(), hash_algo=self._app_config.hashing_algorithm)  # type: ignore
+        info = info or self._probe(model_path, config)
 
         model_path = model_path.resolve()
 

invokeai/app/services/model_load/model_load_base.py

@@ -5,9 +5,10 @@ from abc import ABC, abstractmethod
 from pathlib import Path
 from typing import Callable, Optional
 
-from invokeai.backend.model_manager import AnyModel, AnyModelConfig, SubModelType
+from invokeai.backend.model_manager.config import AnyModelConfig
 from invokeai.backend.model_manager.load import LoadedModel, LoadedModelWithoutConfig
 from invokeai.backend.model_manager.load.model_cache.model_cache import ModelCache
+from invokeai.backend.model_manager.taxonomy import AnyModel, SubModelType
 
 
 class ModelLoadServiceBase(ABC):

invokeai/app/services/model_load/model_load_default.py

@@ -11,7 +11,7 @@ from torch import load as torch_load
 from invokeai.app.services.config import InvokeAIAppConfig
 from invokeai.app.services.invoker import Invoker
 from invokeai.app.services.model_load.model_load_base import ModelLoadServiceBase
-from invokeai.backend.model_manager import AnyModel, AnyModelConfig, SubModelType
+from invokeai.backend.model_manager.config import AnyModelConfig
 from invokeai.backend.model_manager.load import (
     LoadedModel,
     LoadedModelWithoutConfig,
@@ -20,6 +20,7 @@ from invokeai.backend.model_manager.load import (
 )
 from invokeai.backend.model_manager.load.model_cache.model_cache import ModelCache
 from invokeai.backend.model_manager.load.model_loaders.generic_diffusers import GenericDiffusersLoader
+from invokeai.backend.model_manager.taxonomy import AnyModel, SubModelType
 from invokeai.backend.util.devices import TorchDevice
 from invokeai.backend.util.logging import InvokeAILogger
 
@@ -85,8 +86,11 @@ class ModelLoadService(ModelLoadServiceBase):
 
         def torch_load_file(checkpoint: Path) -> AnyModel:
             scan_result = scan_file_path(checkpoint)
-            if scan_result.infected_files != 0 or scan_result.scan_err:
-                raise Exception("The model at {checkpoint} is potentially infected by malware. Aborting load.")
+            if scan_result.infected_files != 0:
+                raise Exception(f"The model at {checkpoint} is potentially infected by malware. Aborting load.")
+            if scan_result.scan_err:
+                raise Exception(f"Error scanning model at {checkpoint} for malware. Aborting load.")
+
             result = torch_load(checkpoint, map_location="cpu")
             return result
 

invokeai/app/services/model_manager/__init__.py

@@ -1,16 +1,12 @@
 """Initialization file for model manager service."""
 
 from invokeai.app.services.model_manager.model_manager_default import ModelManagerService, ModelManagerServiceBase
-from invokeai.backend.model_manager import AnyModel, AnyModelConfig, BaseModelType, ModelType, SubModelType
+from invokeai.backend.model_manager import AnyModelConfig
 from invokeai.backend.model_manager.load import LoadedModel
 
 __all__ = [
     "ModelManagerServiceBase",
     "ModelManagerService",
-    "AnyModel",
     "AnyModelConfig",
-    "BaseModelType",
-    "ModelType",
-    "SubModelType",
     "LoadedModel",
 ]

invokeai/app/services/model_records/model_records_base.py

@@ -14,10 +14,12 @@ from invokeai.app.services.shared.pagination import PaginatedResults
 from invokeai.app.util.model_exclude_null import BaseModelExcludeNull
 from invokeai.backend.model_manager.config import (
     AnyModelConfig,
-    BaseModelType,
-    ClipVariantType,
     ControlAdapterDefaultSettings,
     MainModelDefaultSettings,
+)
+from invokeai.backend.model_manager.taxonomy import (
+    BaseModelType,
+    ClipVariantType,
     ModelFormat,
     ModelSourceType,
     ModelType,
@@ -78,6 +80,7 @@ class ModelRecordChanges(BaseModelExcludeNull):
     type: Optional[ModelType] = Field(description="Type of model", default=None)
     key: Optional[str] = Field(description="Database ID for this model", default=None)
     hash: Optional[str] = Field(description="hash of model file", default=None)
+    file_size: Optional[int] = Field(description="Size of model file", default=None)
     format: Optional[str] = Field(description="format of model file", default=None)
     trigger_phrases: Optional[set[str]] = Field(description="Set of trigger phrases for this model", default=None)
     default_settings: Optional[MainModelDefaultSettings | ControlAdapterDefaultSettings] = Field(