fix(ui): do not bake opacity when rasterizing layer adjustments

- Move migration of model-specific ui_types into BaseInvocation. This
gives us access to the node and field names, so the warnings are more
useful to the end user.
- Ensure we serialize the fields' json_schema_extra with enum values.
This wasn't a problem until now, when it interferes with migrating
ui_type cleanly. It's a transparent change.
- Improve warnings when validating fields (which includes the ui_type
migration logic)

.github/ISSUE_TEMPLATE/BUG_REPORT.yml

@@ -21,6 +21,20 @@ body:
         - label: I have searched the existing issues
           required: true
 
+  - type: dropdown
+    id: install_method
+    attributes:
+      label: Install method
+      description: How did you install Invoke?
+      multiple: false
+      options:
+        - "Invoke's Launcher"
+        - 'Stability Matrix'
+        - 'Pinokio'
+        - 'Manual'
+    validations:
+      required: true
+
   - type: markdown
     attributes:
       value: __Describe your environment__
@@ -76,8 +90,8 @@ body:
     attributes:
       label: Version number
       description: |
-        The version of Invoke you have installed. If it is not the latest version, please update and try again to confirm the issue still exists. If you are testing main, please include the commit hash instead.
-      placeholder: ex. 3.6.1
+        The version of Invoke you have installed. If it is not the [latest version](https://github.com/invoke-ai/InvokeAI/releases/latest), please update and try again to confirm the issue still exists. If you are testing main, please include the commit hash instead.
+      placeholder: ex. v6.0.2
     validations:
       required: true
 
@@ -85,17 +99,17 @@ body:
     id: browser-version
     attributes:
       label: Browser
-      description: Your web browser and version.
+      description: Your web browser and version, if you do not use the Launcher's provided GUI.
       placeholder: ex. Firefox 123.0b3
     validations:
-      required: true
+      required: false
 
   - type: textarea
     id: python-deps
     attributes:
-      label: Python dependencies
+      label: System Information
       description: |
-        If the problem occurred during image generation, click the gear icon at the bottom left corner, click "About", click the copy button and then paste here.
+        Click the gear icon at the bottom left corner, then click "About". Click the copy button and then paste here.
     validations:
       required: false
 

.github/pull_request_template.md

@@ -18,5 +18,6 @@
 
 - [ ] _The PR has a short but descriptive title, suitable for a changelog_
 - [ ] _Tests added / updated (if applicable)_
+- [ ] _❗Changes to a redux slice have a corresponding migration_
 - [ ] _Documentation added / updated (if applicable)_
 - [ ] _Updated `What's New` copy (if doing a release after this PR)_

.github/workflows/build-container.yml

@@ -45,6 +45,9 @@ jobs:
     steps:
       - name: Free up more disk space on the runner
         # https://github.com/actions/runner-images/issues/2840#issuecomment-1284059930
+        # the /mnt dir has 70GBs of free space
+        # /dev/sda1        74G   28K   70G   1% /mnt
+        # According to some online posts the /mnt is not always there, so checking before setting docker to use it
         run: |
           echo "----- Free space before cleanup"
           df -h
@@ -52,6 +55,11 @@ jobs:
           sudo rm -rf "$AGENT_TOOLSDIRECTORY"
           sudo swapoff /mnt/swapfile
           sudo rm -rf /mnt/swapfile
+          if [ -d /mnt ]; then
+            sudo chmod -R 777 /mnt
+            echo '{"data-root": "/mnt/docker-root"}' | sudo tee /etc/docker/daemon.json
+            sudo systemctl restart docker
+          fi
           echo "----- Free space after cleanup"
           df -h
 

.github/workflows/lfs-checks.yml

@@ -0,0 +1,30 @@
+# Checks that large files and LFS-tracked files are properly checked in with pointer format.
+# Uses https://github.com/ppremk/lfs-warning to detect LFS issues.
+
+name: 'lfs checks'
+
+on:
+  push:
+    branches:
+      - 'main'
+  pull_request:
+    types:
+      - 'ready_for_review'
+      - 'opened'
+      - 'synchronize'
+  merge_group:
+  workflow_dispatch:
+
+jobs:
+  lfs-check:
+    runs-on: ubuntu-latest
+    timeout-minutes: 5
+    permissions:
+      # Required to label and comment on the PRs
+      pull-requests: write
+    steps:
+      - name: checkout
+        uses: actions/checkout@v4
+
+      - name: check lfs files
+        uses: ppremk/lfs-warning@v3.3

.github/workflows/typegen-checks.yml

@@ -39,6 +39,18 @@ jobs:
       - name: checkout
         uses: actions/checkout@v4
 
+      - name: Free up more disk space on the runner
+        # https://github.com/actions/runner-images/issues/2840#issuecomment-1284059930
+        run: |
+          echo "----- Free space before cleanup"
+          df -h
+          sudo rm -rf /usr/share/dotnet
+          sudo rm -rf "$AGENT_TOOLSDIRECTORY"
+          sudo swapoff /mnt/swapfile
+          sudo rm -rf /mnt/swapfile
+          echo "----- Free space after cleanup"
+          df -h
+
       - name: check for changed files
         if: ${{ inputs.always_run != true }}
         id: changed-files

.gitignore

@@ -190,3 +190,5 @@ installer/update.bat
 installer/update.sh
 installer/InvokeAI-Installer/
 .aider*
+
+.claude/

docker/.env.sample

@@ -22,6 +22,10 @@
 ## GPU_DRIVER can be set to either `cuda` or `rocm` to enable GPU support in the container accordingly.
 # GPU_DRIVER=cuda #| rocm
 
+## If you are using ROCM, you will need to ensure that the render group within the container and the host system use the same group ID.
+## To obtain the group ID of the render group on the host system, run `getent group render` and grab the number.
+# RENDER_GROUP_ID=
+
 ## CONTAINER_UID can be set to the UID of the user on the host system that should own the files in the container.
 ## It is usually not necessary to change this. Use `id -u` on the host system to find the UID.
 # CONTAINER_UID=1000

docker/Dockerfile

@@ -5,8 +5,7 @@
 FROM docker.io/node:22-slim AS web-builder
 ENV PNPM_HOME="/pnpm"
 ENV PATH="$PNPM_HOME:$PATH"
-RUN corepack use pnpm@8.x
-RUN corepack enable
+RUN corepack use pnpm@10.x && corepack enable
 
 WORKDIR /build
 COPY invokeai/frontend/web/ ./
@@ -44,7 +43,6 @@ ENV \
     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 \
@@ -75,19 +73,17 @@ RUN --mount=type=cache,target=/root/.cache/uv \
     --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"
+    ulimit -n 30000 && \
+    uv sync --extra $GPU_DRIVER --frozen
 
 # 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"
+    ln -s "/usr/share/libdrm/amdgpu.ids" "/opt/amdgpu/share/libdrm/amdgpu.ids" && groupadd render
+
+# 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"
 
 RUN mkdir -p ${INVOKEAI_ROOT} && chown -R ${CONTAINER_UID}:${CONTAINER_GID} ${INVOKEAI_ROOT}
 
@@ -106,8 +102,6 @@ COPY invokeai ${INVOKEAI_SRC}/invokeai
 RUN --mount=type=cache,target=/root/.cache/uv \
     --mount=type=bind,source=pyproject.toml,target=pyproject.toml \
     --mount=type=bind,source=uv.lock,target=uv.lock \
-    if [ "$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 .
+    ulimit -n 30000 && \
+    uv pip install -e .[$GPU_DRIVER]
 

docker/Dockerfile-rocm-full

@@ -0,0 +1,136 @@
+# syntax=docker/dockerfile:1.4
+
+#### Web UI ------------------------------------
+
+FROM docker.io/node:22-slim AS web-builder
+ENV PNPM_HOME="/pnpm"
+ENV PATH="$PNPM_HOME:$PATH"
+RUN corepack use pnpm@8.x
+RUN corepack enable
+
+WORKDIR /build
+COPY invokeai/frontend/web/ ./
+RUN --mount=type=cache,target=/pnpm/store \
+    pnpm install --frozen-lockfile
+RUN npx vite build
+
+## Backend ---------------------------------------
+
+FROM library/ubuntu:24.04
+
+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 \
+    --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 \
+    wget
+
+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 \
+    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}
+
+ARG GPU_DRIVER=cuda
+
+# Install `uv` for package management
+COPY --from=ghcr.io/astral-sh/uv:0.6.9 /uv /uvx /bin/
+
+# 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 \
+    ulimit -n 30000 && \
+    uv sync --extra $GPU_DRIVER --frozen
+
+RUN --mount=type=cache,target=/var/cache/apt \
+    --mount=type=cache,target=/var/lib/apt \
+    if [ "$GPU_DRIVER" = "rocm" ]; then \
+    wget -O /tmp/amdgpu-install.deb \
+    https://repo.radeon.com/amdgpu-install/6.3.4/ubuntu/noble/amdgpu-install_6.3.60304-1_all.deb && \
+    apt install -y /tmp/amdgpu-install.deb && \
+    apt update && \
+    amdgpu-install --usecase=rocm -y && \
+    apt-get autoclean && \
+    apt clean && \
+    rm -rf /tmp/* /var/tmp/* && \
+    usermod -a -G render ubuntu && \
+    usermod -a -G video ubuntu && \
+    echo "\\n/opt/rocm/lib\\n/opt/rocm/lib64" >> /etc/ld.so.conf.d/rocm.conf && \
+    ldconfig && \
+    update-alternatives --auto rocm; \
+    fi
+
+## Heathen711: Leaving this for review input, will remove before merge
+# RUN --mount=type=cache,target=/var/cache/apt \
+#     --mount=type=cache,target=/var/lib/apt \
+#     if [ "$GPU_DRIVER" = "rocm" ]; then \
+#     groupadd render && \
+#     usermod -a -G render ubuntu && \
+#     usermod -a -G video ubuntu; \
+#     fi
+
+## Link amdgpu.ids for ROCm builds
+## contributed by https://github.com/Rubonnek
+# RUN mkdir -p "/opt/amdgpu/share/libdrm" &&\
+#     ln -s "/usr/share/libdrm/amdgpu.ids" "/opt/amdgpu/share/libdrm/amdgpu.ids"
+
+# build patchmatch
+RUN cd /usr/lib/$(uname -p)-linux-gnu/pkgconfig/ && ln -sf opencv4.pc opencv.pc
+RUN python -c "from patchmatch import patch_match"
+
+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 ${INVOKEAI_SRC}/invokeai
+
+# this should not increase image size because we've already installed dependencies
+# in a previous layer
+RUN --mount=type=cache,target=/root/.cache/uv \
+    --mount=type=bind,source=pyproject.toml,target=pyproject.toml \
+    --mount=type=bind,source=uv.lock,target=uv.lock \
+    ulimit -n 30000 && \
+    uv pip install -e .[$GPU_DRIVER]
+

docker/docker-compose.yml

@@ -47,8 +47,9 @@ services:
 
   invokeai-rocm:
     <<: *invokeai
-    devices:
-      - /dev/kfd:/dev/kfd
-      - /dev/dri:/dev/dri
+    environment:
+      - AMD_VISIBLE_DEVICES=all
+      - RENDER_GROUP_ID=${RENDER_GROUP_ID}
+    runtime: amd
     profiles:
       - rocm

docker/docker-entrypoint.sh

@@ -21,6 +21,17 @@ _=$(id ${USER} 2>&1) || useradd -u ${USER_ID} ${USER}
 # ensure the UID is correct
 usermod -u ${USER_ID} ${USER} 1>/dev/null
 
+## ROCM specific configuration
+# render group within the container must match the host render group
+# otherwise the container will not be able to access the host GPU.
+if [[ -v "RENDER_GROUP_ID" ]] && [[ ! -z "${RENDER_GROUP_ID}" ]]; then
+  # ensure the render group exists
+  groupmod -g ${RENDER_GROUP_ID} render
+  usermod -a -G render ${USER}
+  usermod -a -G video ${USER}
+fi
+
+
 ### Set the $PUBLIC_KEY env var to enable SSH access.
 # We do not install openssh-server in the image by default to avoid bloat.
 # but it is useful to have the full SSH server e.g. on Runpod.

docker/run.sh

@@ -13,7 +13,7 @@ run() {
 
   # parse .env file for build args
   build_args=$(awk '$1 ~ /=[^$]/ && $0 !~ /^#/ {print "--build-arg " $0 " "}' .env) &&
-  profile="$(awk -F '=' '/GPU_DRIVER/ {print $2}' .env)"
+  profile="$(awk -F '=' '/GPU_DRIVER=/ {print $2}' .env)"
 
   # default to 'cuda' profile
   [[ -z "$profile" ]] && profile="cuda"
@@ -30,7 +30,7 @@ run() {
 
   printf "%s\n" "starting service $service_name"
   docker compose --profile "$profile" up -d "$service_name"
-  docker compose logs -f
+  docker compose --profile "$profile" logs -f
 }
 
 run

docs/contributing/MODEL_MANAGER.md

@@ -265,7 +265,7 @@ If the key is unrecognized, this call raises an
 
 #### exists(key) -> AnyModelConfig
 
-Returns True if a model with the given key exists in the databsae.
+Returns True if a model with the given key exists in the database.
 
 #### search_by_path(path) -> AnyModelConfig
 
@@ -718,7 +718,7 @@ When downloading remote models is implemented, additional
 configuration information, such as list of trigger terms, will be
 retrieved from the HuggingFace and Civitai model repositories.
 
-The probed values can be overriden by providing a dictionary in the
+The probed values can be overridden by providing a dictionary in the
 optional `config` argument passed to `import_model()`. You may provide
 overriding values for any of the model's configuration
 attributes. Here is an example of setting the
@@ -841,7 +841,7 @@ variable.
 
 #### installer.start(invoker)
 
-The `start` method is called by the API intialization routines when
+The `start` method is called by the API initialization routines when
 the API starts up. Its effect is to call `sync_to_config()` to
 synchronize the model record store database with what's currently on
 disk.

docs/contributing/contributors.md

@@ -16,7 +16,7 @@ We thank [all contributors](https://github.com/invoke-ai/InvokeAI/graphs/contrib
 - @psychedelicious (Spencer Mabrito) - Web Team Leader
 - @joshistoast (Josh Corbett) - Web Development
 - @cheerio (Mary Rogers) - Lead Engineer & Web App Development
-- @ebr (Eugene Brodsky) - Cloud/DevOps/Sofware engineer; your friendly neighbourhood cluster-autoscaler
+- @ebr (Eugene Brodsky) - Cloud/DevOps/Software engineer; your friendly neighbourhood cluster-autoscaler
 - @sunija - Standalone version
 - @brandon (Brandon Rising) - Platform, Infrastructure, Backend Systems
 - @ryanjdick (Ryan Dick) - Machine Learning & Training

docs/contributing/dev-environment.md

@@ -41,7 +41,7 @@ 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 3.12 --python-preference only-managed --index=https://download.pytorch.org/whl/cu126 --reinstall
+      uv pip install -e ".[dev,test,docs,xformers]" --python 3.12 --python-preference only-managed --index=https://download.pytorch.org/whl/cu128 --reinstall
       ```
 
 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.
@@ -50,11 +50,11 @@ If you just want to use Invoke, you should use the [launcher][launcher link].
 
       If you only want to edit the docs, you can stop here and skip to the **Documentation** section below.
 
-7. Install the frontend dev toolchain:
+7. Install the frontend dev toolchain, paying attention to versions:
 
-      - [`nodejs`](https://nodejs.org/) (v20+)
+      - [`nodejs`](https://nodejs.org/) (tested on LTS, v22)
 
-      - [`pnpm`](https://pnpm.io/8.x/installation) (must be v8 - not v9!)
+      - [`pnpm`](https://pnpm.io/installation) (tested on v10)
 
 8. Do a production build of the frontend:
 

docs/contributing/frontend/workflows.md

@@ -297,7 +297,7 @@ Migration logic is in [migrations.ts].
 <!-- links -->
 
 [pydantic]: https://github.com/pydantic/pydantic 'pydantic'
-[zod]: https://github.com/colinhacks/zod 'zod/v4'
+[zod]: https://github.com/colinhacks/zod 'zod'
 [openapi-types]: https://github.com/kogosoftwarellc/open-api/tree/main/packages/openapi-types 'openapi-types'
 [reactflow]: https://github.com/xyflow/xyflow 'reactflow'
 [reactflow-concepts]: https://reactflow.dev/learn/concepts/terms-and-definitions

docs/installation/manual.md

@@ -69,34 +69,34 @@ The following commands vary depending on the version of Invoke being installed a
     - 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.
+7. Determine the torch backend to use for installation, if any. This is necessary to get the right version of torch installed. This is acheived by using [UV's built in torch support.](https://docs.astral.sh/uv/guides/integration/pytorch/#automatic-backend-selection)
 
     === "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.**
+        - If you are on Windows or Linux with an Nvidia GPU, use `--torch-backend=cu128`.
+        - If you are on Linux with no GPU, use `--torch-backend=cpu`.
+        - If you are on Linux with an AMD GPU, use `--torch-backend=rocm6.3`.
+        - **In all other cases, do not use a torch backend.**
 
     === "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`.
+        - If you are on Windows or Linux with an Nvidia GPU, use `--torch-backend=cu126`.
+        - If you are on Linux with no GPU, use `--torch-backend=cpu`.
+        - If you are on Linux with an AMD GPU, use `--torch-backend=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`.
-        - If you are on Linux with an AMD GPU, use `https://download.pytorch.org/whl/rocm6.1`.
+        - If you are on Windows with an Nvidia GPU, use `--torch-backend=cu124`.
+        - If you are on Linux with no GPU, use `--torch-backend=cpu`.
+        - If you are on Linux with an AMD GPU, use `--torch-backend=rocm6.1`.
         - **In all other cases, do not use an index.**
 
     === "Invoke v4"
 
-        - 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`.
-        - If you are on Linux with an AMD GPU, use `https://download.pytorch.org/whl/rocm5.2`.
+        - If you are on Windows with an Nvidia GPU, use `--torch-backend=cu124`.
+        - If you are on Linux with no GPU, use `--torch-backend=cpu`.
+        - If you are on Linux with an AMD GPU, use `--torch-backend=rocm5.2`.
         - **In all other cases, do not use an index.**
 
 8. Install the `invokeai` package. Substitute the package specifier and version.
@@ -105,10 +105,10 @@ The following commands vary depending on the version of Invoke being installed a
     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:
+    If you determined you needed to use a torch backend in the previous step, you'll need to set the backend like this:
 
     ```sh
-    uv pip install <PACKAGE_SPECIFIER>==<VERSION> --python 3.12 --python-preference only-managed --index=<INDEX_URL> --force-reinstall
+    uv pip install <PACKAGE_SPECIFIER>==<VERSION> --python 3.12 --python-preference only-managed --torch-backend=<VERSION> --force-reinstall
     ```
 
 9. Deactivate and reactivate your venv so that the invokeai-specific commands become available in the environment:

docs/installation/quick_start.md

@@ -33,30 +33,45 @@ Hardware requirements vary significantly depending on model and image output siz
 
 More detail on system requirements can be found [here](./requirements.md).
 
-## Step 2: Download
+## Step 2: Download and Set Up the Launcher
 
-Download the most recent launcher for your operating system:
+The Launcher manages your Invoke install. Follow these instructions to download and set up the Launcher.
 
-- [Download for Windows](https://download.invoke.ai/Invoke%20Community%20Edition.exe)
-- [Download for macOS](https://download.invoke.ai/Invoke%20Community%20Edition.dmg)
-- [Download for Linux](https://download.invoke.ai/Invoke%20Community%20Edition.AppImage)
+!!! info "Instructions for each OS"
 
-## Step 3: Install or Update
+    === "Windows"
 
-Run the launcher you just downloaded, click **Install** and follow the instructions to get set up.
+        - [Download for Windows](https://github.com/invoke-ai/launcher/releases/latest/download/Invoke.Community.Edition.Setup.latest.exe)
+        - Run the `EXE` to install the Launcher and start it.
+        - A desktop shortcut will be created; use this to run the Launcher in the future.
+        - You can delete the `EXE` file you downloaded.
+
+    === "macOS"
+
+        - [Download for macOS](https://github.com/invoke-ai/launcher/releases/latest/download/Invoke.Community.Edition-latest-arm64.dmg)
+        - Open the `DMG` and drag the app into `Applications`.
+        - Run the Launcher using its entry in `Applications`.
+        - You can delete the `DMG` file you downloaded.
+
+    === "Linux"
+
+        - [Download for Linux](https://github.com/invoke-ai/launcher/releases/latest/download/Invoke.Community.Edition-latest.AppImage)
+        - You may need to edit the `AppImage` file properties and make it executable.
+        - Optionally move the file to a location that does not require admin privileges and add a desktop shortcut for it.
+        - Run the Launcher by double-clicking the `AppImage` or the shortcut you made.
+
+## Step 3: Install Invoke
+
+Run the Launcher you just set up if you haven't already. Click **Install** and follow the instructions to install (or update) Invoke.
 
 If you have an existing Invoke installation, you can select it and let the launcher manage the install. You'll be able to update or launch the installation.
 
-!!! warning "Problem running the launcher on macOS"
+!!! tip "Updating"
 
-    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:
+    The Launcher will check for updates for itself _and_ Invoke.
 
-    - 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`.
-
-    You should now be able to run the launcher.
+    - When the Launcher detects an update is available for itself, you'll get a small popup window. Click through this and the Launcher will update itself.
+    - When the Launcher detects an update for Invoke, you'll see a small green alert in the Launcher. Click that and follow the instructions to update Invoke.
 
 ## Step 4: Launch
 

docs/nodes/NODES.md

@@ -41,7 +41,7 @@ Nodes have a "Use Cache" option in their footer. This allows for performance imp
 
 There are several node grouping concepts that can be examined with a narrow focus. These (and other) groupings can be pieced together to make up functional graph setups, and are important to understanding how groups of nodes work together as part of a whole. Note that the screenshots below aren't examples of complete functioning node graphs (see Examples).
 
-### Noise
+### Create Latent Noise
 
 An initial noise tensor is necessary for the latent diffusion process. As a result, the Denoising node requires a noise node input.  
 

invokeai/app/api/dependencies.py

@@ -10,6 +10,7 @@ from invokeai.app.services.board_images.board_images_default import BoardImagesS
 from invokeai.app.services.board_records.board_records_sqlite import SqliteBoardRecordStorage
 from invokeai.app.services.boards.boards_default import BoardService
 from invokeai.app.services.bulk_download.bulk_download_default import BulkDownloadService
+from invokeai.app.services.client_state_persistence.client_state_persistence_sqlite import ClientStatePersistenceSqlite
 from invokeai.app.services.config.config_default import InvokeAIAppConfig
 from invokeai.app.services.download.download_default import DownloadQueueService
 from invokeai.app.services.events.events_fastapievents import FastAPIEventService
@@ -151,6 +152,7 @@ class ApiDependencies:
         style_preset_records = SqliteStylePresetRecordsStorage(db=db)
         style_preset_image_files = StylePresetImageFileStorageDisk(style_presets_folder / "images")
         workflow_thumbnails = WorkflowThumbnailFileStorageDisk(workflow_thumbnails_folder)
+        client_state_persistence = ClientStatePersistenceSqlite(db=db)
 
         services = InvocationServices(
             board_image_records=board_image_records,
@@ -181,6 +183,7 @@ class ApiDependencies:
             style_preset_records=style_preset_records,
             style_preset_image_files=style_preset_image_files,
             workflow_thumbnails=workflow_thumbnails,
+            client_state_persistence=client_state_persistence,
         )
 
         ApiDependencies.invoker = Invoker(services)

invokeai/app/api/routers/board_videos.py

@@ -0,0 +1,39 @@
+from fastapi import Body, HTTPException
+from fastapi.routing import APIRouter
+
+from invokeai.app.services.videos_common import AddVideosToBoardResult, RemoveVideosFromBoardResult
+
+board_videos_router = APIRouter(prefix="/v1/board_videos", tags=["boards"])
+
+
+@board_videos_router.post(
+    "/batch",
+    operation_id="add_videos_to_board",
+    responses={
+        201: {"description": "Videos were added to board successfully"},
+    },
+    status_code=201,
+    response_model=AddVideosToBoardResult,
+)
+async def add_videos_to_board(
+    board_id: str = Body(description="The id of the board to add to"),
+    video_ids: list[str] = Body(description="The ids of the videos to add", embed=True),
+) -> AddVideosToBoardResult:
+    """Adds a list of videos to a board"""
+    raise HTTPException(status_code=501, detail="Not implemented")
+
+
+@board_videos_router.post(
+    "/batch/delete",
+    operation_id="remove_videos_from_board",
+    responses={
+        201: {"description": "Videos were removed from board successfully"},
+    },
+    status_code=201,
+    response_model=RemoveVideosFromBoardResult,
+)
+async def remove_videos_from_board(
+    video_ids: list[str] = Body(description="The ids of the videos to remove", embed=True),
+) -> RemoveVideosFromBoardResult:
+    """Removes a list of videos from their board, if they had one"""
+    raise HTTPException(status_code=501, detail="Not implemented")

invokeai/app/api/routers/client_state.py

@@ -0,0 +1,58 @@
+from fastapi import Body, HTTPException, Path, Query
+from fastapi.routing import APIRouter
+
+from invokeai.app.api.dependencies import ApiDependencies
+from invokeai.backend.util.logging import logging
+
+client_state_router = APIRouter(prefix="/v1/client_state", tags=["client_state"])
+
+
+@client_state_router.get(
+    "/{queue_id}/get_by_key",
+    operation_id="get_client_state_by_key",
+    response_model=str | None,
+)
+async def get_client_state_by_key(
+    queue_id: str = Path(description="The queue id to perform this operation on"),
+    key: str = Query(..., description="Key to get"),
+) -> str | None:
+    """Gets the client state"""
+    try:
+        return ApiDependencies.invoker.services.client_state_persistence.get_by_key(queue_id, key)
+    except Exception as e:
+        logging.error(f"Error getting client state: {e}")
+        raise HTTPException(status_code=500, detail="Error setting client state")
+
+
+@client_state_router.post(
+    "/{queue_id}/set_by_key",
+    operation_id="set_client_state",
+    response_model=str,
+)
+async def set_client_state(
+    queue_id: str = Path(description="The queue id to perform this operation on"),
+    key: str = Query(..., description="Key to set"),
+    value: str = Body(..., description="Stringified value to set"),
+) -> str:
+    """Sets the client state"""
+    try:
+        return ApiDependencies.invoker.services.client_state_persistence.set_by_key(queue_id, key, value)
+    except Exception as e:
+        logging.error(f"Error setting client state: {e}")
+        raise HTTPException(status_code=500, detail="Error setting client state")
+
+
+@client_state_router.post(
+    "/{queue_id}/delete",
+    operation_id="delete_client_state",
+    responses={204: {"description": "Client state deleted"}},
+)
+async def delete_client_state(
+    queue_id: str = Path(description="The queue id to perform this operation on"),
+) -> None:
+    """Deletes the client state"""
+    try:
+        ApiDependencies.invoker.services.client_state_persistence.delete(queue_id)
+    except Exception as e:
+        logging.error(f"Error deleting client state: {e}")
+        raise HTTPException(status_code=500, detail="Error deleting client state")

invokeai/app/api/routers/session_queue.py

@@ -7,7 +7,6 @@ from pydantic import BaseModel, Field
 from invokeai.app.api.dependencies import ApiDependencies
 from invokeai.app.services.session_processor.session_processor_common import SessionProcessorStatus
 from invokeai.app.services.session_queue.session_queue_common import (
-    QUEUE_ITEM_STATUS,
     Batch,
     BatchStatus,
     CancelAllExceptCurrentResult,
@@ -18,6 +17,7 @@ from invokeai.app.services.session_queue.session_queue_common import (
     DeleteByDestinationResult,
     EnqueueBatchResult,
     FieldIdentifier,
+    ItemIdsResult,
     PruneResult,
     RetryItemsResult,
     SessionQueueCountsByDestination,
@@ -25,7 +25,7 @@ from invokeai.app.services.session_queue.session_queue_common import (
     SessionQueueItemNotFoundError,
     SessionQueueStatus,
 )
-from invokeai.app.services.shared.pagination import CursorPaginatedResults
+from invokeai.app.services.shared.sqlite.sqlite_common import SQLiteDirection
 
 session_queue_router = APIRouter(prefix="/v1/queue", tags=["queue"])
 
@@ -68,36 +68,6 @@ async def enqueue_batch(
         raise HTTPException(status_code=500, detail=f"Unexpected error while enqueuing batch: {e}")
 
 
-@session_queue_router.get(
-    "/{queue_id}/list",
-    operation_id="list_queue_items",
-    responses={
-        200: {"model": CursorPaginatedResults[SessionQueueItem]},
-    },
-)
-async def list_queue_items(
-    queue_id: str = Path(description="The queue id to perform this operation on"),
-    limit: int = Query(default=50, description="The number of items to fetch"),
-    status: Optional[QUEUE_ITEM_STATUS] = Query(default=None, description="The status of items to fetch"),
-    cursor: Optional[int] = Query(default=None, description="The pagination cursor"),
-    priority: int = Query(default=0, description="The pagination cursor priority"),
-    destination: Optional[str] = Query(default=None, description="The destination of queue items to fetch"),
-) -> CursorPaginatedResults[SessionQueueItem]:
-    """Gets cursor-paginated queue items"""
-
-    try:
-        return ApiDependencies.invoker.services.session_queue.list_queue_items(
-            queue_id=queue_id,
-            limit=limit,
-            status=status,
-            cursor=cursor,
-            priority=priority,
-            destination=destination,
-        )
-    except Exception as e:
-        raise HTTPException(status_code=500, detail=f"Unexpected error while listing all items: {e}")
-
-
 @session_queue_router.get(
     "/{queue_id}/list_all",
     operation_id="list_all_queue_items",
@@ -119,6 +89,56 @@ async def list_all_queue_items(
         raise HTTPException(status_code=500, detail=f"Unexpected error while listing all queue items: {e}")
 
 
+@session_queue_router.get(
+    "/{queue_id}/item_ids",
+    operation_id="get_queue_item_ids",
+    responses={
+        200: {"model": ItemIdsResult},
+    },
+)
+async def get_queue_item_ids(
+    queue_id: str = Path(description="The queue id to perform this operation on"),
+    order_dir: SQLiteDirection = Query(default=SQLiteDirection.Descending, description="The order of sort"),
+) -> ItemIdsResult:
+    """Gets all queue item ids that match the given parameters"""
+    try:
+        return ApiDependencies.invoker.services.session_queue.get_queue_item_ids(queue_id=queue_id, order_dir=order_dir)
+    except Exception as e:
+        raise HTTPException(status_code=500, detail=f"Unexpected error while listing all queue item ids: {e}")
+
+
+@session_queue_router.post(
+    "/{queue_id}/items_by_ids",
+    operation_id="get_queue_items_by_item_ids",
+    responses={200: {"model": list[SessionQueueItem]}},
+)
+async def get_queue_items_by_item_ids(
+    queue_id: str = Path(description="The queue id to perform this operation on"),
+    item_ids: list[int] = Body(
+        embed=True, description="Object containing list of queue item ids to fetch queue items for"
+    ),
+) -> list[SessionQueueItem]:
+    """Gets queue items for the specified queue item ids. Maintains order of item ids."""
+    try:
+        session_queue_service = ApiDependencies.invoker.services.session_queue
+
+        # Fetch queue items preserving the order of requested item ids
+        queue_items: list[SessionQueueItem] = []
+        for item_id in item_ids:
+            try:
+                queue_item = session_queue_service.get_queue_item(item_id=item_id)
+                if queue_item.queue_id != queue_id:  # Auth protection for items from other queues
+                    continue
+                queue_items.append(queue_item)
+            except Exception:
+                # Skip missing queue items - they may have been deleted between item id fetch and queue item fetch
+                continue
+
+        return queue_items
+    except Exception:
+        raise HTTPException(status_code=500, detail="Failed to get queue items")
+
+
 @session_queue_router.put(
     "/{queue_id}/processor/resume",
     operation_id="resume",
@@ -354,7 +374,10 @@ async def get_queue_item(
 ) -> SessionQueueItem:
     """Gets a queue item"""
     try:
-        return ApiDependencies.invoker.services.session_queue.get_queue_item(item_id)
+        queue_item = ApiDependencies.invoker.services.session_queue.get_queue_item(item_id=item_id)
+        if queue_item.queue_id != queue_id:
+            raise HTTPException(status_code=404, detail=f"Queue item with id {item_id} not found in queue {queue_id}")
+        return queue_item
     except SessionQueueItemNotFoundError:
         raise HTTPException(status_code=404, detail=f"Queue item with id {item_id} not found in queue {queue_id}")
     except Exception as e:

invokeai/app/api/routers/videos.py

@@ -0,0 +1,119 @@
+from typing import Optional
+
+from fastapi import Body, HTTPException, Path, Query
+from fastapi.routing import APIRouter
+
+from invokeai.app.services.shared.pagination import OffsetPaginatedResults
+from invokeai.app.services.shared.sqlite.sqlite_common import SQLiteDirection
+from invokeai.app.services.videos_common import (
+    DeleteVideosResult,
+    StarredVideosResult,
+    UnstarredVideosResult,
+    VideoDTO,
+    VideoIdsResult,
+    VideoRecordChanges,
+)
+
+videos_router = APIRouter(prefix="/v1/videos", tags=["videos"])
+
+
+@videos_router.patch(
+    "/i/{video_id}",
+    operation_id="update_video",
+    response_model=VideoDTO,
+)
+async def update_video(
+    video_id: str = Path(description="The id of the video to update"),
+    video_changes: VideoRecordChanges = Body(description="The changes to apply to the video"),
+) -> VideoDTO:
+    """Updates a video"""
+
+    raise HTTPException(status_code=501, detail="Not implemented")
+
+
+@videos_router.get(
+    "/i/{video_id}",
+    operation_id="get_video_dto",
+    response_model=VideoDTO,
+)
+async def get_video_dto(
+    video_id: str = Path(description="The id of the video to get"),
+) -> VideoDTO:
+    """Gets a video's DTO"""
+
+    raise HTTPException(status_code=501, detail="Not implemented")
+
+
+@videos_router.post("/delete", operation_id="delete_videos_from_list", response_model=DeleteVideosResult)
+async def delete_videos_from_list(
+    video_ids: list[str] = Body(description="The list of ids of videos to delete", embed=True),
+) -> DeleteVideosResult:
+    raise HTTPException(status_code=501, detail="Not implemented")
+
+
+@videos_router.post("/star", operation_id="star_videos_in_list", response_model=StarredVideosResult)
+async def star_videos_in_list(
+    video_ids: list[str] = Body(description="The list of ids of videos to star", embed=True),
+) -> StarredVideosResult:
+    raise HTTPException(status_code=501, detail="Not implemented")
+
+
+@videos_router.post("/unstar", operation_id="unstar_videos_in_list", response_model=UnstarredVideosResult)
+async def unstar_videos_in_list(
+    video_ids: list[str] = Body(description="The list of ids of videos to unstar", embed=True),
+) -> UnstarredVideosResult:
+    raise HTTPException(status_code=501, detail="Not implemented")
+
+
+@videos_router.delete("/uncategorized", operation_id="delete_uncategorized_videos", response_model=DeleteVideosResult)
+async def delete_uncategorized_videos() -> DeleteVideosResult:
+    """Deletes all videos that are uncategorized"""
+
+    raise HTTPException(status_code=501, detail="Not implemented")
+
+
+@videos_router.get("/", operation_id="list_video_dtos", response_model=OffsetPaginatedResults[VideoDTO])
+async def list_video_dtos(
+    is_intermediate: Optional[bool] = Query(default=None, description="Whether to list intermediate videos."),
+    board_id: Optional[str] = Query(
+        default=None,
+        description="The board id to filter by. Use 'none' to find videos without a board.",
+    ),
+    offset: int = Query(default=0, description="The page offset"),
+    limit: int = Query(default=10, description="The number of videos per page"),
+    order_dir: SQLiteDirection = Query(default=SQLiteDirection.Descending, description="The order of sort"),
+    starred_first: bool = Query(default=True, description="Whether to sort by starred videos first"),
+    search_term: Optional[str] = Query(default=None, description="The term to search for"),
+) -> OffsetPaginatedResults[VideoDTO]:
+    """Lists video DTOs"""
+
+    raise HTTPException(status_code=501, detail="Not implemented")
+
+
+@videos_router.get("/ids", operation_id="get_video_ids")
+async def get_video_ids(
+    is_intermediate: Optional[bool] = Query(default=None, description="Whether to list intermediate videos."),
+    board_id: Optional[str] = Query(
+        default=None,
+        description="The board id to filter by. Use 'none' to find videos without a board.",
+    ),
+    order_dir: SQLiteDirection = Query(default=SQLiteDirection.Descending, description="The order of sort"),
+    starred_first: bool = Query(default=True, description="Whether to sort by starred videos first"),
+    search_term: Optional[str] = Query(default=None, description="The term to search for"),
+) -> VideoIdsResult:
+    """Gets ordered list of video ids with metadata for optimistic updates"""
+
+    raise HTTPException(status_code=501, detail="Not implemented")
+
+
+@videos_router.post(
+    "/videos_by_ids",
+    operation_id="get_videos_by_ids",
+    responses={200: {"model": list[VideoDTO]}},
+)
+async def get_videos_by_ids(
+    video_ids: list[str] = Body(embed=True, description="Object containing list of video ids to fetch DTOs for"),
+) -> list[VideoDTO]:
+    """Gets video DTOs for the specified video ids. Maintains order of input ids."""
+
+    raise HTTPException(status_code=501, detail="Not implemented")

invokeai/app/api_app.py

@@ -18,7 +18,9 @@ from invokeai.app.api.no_cache_staticfiles import NoCacheStaticFiles
 from invokeai.app.api.routers import (
     app_info,
     board_images,
+    board_videos,
     boards,
+    client_state,
     download_queue,
     images,
     model_manager,
@@ -26,6 +28,7 @@ from invokeai.app.api.routers import (
     session_queue,
     style_presets,
     utilities,
+    videos,
     workflows,
 )
 from invokeai.app.api.sockets import SocketIO
@@ -124,13 +127,16 @@ app.include_router(utilities.utilities_router, prefix="/api")
 app.include_router(model_manager.model_manager_router, prefix="/api")
 app.include_router(download_queue.download_queue_router, prefix="/api")
 app.include_router(images.images_router, prefix="/api")
+app.include_router(videos.videos_router, prefix="/api")
 app.include_router(boards.boards_router, prefix="/api")
 app.include_router(board_images.board_images_router, prefix="/api")
+app.include_router(board_videos.board_videos_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")
 app.include_router(style_presets.style_presets_router, prefix="/api")
+app.include_router(client_state.client_state_router, prefix="/api")
 
 app.openapi = get_openapi_func(app)
 
@@ -155,6 +161,12 @@ def overridden_redoc() -> HTMLResponse:
 
 web_root_path = Path(list(web_dir.__path__)[0])
 
+if app_config.unsafe_disable_picklescan:
+    logger.warning(
+        "The unsafe_disable_picklescan option is enabled. This disables malware scanning while installing and"
+        "loading models, which may allow malicious code to be executed. Use at your own risk."
+    )
+
 try:
     app.mount("/", NoCacheStaticFiles(directory=Path(web_root_path, "dist"), html=True), name="ui")
 except RuntimeError:

invokeai/app/invocations/baseinvocation.py

@@ -36,6 +36,9 @@ from pydantic_core import PydanticUndefined
 from invokeai.app.invocations.fields import (
     FieldKind,
     Input,
+    InputFieldJSONSchemaExtra,
+    UIType,
+    migrate_model_ui_type,
 )
 from invokeai.app.services.config.config_default import get_config
 from invokeai.app.services.shared.invocation_context import InvocationContext
@@ -256,7 +259,9 @@ class BaseInvocation(ABC, BaseModel):
     is_intermediate: bool = Field(
         default=False,
         description="Whether or not this is an intermediate invocation.",
-        json_schema_extra={"ui_type": "IsIntermediate", "field_kind": FieldKind.NodeAttribute},
+        json_schema_extra=InputFieldJSONSchemaExtra(
+            input=Input.Direct, field_kind=FieldKind.NodeAttribute, ui_type=UIType._IsIntermediate
+        ).model_dump(exclude_none=True),
     )
     use_cache: bool = Field(
         default=True,
@@ -445,6 +450,15 @@ with warnings.catch_warnings():
     RESERVED_PYDANTIC_FIELD_NAMES = {m[0] for m in inspect.getmembers(_Model())}
 
 
+def is_enum_member(value: Any, enum_class: type[Enum]) -> bool:
+    """Checks if a value is a member of an enum class."""
+    try:
+        enum_class(value)
+        return True
+    except ValueError:
+        return False
+
+
 def validate_fields(model_fields: dict[str, FieldInfo], model_type: str) -> None:
     """
     Validates the fields of an invocation or invocation output:
@@ -456,51 +470,99 @@ def validate_fields(model_fields: dict[str, FieldInfo], model_type: str) -> None
     """
     for name, field in model_fields.items():
         if name in RESERVED_PYDANTIC_FIELD_NAMES:
-            raise InvalidFieldError(f'Invalid field name "{name}" on "{model_type}" (reserved by pydantic)')
+            raise InvalidFieldError(f"{model_type}.{name}: Invalid field name (reserved by pydantic)")
 
         if not field.annotation:
-            raise InvalidFieldError(f'Invalid field type "{name}" on "{model_type}" (missing annotation)')
+            raise InvalidFieldError(f"{model_type}.{name}: Invalid field type (missing annotation)")
 
         if not isinstance(field.json_schema_extra, dict):
-            raise InvalidFieldError(
-                f'Invalid field definition for "{name}" on "{model_type}" (missing json_schema_extra dict)'
-            )
+            raise InvalidFieldError(f"{model_type}.{name}: Invalid field definition (missing json_schema_extra dict)")
 
         field_kind = field.json_schema_extra.get("field_kind", None)
 
         # must have a field_kind
-        if not isinstance(field_kind, FieldKind):
+        if not is_enum_member(field_kind, FieldKind):
             raise InvalidFieldError(
-                f'Invalid field definition for "{name}" on "{model_type}" (maybe it\'s not an InputField or OutputField?)'
+                f"{model_type}.{name}: Invalid field definition for (maybe it's not an InputField or OutputField?)"
             )
 
-        if field_kind is FieldKind.Input and (
+        if field_kind == FieldKind.Input.value and (
             name in RESERVED_NODE_ATTRIBUTE_FIELD_NAMES or name in RESERVED_INPUT_FIELD_NAMES
         ):
-            raise InvalidFieldError(f'Invalid field name "{name}" on "{model_type}" (reserved input field name)')
+            raise InvalidFieldError(f"{model_type}.{name}: Invalid field name (reserved input field name)")
 
-        if field_kind is FieldKind.Output and name in RESERVED_OUTPUT_FIELD_NAMES:
-            raise InvalidFieldError(f'Invalid field name "{name}" on "{model_type}" (reserved output field name)')
+        if field_kind == FieldKind.Output.value and name in RESERVED_OUTPUT_FIELD_NAMES:
+            raise InvalidFieldError(f"{model_type}.{name}: Invalid field name (reserved output field name)")
 
-        if (field_kind is FieldKind.Internal) and name not in RESERVED_INPUT_FIELD_NAMES:
-            raise InvalidFieldError(
-                f'Invalid field name "{name}" on "{model_type}" (internal field without reserved name)'
-            )
+        if field_kind == FieldKind.Internal.value and name not in RESERVED_INPUT_FIELD_NAMES:
+            raise InvalidFieldError(f"{model_type}.{name}: Invalid field name (internal field without reserved name)")
 
         # node attribute fields *must* be in the reserved list
         if (
-            field_kind is FieldKind.NodeAttribute
+            field_kind == FieldKind.NodeAttribute.value
             and name not in RESERVED_NODE_ATTRIBUTE_FIELD_NAMES
             and name not in RESERVED_OUTPUT_FIELD_NAMES
         ):
             raise InvalidFieldError(
-                f'Invalid field name "{name}" on "{model_type}" (node attribute field without reserved name)'
+                f"{model_type}.{name}: Invalid field name (node attribute field without reserved name)"
             )
 
         ui_type = field.json_schema_extra.get("ui_type", None)
-        if isinstance(ui_type, str) and ui_type.startswith("DEPRECATED_"):
-            logger.warning(f'"UIType.{ui_type.split("_")[-1]}" is deprecated, ignoring')
-            field.json_schema_extra.pop("ui_type")
+        ui_model_base = field.json_schema_extra.get("ui_model_base", None)
+        ui_model_type = field.json_schema_extra.get("ui_model_type", None)
+        ui_model_variant = field.json_schema_extra.get("ui_model_variant", None)
+        ui_model_format = field.json_schema_extra.get("ui_model_format", None)
+
+        if ui_type is not None:
+            # There are 3 cases where we may need to take action:
+            #
+            # 1. The ui_type is a migratable, deprecated value. For example, ui_type=UIType.MainModel value is
+            #    deprecated and should be migrated to:
+            #       - ui_model_base=[BaseModelType.StableDiffusion1, BaseModelType.StableDiffusion2]
+            #       - ui_model_type=[ModelType.Main]
+            #
+            # 2. ui_type was set in conjunction with any of the new ui_model_[base|type|variant|format] fields, which
+            #    is not allowed (they are mutually exclusive). In this case, we ignore ui_type and log a warning.
+            #
+            # 3. ui_type is a deprecated value that is not migratable. For example, ui_type=UIType.Image is deprecated;
+            #    Image fields are now automatically detected based on the field's type annotation. In this case, we
+            #    ignore ui_type and log a warning.
+            #
+            # The cases must be checked in this order to ensure proper handling.
+
+            # Easier to work with as an enum
+            ui_type = UIType(ui_type)
+
+            # The enum member values are not always the same as their names - we want to log the name so the user can
+            # easily review their code and see where the deprecated enum member is used.
+            human_readable_name = f"UIType.{ui_type.name}"
+
+            # Case 1: migratable deprecated value
+            did_migrate = migrate_model_ui_type(ui_type, field.json_schema_extra)
+
+            if did_migrate:
+                logger.warning(
+                    f'{model_type}.{name}: Migrated deprecated "ui_type" "{human_readable_name}" to new ui_model_[base|type|variant|format] fields'
+                )
+                field.json_schema_extra.pop("ui_type")
+
+            # Case 2: mutually exclusive with new fields
+            elif (
+                ui_model_base is not None
+                or ui_model_type is not None
+                or ui_model_variant is not None
+                or ui_model_format is not None
+            ):
+                logger.warning(
+                    f'{model_type}.{name}: "ui_type" is mutually exclusive with "ui_model_[base|type|format|variant]", ignoring "ui_type"'
+                )
+                field.json_schema_extra.pop("ui_type")
+
+            # Case 3: deprecated value that is not migratable
+            elif ui_type.startswith("DEPRECATED_"):
+                logger.warning(f'{model_type}.{name}: Deprecated "ui_type" "{human_readable_name}", ignoring')
+                field.json_schema_extra.pop("ui_type")
+
     return None
 
 

invokeai/app/invocations/cogview4_image_to_latents.py

@@ -17,6 +17,7 @@ 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
+from invokeai.backend.util.vae_working_memory import estimate_vae_working_memory_cogview4
 
 # TODO(ryand): This is effectively a copy of SD3ImageToLatentsInvocation and a subset of ImageToLatentsInvocation. We
 # should refactor to avoid this duplication.
@@ -38,7 +39,11 @@ class CogView4ImageToLatentsInvocation(BaseInvocation, WithMetadata, WithBoard):
 
     @staticmethod
     def vae_encode(vae_info: LoadedModel, image_tensor: torch.Tensor) -> torch.Tensor:
-        with vae_info as vae:
+        assert isinstance(vae_info.model, AutoencoderKL)
+        estimated_working_memory = estimate_vae_working_memory_cogview4(
+            operation="encode", image_tensor=image_tensor, vae=vae_info.model
+        )
+        with vae_info.model_on_device(working_mem_bytes=estimated_working_memory) as (_, vae):
             assert isinstance(vae, AutoencoderKL)
 
             vae.disable_tiling()
@@ -62,6 +67,8 @@ class CogView4ImageToLatentsInvocation(BaseInvocation, WithMetadata, WithBoard):
             image_tensor = einops.rearrange(image_tensor, "c h w -> 1 c h w")
 
         vae_info = context.models.load(self.vae.vae)
+        assert isinstance(vae_info.model, AutoencoderKL)
+
         latents = self.vae_encode(vae_info=vae_info, image_tensor=image_tensor)
 
         latents = latents.to("cpu")

invokeai/app/invocations/cogview4_latents_to_image.py

@@ -6,7 +6,6 @@ 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,
@@ -20,6 +19,7 @@ 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
+from invokeai.backend.util.vae_working_memory import estimate_vae_working_memory_cogview4
 
 # TODO(ryand): This is effectively a copy of SD3LatentsToImageInvocation and a subset of LatentsToImageInvocation. We
 # should refactor to avoid this duplication.
@@ -39,22 +39,15 @@ class CogView4LatentsToImageInvocation(BaseInvocation, WithMetadata, WithBoard):
     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)
+        estimated_working_memory = estimate_vae_working_memory_cogview4(
+            operation="decode", image_tensor=latents, vae=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),

invokeai/app/invocations/cogview4_model_loader.py

@@ -5,7 +5,7 @@ from invokeai.app.invocations.baseinvocation import (
     invocation,
     invocation_output,
 )
-from invokeai.app.invocations.fields import FieldDescriptions, Input, InputField, OutputField, UIType
+from invokeai.app.invocations.fields import FieldDescriptions, Input, InputField, OutputField
 from invokeai.app.invocations.model import (
     GlmEncoderField,
     ModelIdentifierField,
@@ -14,6 +14,7 @@ from invokeai.app.invocations.model import (
 )
 from invokeai.app.services.shared.invocation_context import InvocationContext
 from invokeai.backend.model_manager.config import SubModelType
+from invokeai.backend.model_manager.taxonomy import BaseModelType, ModelType
 
 
 @invocation_output("cogview4_model_loader_output")
@@ -38,8 +39,9 @@ class CogView4ModelLoaderInvocation(BaseInvocation):
 
     model: ModelIdentifierField = InputField(
         description=FieldDescriptions.cogview4_model,
-        ui_type=UIType.CogView4MainModel,
         input=Input.Direct,
+        ui_model_base=BaseModelType.CogView4,
+        ui_model_type=ModelType.Main,
     )
 
     def invoke(self, context: InvocationContext) -> CogView4ModelLoaderOutput:

invokeai/app/invocations/controlnet.py

@@ -16,7 +16,6 @@ from invokeai.app.invocations.fields import (
     ImageField,
     InputField,
     OutputField,
-    UIType,
 )
 from invokeai.app.invocations.model import ModelIdentifierField
 from invokeai.app.invocations.primitives import ImageOutput
@@ -28,6 +27,7 @@ from invokeai.app.util.controlnet_utils import (
     heuristic_resize_fast,
 )
 from invokeai.backend.image_util.util import np_to_pil, pil_to_np
+from invokeai.backend.model_manager.taxonomy import BaseModelType, ModelType
 
 
 class ControlField(BaseModel):
@@ -63,13 +63,17 @@ class ControlOutput(BaseInvocationOutput):
     control: ControlField = OutputField(description=FieldDescriptions.control)
 
 
-@invocation("controlnet", title="ControlNet - SD1.5, SDXL", tags=["controlnet"], category="controlnet", version="1.1.3")
+@invocation(
+    "controlnet", title="ControlNet - SD1.5, SD2, 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
+        description=FieldDescriptions.controlnet_model,
+        ui_model_base=[BaseModelType.StableDiffusion1, BaseModelType.StableDiffusion2, BaseModelType.StableDiffusionXL],
+        ui_model_type=ModelType.ControlNet,
     )
     control_weight: Union[float, List[float]] = InputField(
         default=1.0, ge=-1, le=2, description="The weight given to the ControlNet"

invokeai/app/invocations/fields.py

@@ -1,11 +1,19 @@
 from enum import Enum
 from typing import Any, Callable, Optional, Tuple
 
-from pydantic import BaseModel, ConfigDict, Field, RootModel, TypeAdapter, model_validator
+from pydantic import BaseModel, ConfigDict, Field, RootModel, TypeAdapter
 from pydantic.fields import _Unset
 from pydantic_core import PydanticUndefined
 
 from invokeai.app.util.metaenum import MetaEnum
+from invokeai.backend.image_util.segment_anything.shared import BoundingBox
+from invokeai.backend.model_manager.taxonomy import (
+    BaseModelType,
+    ClipVariantType,
+    ModelFormat,
+    ModelType,
+    ModelVariantType,
+)
 from invokeai.backend.util.logging import InvokeAILogger
 
 logger = InvokeAILogger.get_logger()
@@ -38,35 +46,6 @@ class UIType(str, Enum, metaclass=MetaEnum):
     used, and the type will be ignored. They are included here for backwards compatibility.
     """
 
-    # region Model Field Types
-    MainModel = "MainModelField"
-    CogView4MainModel = "CogView4MainModelField"
-    FluxMainModel = "FluxMainModelField"
-    SD3MainModel = "SD3MainModelField"
-    SDXLMainModel = "SDXLMainModelField"
-    SDXLRefinerModel = "SDXLRefinerModelField"
-    ONNXModel = "ONNXModelField"
-    VAEModel = "VAEModelField"
-    FluxVAEModel = "FluxVAEModelField"
-    LoRAModel = "LoRAModelField"
-    ControlNetModel = "ControlNetModelField"
-    IPAdapterModel = "IPAdapterModelField"
-    T2IAdapterModel = "T2IAdapterModelField"
-    T5EncoderModel = "T5EncoderModelField"
-    CLIPEmbedModel = "CLIPEmbedModelField"
-    CLIPLEmbedModel = "CLIPLEmbedModelField"
-    CLIPGEmbedModel = "CLIPGEmbedModelField"
-    SpandrelImageToImageModel = "SpandrelImageToImageModelField"
-    ControlLoRAModel = "ControlLoRAModelField"
-    SigLipModel = "SigLipModelField"
-    FluxReduxModel = "FluxReduxModelField"
-    LlavaOnevisionModel = "LLaVAModelField"
-    Imagen3Model = "Imagen3ModelField"
-    Imagen4Model = "Imagen4ModelField"
-    ChatGPT4oModel = "ChatGPT4oModelField"
-    FluxKontextModel = "FluxKontextModelField"
-    # endregion
-
     # region Misc Field Types
     Scheduler = "SchedulerField"
     Any = "AnyField"
@@ -75,6 +54,7 @@ class UIType(str, Enum, metaclass=MetaEnum):
     # region Internal Field Types
     _Collection = "CollectionField"
     _CollectionItem = "CollectionItemField"
+    _IsIntermediate = "IsIntermediate"
     # endregion
 
     # region DEPRECATED
@@ -112,13 +92,44 @@ class UIType(str, Enum, metaclass=MetaEnum):
     CollectionItem = "DEPRECATED_CollectionItem"
     Enum = "DEPRECATED_Enum"
     WorkflowField = "DEPRECATED_WorkflowField"
-    IsIntermediate = "DEPRECATED_IsIntermediate"
     BoardField = "DEPRECATED_BoardField"
     MetadataItem = "DEPRECATED_MetadataItem"
     MetadataItemCollection = "DEPRECATED_MetadataItemCollection"
     MetadataItemPolymorphic = "DEPRECATED_MetadataItemPolymorphic"
     MetadataDict = "DEPRECATED_MetadataDict"
 
+    # Deprecated Model Field Types - use ui_model_[base|type|variant|format] instead
+    MainModel = "DEPRECATED_MainModelField"
+    CogView4MainModel = "DEPRECATED_CogView4MainModelField"
+    FluxMainModel = "DEPRECATED_FluxMainModelField"
+    SD3MainModel = "DEPRECATED_SD3MainModelField"
+    SDXLMainModel = "DEPRECATED_SDXLMainModelField"
+    SDXLRefinerModel = "DEPRECATED_SDXLRefinerModelField"
+    ONNXModel = "DEPRECATED_ONNXModelField"
+    VAEModel = "DEPRECATED_VAEModelField"
+    FluxVAEModel = "DEPRECATED_FluxVAEModelField"
+    LoRAModel = "DEPRECATED_LoRAModelField"
+    ControlNetModel = "DEPRECATED_ControlNetModelField"
+    IPAdapterModel = "DEPRECATED_IPAdapterModelField"
+    T2IAdapterModel = "DEPRECATED_T2IAdapterModelField"
+    T5EncoderModel = "DEPRECATED_T5EncoderModelField"
+    CLIPEmbedModel = "DEPRECATED_CLIPEmbedModelField"
+    CLIPLEmbedModel = "DEPRECATED_CLIPLEmbedModelField"
+    CLIPGEmbedModel = "DEPRECATED_CLIPGEmbedModelField"
+    SpandrelImageToImageModel = "DEPRECATED_SpandrelImageToImageModelField"
+    ControlLoRAModel = "DEPRECATED_ControlLoRAModelField"
+    SigLipModel = "DEPRECATED_SigLipModelField"
+    FluxReduxModel = "DEPRECATED_FluxReduxModelField"
+    LlavaOnevisionModel = "DEPRECATED_LLaVAModelField"
+    Imagen3Model = "DEPRECATED_Imagen3ModelField"
+    Imagen4Model = "DEPRECATED_Imagen4ModelField"
+    ChatGPT4oModel = "DEPRECATED_ChatGPT4oModelField"
+    Gemini2_5Model = "DEPRECATED_Gemini2_5ModelField"
+    FluxKontextModel = "DEPRECATED_FluxKontextModelField"
+    Veo3Model = "DEPRECATED_Veo3ModelField"
+    RunwayModel = "DEPRECATED_RunwayModelField"
+    # endregion
+
 
 class UIComponent(str, Enum, metaclass=MetaEnum):
     """
@@ -224,6 +235,12 @@ class ImageField(BaseModel):
     image_name: str = Field(description="The name of the image")
 
 
+class VideoField(BaseModel):
+    """A video primitive field"""
+
+    video_id: str = Field(description="The id of the video")
+
+
 class BoardField(BaseModel):
     """A board primitive field"""
 
@@ -321,14 +338,9 @@ class ConditioningField(BaseModel):
     )
 
 
-class BoundingBoxField(BaseModel):
+class BoundingBoxField(BoundingBox):
     """A bounding box primitive value."""
 
-    x_min: int = Field(ge=0, description="The minimum x-coordinate of the bounding box (inclusive).")
-    x_max: int = Field(ge=0, description="The maximum x-coordinate of the bounding box (exclusive).")
-    y_min: int = Field(ge=0, description="The minimum y-coordinate of the bounding box (inclusive).")
-    y_max: int = Field(ge=0, description="The maximum y-coordinate of the bounding box (exclusive).")
-
     score: Optional[float] = Field(
         default=None,
         ge=0.0,
@@ -337,21 +349,6 @@ class BoundingBoxField(BaseModel):
         "when the bounding box was produced by a detector and has an associated confidence score.",
     )
 
-    @model_validator(mode="after")
-    def check_coords(self):
-        if self.x_min > self.x_max:
-            raise ValueError(f"x_min ({self.x_min}) is greater than x_max ({self.x_max}).")
-        if self.y_min > self.y_max:
-            raise ValueError(f"y_min ({self.y_min}) is greater than y_max ({self.y_max}).")
-        return self
-
-    def tuple(self) -> Tuple[int, int, int, int]:
-        """
-        Returns the bounding box as a tuple suitable for use with PIL's `Image.crop()` method.
-        This method returns a tuple of the form (left, upper, right, lower) == (x_min, y_min, x_max, y_max).
-        """
-        return (self.x_min, self.y_min, self.x_max, self.y_max)
-
 
 class MetadataField(RootModel[dict[str, Any]]):
     """
@@ -418,10 +415,15 @@ class InputFieldJSONSchemaExtra(BaseModel):
     ui_component: Optional[UIComponent] = None
     ui_order: Optional[int] = None
     ui_choice_labels: Optional[dict[str, str]] = None
+    ui_model_base: Optional[list[BaseModelType]] = None
+    ui_model_type: Optional[list[ModelType]] = None
+    ui_model_variant: Optional[list[ClipVariantType | ModelVariantType]] = None
+    ui_model_format: Optional[list[ModelFormat]] = None
 
     model_config = ConfigDict(
         validate_assignment=True,
         json_schema_serialization_defaults_required=True,
+        use_enum_values=True,
     )
 
 
@@ -474,16 +476,121 @@ class OutputFieldJSONSchemaExtra(BaseModel):
     """
 
     field_kind: FieldKind
-    ui_hidden: bool
-    ui_type: Optional[UIType]
-    ui_order: Optional[int]
+    ui_hidden: bool = False
+    ui_order: Optional[int] = None
+    ui_type: Optional[UIType] = None
 
     model_config = ConfigDict(
         validate_assignment=True,
         json_schema_serialization_defaults_required=True,
+        use_enum_values=True,
     )
 
 
+def migrate_model_ui_type(ui_type: UIType | str, json_schema_extra: dict[str, Any]) -> bool:
+    """Migrate deprecated model-specifier ui_type values to new-style ui_model_[base|type|variant|format] in json_schema_extra."""
+    if not isinstance(ui_type, UIType):
+        ui_type = UIType(ui_type)
+
+    ui_model_type: list[ModelType] | None = None
+    ui_model_base: list[BaseModelType] | None = None
+    ui_model_format: list[ModelFormat] | None = None
+    ui_model_variant: list[ClipVariantType | ModelVariantType] | None = None
+
+    match ui_type:
+        case UIType.MainModel:
+            ui_model_base = [BaseModelType.StableDiffusion1, BaseModelType.StableDiffusion2]
+            ui_model_type = [ModelType.Main]
+        case UIType.CogView4MainModel:
+            ui_model_base = [BaseModelType.CogView4]
+            ui_model_type = [ModelType.Main]
+        case UIType.FluxMainModel:
+            ui_model_base = [BaseModelType.Flux]
+            ui_model_type = [ModelType.Main]
+        case UIType.SD3MainModel:
+            ui_model_base = [BaseModelType.StableDiffusion3]
+            ui_model_type = [ModelType.Main]
+        case UIType.SDXLMainModel:
+            ui_model_base = [BaseModelType.StableDiffusionXL]
+            ui_model_type = [ModelType.Main]
+        case UIType.SDXLRefinerModel:
+            ui_model_base = [BaseModelType.StableDiffusionXLRefiner]
+            ui_model_type = [ModelType.Main]
+        case UIType.VAEModel:
+            ui_model_type = [ModelType.VAE]
+        case UIType.FluxVAEModel:
+            ui_model_base = [BaseModelType.Flux]
+            ui_model_type = [ModelType.VAE]
+        case UIType.LoRAModel:
+            ui_model_type = [ModelType.LoRA]
+        case UIType.ControlNetModel:
+            ui_model_type = [ModelType.ControlNet]
+        case UIType.IPAdapterModel:
+            ui_model_type = [ModelType.IPAdapter]
+        case UIType.T2IAdapterModel:
+            ui_model_type = [ModelType.T2IAdapter]
+        case UIType.T5EncoderModel:
+            ui_model_type = [ModelType.T5Encoder]
+        case UIType.CLIPEmbedModel:
+            ui_model_type = [ModelType.CLIPEmbed]
+        case UIType.CLIPLEmbedModel:
+            ui_model_type = [ModelType.CLIPEmbed]
+            ui_model_variant = [ClipVariantType.L]
+        case UIType.CLIPGEmbedModel:
+            ui_model_type = [ModelType.CLIPEmbed]
+            ui_model_variant = [ClipVariantType.G]
+        case UIType.SpandrelImageToImageModel:
+            ui_model_type = [ModelType.SpandrelImageToImage]
+        case UIType.ControlLoRAModel:
+            ui_model_type = [ModelType.ControlLoRa]
+        case UIType.SigLipModel:
+            ui_model_type = [ModelType.SigLIP]
+        case UIType.FluxReduxModel:
+            ui_model_type = [ModelType.FluxRedux]
+        case UIType.LlavaOnevisionModel:
+            ui_model_type = [ModelType.LlavaOnevision]
+        case UIType.Imagen3Model:
+            ui_model_base = [BaseModelType.Imagen3]
+            ui_model_type = [ModelType.Main]
+        case UIType.Imagen4Model:
+            ui_model_base = [BaseModelType.Imagen4]
+            ui_model_type = [ModelType.Main]
+        case UIType.ChatGPT4oModel:
+            ui_model_base = [BaseModelType.ChatGPT4o]
+            ui_model_type = [ModelType.Main]
+        case UIType.Gemini2_5Model:
+            ui_model_base = [BaseModelType.Gemini2_5]
+            ui_model_type = [ModelType.Main]
+        case UIType.FluxKontextModel:
+            ui_model_base = [BaseModelType.FluxKontext]
+            ui_model_type = [ModelType.Main]
+        case UIType.Veo3Model:
+            ui_model_base = [BaseModelType.Veo3]
+            ui_model_type = [ModelType.Video]
+        case UIType.RunwayModel:
+            ui_model_base = [BaseModelType.Runway]
+            ui_model_type = [ModelType.Video]
+        case _:
+            pass
+
+    did_migrate = False
+
+    if ui_model_type is not None:
+        json_schema_extra["ui_model_type"] = [m.value for m in ui_model_type]
+        did_migrate = True
+    if ui_model_base is not None:
+        json_schema_extra["ui_model_base"] = [m.value for m in ui_model_base]
+        did_migrate = True
+    if ui_model_format is not None:
+        json_schema_extra["ui_model_format"] = [m.value for m in ui_model_format]
+        did_migrate = True
+    if ui_model_variant is not None:
+        json_schema_extra["ui_model_variant"] = [m.value for m in ui_model_variant]
+        did_migrate = True
+
+    return did_migrate
+
+
 def InputField(
     # copied from pydantic's Field
     # TODO: Can we support default_factory?
@@ -510,35 +617,63 @@ def InputField(
     ui_hidden: Optional[bool] = None,
     ui_order: Optional[int] = None,
     ui_choice_labels: Optional[dict[str, str]] = None,
+    ui_model_base: Optional[BaseModelType | list[BaseModelType]] = None,
+    ui_model_type: Optional[ModelType | list[ModelType]] = None,
+    ui_model_variant: Optional[ClipVariantType | ModelVariantType | list[ClipVariantType | ModelVariantType]] = None,
+    ui_model_format: Optional[ModelFormat | list[ModelFormat]] = None,
 ) -> Any:
     """
     Creates an input field for an invocation.
 
-    This is a wrapper for Pydantic's [Field](https://docs.pydantic.dev/latest/api/fields/#pydantic.fields.Field) \
+    This is a wrapper for Pydantic's [Field](https://docs.pydantic.dev/latest/api/fields/#pydantic.fields.Field)
     that adds a few extra parameters to support graph execution and the node editor UI.
 
-    :param Input input: [Input.Any] The kind of input this field requires. \
-      `Input.Direct` means a value must be provided on instantiation. \
-      `Input.Connection` means the value must be provided by a connection. \
-      `Input.Any` means either will do.
+    If the field is a `ModelIdentifierField`, use the `ui_model_[base|type|variant|format]` args to filter the model list
+    in the Workflow Editor. Otherwise, use `ui_type` to provide extra type hints for the UI.
 
-    :param UIType ui_type: [None] Optionally provides an extra type hint for the UI. \
-      In some situations, the field's type is not enough to infer the correct UI type. \
-      For example, model selection fields should render a dropdown UI component to select a model. \
-      Internally, there is no difference between SD-1, SD-2 and SDXL model fields, they all use \
-      `MainModelField`. So to ensure the base-model-specific UI is rendered, you can use \
-      `UIType.SDXLMainModelField` to indicate that the field is an SDXL main model field.
+    Don't use both `ui_type` and `ui_model_[base|type|variant|format]` - if both are provided, a warning will be
+    logged and `ui_type` will be ignored.
 
-    :param UIComponent ui_component: [None] Optionally specifies a specific component to use in the UI. \
-      The UI will always render a suitable component, but sometimes you want something different than the default. \
-      For example, a `string` field will default to a single-line input, but you may want a multi-line textarea instead. \
-      For this case, you could provide `UIComponent.Textarea`.
+    Args:
+        input: The kind of input this field requires.
+        - `Input.Direct` means a value must be provided on instantiation.
+        - `Input.Connection` means the value must be provided by a connection.
+        - `Input.Any` means either will do.
 
-    :param bool ui_hidden: [False] Specifies whether or not this field should be hidden in the UI.
+        ui_type: Optionally provides an extra type hint for the UI. In some situations, the field's type is not enough
+        to infer the correct UI type. For example, Scheduler fields are enums, but we want to render a special scheduler
+        dropdown in the UI. Use `UIType.Scheduler` to indicate this.
 
-    :param int ui_order: [None] Specifies the order in which this field should be rendered in the UI.
+        ui_component: Optionally specifies a specific component to use in the UI. The UI will always render a suitable
+        component, but sometimes you want something different than the default. For example, a `string` field will
+        default to a single-line input, but you may want a multi-line textarea instead. In this case, you could use
+        `UIComponent.Textarea`.
 
-    :param dict[str, str] ui_choice_labels: [None] Specifies the labels to use for the choices in an enum field.
+        ui_hidden: Specifies whether or not this field should be hidden in the UI.
+
+        ui_order: Specifies the order in which this field should be rendered in the UI. If omitted, the field will be
+        rendered after all fields with an explicit order, in the order they are defined in the Invocation class.
+
+        ui_model_base: Specifies the base model architectures to filter the model list by in the Workflow Editor. For
+        example, `ui_model_base=BaseModelType.StableDiffusionXL` will show only SDXL architecture models. This arg is
+        only valid if this Input field is annotated as a `ModelIdentifierField`.
+
+        ui_model_type: Specifies the model type(s) to filter the model list by in the Workflow Editor. For example,
+        `ui_model_type=ModelType.VAE` will show only VAE models. This arg is only valid if this Input field is
+        annotated as a `ModelIdentifierField`.
+
+        ui_model_variant: Specifies the model variant(s) to filter the model list by in the Workflow Editor. For example,
+        `ui_model_variant=ModelVariantType.Inpainting` will show only inpainting models. This arg is only valid if this
+        Input field is annotated as a `ModelIdentifierField`.
+
+        ui_model_format: Specifies the model format(s) to filter the model list by in the Workflow Editor. For example,
+        `ui_model_format=ModelFormat.Diffusers` will show only models in the diffusers format. This arg is only valid
+        if this Input field is annotated as a `ModelIdentifierField`.
+
+        ui_choice_labels: Specifies the labels to use for the choices in an enum field. If omitted, the enum values
+        will be used. This arg is only valid if the field is annotated with as a `Literal`. For example,
+        `Literal["choice1", "choice2", "choice3"]` with `ui_choice_labels={"choice1": "Choice 1", "choice2": "Choice 2",
+        "choice3": "Choice 3"}` will render a dropdown with the labels "Choice 1", "Choice 2" and "Choice 3".
     """
 
     json_schema_extra_ = InputFieldJSONSchemaExtra(
@@ -546,8 +681,6 @@ def InputField(
         field_kind=FieldKind.Input,
     )
 
-    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:
@@ -556,6 +689,28 @@ def InputField(
         json_schema_extra_.ui_order = ui_order
     if ui_choice_labels is not None:
         json_schema_extra_.ui_choice_labels = ui_choice_labels
+    if ui_model_base is not None:
+        if isinstance(ui_model_base, list):
+            json_schema_extra_.ui_model_base = ui_model_base
+        else:
+            json_schema_extra_.ui_model_base = [ui_model_base]
+    if ui_model_type is not None:
+        if isinstance(ui_model_type, list):
+            json_schema_extra_.ui_model_type = ui_model_type
+        else:
+            json_schema_extra_.ui_model_type = [ui_model_type]
+    if ui_model_variant is not None:
+        if isinstance(ui_model_variant, list):
+            json_schema_extra_.ui_model_variant = ui_model_variant
+        else:
+            json_schema_extra_.ui_model_variant = [ui_model_variant]
+    if ui_model_format is not None:
+        if isinstance(ui_model_format, list):
+            json_schema_extra_.ui_model_format = ui_model_format
+        else:
+            json_schema_extra_.ui_model_format = [ui_model_format]
+    if ui_type is not None:
+        json_schema_extra_.ui_type = ui_type
 
     """
     There is a conflict between the typing of invocation definitions and the typing of an invocation's
@@ -657,20 +812,20 @@ def OutputField(
     """
     Creates an output field for an invocation output.
 
-    This is a wrapper for Pydantic's [Field](https://docs.pydantic.dev/1.10/usage/schema/#field-customization) \
+    This is a wrapper for Pydantic's [Field](https://docs.pydantic.dev/1.10/usage/schema/#field-customization)
     that adds a few extra parameters to support graph execution and the node editor UI.
 
-    :param UIType ui_type: [None] Optionally provides an extra type hint for the UI. \
-      In some situations, the field's type is not enough to infer the correct UI type. \
-      For example, model selection fields should render a dropdown UI component to select a model. \
-      Internally, there is no difference between SD-1, SD-2 and SDXL model fields, they all use \
-      `MainModelField`. So to ensure the base-model-specific UI is rendered, you can use \
-      `UIType.SDXLMainModelField` to indicate that the field is an SDXL main model field.
+    Args:
+        ui_type: Optionally provides an extra type hint for the UI. In some situations, the field's type is not enough
+        to infer the correct UI type. For example, Scheduler fields are enums, but we want to render a special scheduler
+        dropdown in the UI. Use `UIType.Scheduler` to indicate this.
 
-    :param bool ui_hidden: [False] Specifies whether or not this field should be hidden in the UI. \
+        ui_hidden: Specifies whether or not this field should be hidden in the UI.
 
-    :param int ui_order: [None] Specifies the order in which this field should be rendered in the UI. \
+        ui_order: Specifies the order in which this field should be rendered in the UI. If omitted, the field will be
+        rendered after all fields with an explicit order, in the order they are defined in the Invocation class.
     """
+
     return Field(
         default=default,
         title=title,
@@ -688,9 +843,9 @@ def OutputField(
         min_length=min_length,
         max_length=max_length,
         json_schema_extra=OutputFieldJSONSchemaExtra(
-            ui_type=ui_type,
             ui_hidden=ui_hidden,
             ui_order=ui_order,
+            ui_type=ui_type,
             field_kind=FieldKind.Output,
         ).model_dump(exclude_none=True),
     )

invokeai/app/invocations/flux_control_lora_loader.py

@@ -4,9 +4,10 @@ from invokeai.app.invocations.baseinvocation import (
     invocation,
     invocation_output,
 )
-from invokeai.app.invocations.fields import FieldDescriptions, ImageField, InputField, OutputField, UIType
+from invokeai.app.invocations.fields import FieldDescriptions, ImageField, InputField, OutputField
 from invokeai.app.invocations.model import ControlLoRAField, ModelIdentifierField
 from invokeai.app.services.shared.invocation_context import InvocationContext
+from invokeai.backend.model_manager.taxonomy import BaseModelType, ModelType
 
 
 @invocation_output("flux_control_lora_loader_output")
@@ -29,7 +30,10 @@ class FluxControlLoRALoaderInvocation(BaseInvocation):
     """LoRA model and Image to use with FLUX transformer generation."""
 
     lora: ModelIdentifierField = InputField(
-        description=FieldDescriptions.control_lora_model, title="Control LoRA", ui_type=UIType.ControlLoRAModel
+        description=FieldDescriptions.control_lora_model,
+        title="Control LoRA",
+        ui_model_base=BaseModelType.Flux,
+        ui_model_type=ModelType.ControlLoRa,
     )
     image: ImageField = InputField(description="The image to encode.")
     weight: float = InputField(description="The weight of the LoRA.", default=1.0)

invokeai/app/invocations/flux_controlnet.py

@@ -6,11 +6,12 @@ from invokeai.app.invocations.baseinvocation import (
     invocation,
     invocation_output,
 )
-from invokeai.app.invocations.fields import FieldDescriptions, ImageField, InputField, OutputField, UIType
+from invokeai.app.invocations.fields import FieldDescriptions, ImageField, InputField, OutputField
 from invokeai.app.invocations.model import ModelIdentifierField
 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_RESIZE_VALUES
+from invokeai.backend.model_manager.taxonomy import BaseModelType, ModelType
 
 
 class FluxControlNetField(BaseModel):
@@ -57,7 +58,9 @@ class FluxControlNetInvocation(BaseInvocation):
 
     image: ImageField = InputField(description="The control image")
     control_model: ModelIdentifierField = InputField(
-        description=FieldDescriptions.controlnet_model, ui_type=UIType.ControlNetModel
+        description=FieldDescriptions.controlnet_model,
+        ui_model_base=BaseModelType.Flux,
+        ui_model_type=ModelType.ControlNet,
     )
     control_weight: float | list[float] = InputField(
         default=1.0, ge=-1, le=2, description="The weight given to the ControlNet"

invokeai/app/invocations/flux_denoise.py

@@ -63,7 +63,7 @@ from invokeai.backend.util.devices import TorchDevice
     title="FLUX Denoise",
     tags=["image", "flux"],
     category="image",
-    version="4.0.0",
+    version="4.1.0",
 )
 class FluxDenoiseInvocation(BaseInvocation):
     """Run denoising process with a FLUX transformer model."""
@@ -153,7 +153,7 @@ class FluxDenoiseInvocation(BaseInvocation):
         description=FieldDescriptions.ip_adapter, title="IP-Adapter", default=None, input=Input.Connection
     )
 
-    kontext_conditioning: Optional[FluxKontextConditioningField] = InputField(
+    kontext_conditioning: FluxKontextConditioningField | list[FluxKontextConditioningField] | None = InputField(
         default=None,
         description="FLUX Kontext conditioning (reference image).",
         input=Input.Connection,
@@ -328,6 +328,21 @@ class FluxDenoiseInvocation(BaseInvocation):
             cfg_scale_end_step=self.cfg_scale_end_step,
         )
 
+        kontext_extension = None
+        if self.kontext_conditioning:
+            if not self.controlnet_vae:
+                raise ValueError("A VAE (e.g., controlnet_vae) must be provided to use Kontext conditioning.")
+
+            kontext_extension = KontextExtension(
+                context=context,
+                kontext_conditioning=self.kontext_conditioning
+                if isinstance(self.kontext_conditioning, list)
+                else [self.kontext_conditioning],
+                vae_field=self.controlnet_vae,
+                device=TorchDevice.choose_torch_device(),
+                dtype=inference_dtype,
+            )
+
         with ExitStack() as exit_stack:
             # Prepare ControlNet extensions.
             # Note: We do this before loading the transformer model to minimize peak memory (see implementation).
@@ -385,19 +400,6 @@ class FluxDenoiseInvocation(BaseInvocation):
                 dtype=inference_dtype,
             )
 
-            kontext_extension = None
-            if self.kontext_conditioning is not None:
-                if not self.controlnet_vae:
-                    raise ValueError("A VAE (e.g., controlnet_vae) must be provided to use Kontext conditioning.")
-
-                kontext_extension = KontextExtension(
-                    context=context,
-                    kontext_conditioning=self.kontext_conditioning,
-                    vae_field=self.controlnet_vae,
-                    device=TorchDevice.choose_torch_device(),
-                    dtype=inference_dtype,
-                )
-
             # Prepare Kontext conditioning if provided
             img_cond_seq = None
             img_cond_seq_ids = None

invokeai/app/invocations/flux_ip_adapter.py

@@ -5,7 +5,7 @@ from pydantic import field_validator, model_validator
 from typing_extensions import Self
 
 from invokeai.app.invocations.baseinvocation import BaseInvocation, invocation
-from invokeai.app.invocations.fields import InputField, UIType
+from invokeai.app.invocations.fields import InputField
 from invokeai.app.invocations.ip_adapter import (
     CLIP_VISION_MODEL_MAP,
     IPAdapterField,
@@ -20,6 +20,7 @@ from invokeai.backend.model_manager.config import (
     IPAdapterCheckpointConfig,
     IPAdapterInvokeAIConfig,
 )
+from invokeai.backend.model_manager.taxonomy import BaseModelType, ModelType
 
 
 @invocation(
@@ -36,7 +37,10 @@ class FluxIPAdapterInvocation(BaseInvocation):
 
     image: ImageField = InputField(description="The IP-Adapter image prompt(s).")
     ip_adapter_model: ModelIdentifierField = InputField(
-        description="The IP-Adapter model.", title="IP-Adapter Model", ui_type=UIType.IPAdapterModel
+        description="The IP-Adapter model.",
+        title="IP-Adapter Model",
+        ui_model_base=BaseModelType.Flux,
+        ui_model_type=ModelType.IPAdapter,
     )
     # Currently, the only known ViT model used by FLUX IP-Adapters is ViT-L.
     clip_vision_model: Literal["ViT-L"] = InputField(description="CLIP Vision model to use.", default="ViT-L")

invokeai/app/invocations/flux_lora_loader.py

@@ -6,10 +6,10 @@ from invokeai.app.invocations.baseinvocation import (
     invocation,
     invocation_output,
 )
-from invokeai.app.invocations.fields import FieldDescriptions, Input, InputField, OutputField, UIType
+from invokeai.app.invocations.fields import FieldDescriptions, Input, InputField, OutputField
 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.taxonomy import BaseModelType
+from invokeai.backend.model_manager.taxonomy import BaseModelType, ModelType
 
 
 @invocation_output("flux_lora_loader_output")
@@ -36,7 +36,10 @@ class FluxLoRALoaderInvocation(BaseInvocation):
     """Apply a LoRA model to a FLUX transformer and/or text encoder."""
 
     lora: ModelIdentifierField = InputField(
-        description=FieldDescriptions.lora_model, title="LoRA", ui_type=UIType.LoRAModel
+        description=FieldDescriptions.lora_model,
+        title="LoRA",
+        ui_model_base=BaseModelType.Flux,
+        ui_model_type=ModelType.LoRA,
     )
     weight: float = InputField(default=0.75, description=FieldDescriptions.lora_weight)
     transformer: TransformerField | None = InputField(

invokeai/app/invocations/flux_model_loader.py

@@ -6,7 +6,7 @@ from invokeai.app.invocations.baseinvocation import (
     invocation,
     invocation_output,
 )
-from invokeai.app.invocations.fields import FieldDescriptions, Input, InputField, OutputField, UIType
+from invokeai.app.invocations.fields import FieldDescriptions, Input, InputField, OutputField
 from invokeai.app.invocations.model import CLIPField, ModelIdentifierField, T5EncoderField, TransformerField, VAEField
 from invokeai.app.services.shared.invocation_context import InvocationContext
 from invokeai.app.util.t5_model_identifier import (
@@ -17,7 +17,7 @@ from invokeai.backend.flux.util import max_seq_lengths
 from invokeai.backend.model_manager.config import (
     CheckpointConfigBase,
 )
-from invokeai.backend.model_manager.taxonomy import SubModelType
+from invokeai.backend.model_manager.taxonomy import BaseModelType, ModelType, SubModelType
 
 
 @invocation_output("flux_model_loader_output")
@@ -46,23 +46,30 @@ class FluxModelLoaderInvocation(BaseInvocation):
 
     model: ModelIdentifierField = InputField(
         description=FieldDescriptions.flux_model,
-        ui_type=UIType.FluxMainModel,
         input=Input.Direct,
+        ui_model_base=BaseModelType.Flux,
+        ui_model_type=ModelType.Main,
     )
 
     t5_encoder_model: ModelIdentifierField = InputField(
-        description=FieldDescriptions.t5_encoder, ui_type=UIType.T5EncoderModel, input=Input.Direct, title="T5 Encoder"
+        description=FieldDescriptions.t5_encoder,
+        input=Input.Direct,
+        title="T5 Encoder",
+        ui_model_type=ModelType.T5Encoder,
     )
 
     clip_embed_model: ModelIdentifierField = InputField(
         description=FieldDescriptions.clip_embed_model,
-        ui_type=UIType.CLIPEmbedModel,
         input=Input.Direct,
         title="CLIP Embed",
+        ui_model_type=ModelType.CLIPEmbed,
     )
 
     vae_model: ModelIdentifierField = InputField(
-        description=FieldDescriptions.vae_model, ui_type=UIType.FluxVAEModel, title="VAE"
+        description=FieldDescriptions.vae_model,
+        title="VAE",
+        ui_model_base=BaseModelType.Flux,
+        ui_model_type=ModelType.VAE,
     )
 
     def invoke(self, context: InvocationContext) -> FluxModelLoaderOutput:

invokeai/app/invocations/flux_redux.py

@@ -18,7 +18,6 @@ from invokeai.app.invocations.fields import (
     InputField,
     OutputField,
     TensorField,
-    UIType,
 )
 from invokeai.app.invocations.model import ModelIdentifierField
 from invokeai.app.invocations.primitives import ImageField
@@ -64,7 +63,8 @@ class FluxReduxInvocation(BaseInvocation):
     redux_model: ModelIdentifierField = InputField(
         description="The FLUX Redux model to use.",
         title="FLUX Redux Model",
-        ui_type=UIType.FluxReduxModel,
+        ui_model_base=BaseModelType.Flux,
+        ui_model_type=ModelType.FluxRedux,
     )
     downsampling_factor: int = InputField(
         ge=1,

invokeai/app/invocations/flux_vae_decode.py

@@ -3,7 +3,6 @@ from einops import rearrange
 from PIL import Image
 
 from invokeai.app.invocations.baseinvocation import BaseInvocation, invocation
-from invokeai.app.invocations.constants import LATENT_SCALE_FACTOR
 from invokeai.app.invocations.fields import (
     FieldDescriptions,
     Input,
@@ -18,6 +17,7 @@ from invokeai.app.services.shared.invocation_context import InvocationContext
 from invokeai.backend.flux.modules.autoencoder import AutoEncoder
 from invokeai.backend.model_manager.load.load_base import LoadedModel
 from invokeai.backend.util.devices import TorchDevice
+from invokeai.backend.util.vae_working_memory import estimate_vae_working_memory_flux
 
 
 @invocation(
@@ -39,17 +39,11 @@ class FluxVaeDecodeInvocation(BaseInvocation, WithMetadata, WithBoard):
         input=Input.Connection,
     )
 
-    def _estimate_working_memory(self, latents: torch.Tensor, vae: AutoEncoder) -> 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)
-
     def _vae_decode(self, vae_info: LoadedModel, latents: torch.Tensor) -> Image.Image:
-        estimated_working_memory = self._estimate_working_memory(latents, vae_info.model)
+        assert isinstance(vae_info.model, AutoEncoder)
+        estimated_working_memory = estimate_vae_working_memory_flux(
+            operation="decode", image_tensor=latents, vae=vae_info.model
+        )
         with vae_info.model_on_device(working_mem_bytes=estimated_working_memory) as (_, vae):
             assert isinstance(vae, AutoEncoder)
             vae_dtype = next(iter(vae.parameters())).dtype

invokeai/app/invocations/flux_vae_encode.py

@@ -15,6 +15,7 @@ from invokeai.backend.flux.modules.autoencoder import AutoEncoder
 from invokeai.backend.model_manager import LoadedModel
 from invokeai.backend.stable_diffusion.diffusers_pipeline import image_resized_to_grid_as_tensor
 from invokeai.backend.util.devices import TorchDevice
+from invokeai.backend.util.vae_working_memory import estimate_vae_working_memory_flux
 
 
 @invocation(
@@ -41,8 +42,12 @@ class FluxVaeEncodeInvocation(BaseInvocation):
         # TODO(ryand): Write a util function for generating random tensors that is consistent across devices / dtypes.
         # There's a starting point in get_noise(...), but it needs to be extracted and generalized. This function
         # should be used for VAE encode sampling.
+        assert isinstance(vae_info.model, AutoEncoder)
+        estimated_working_memory = estimate_vae_working_memory_flux(
+            operation="encode", image_tensor=image_tensor, vae=vae_info.model
+        )
         generator = torch.Generator(device=TorchDevice.choose_torch_device()).manual_seed(0)
-        with vae_info as vae:
+        with vae_info.model_on_device(working_mem_bytes=estimated_working_memory) as (_, vae):
             assert isinstance(vae, AutoEncoder)
             vae_dtype = next(iter(vae.parameters())).dtype
             image_tensor = image_tensor.to(device=TorchDevice.choose_torch_device(), dtype=vae_dtype)

invokeai/app/invocations/image.py

@@ -1347,3 +1347,96 @@ class PasteImageIntoBoundingBoxInvocation(BaseInvocation, WithMetadata, WithBoar
 
         image_dto = context.images.save(image=target_image)
         return ImageOutput.build(image_dto)
+
+
+@invocation(
+    "flux_kontext_image_prep",
+    title="FLUX Kontext Image Prep",
+    tags=["image", "concatenate", "flux", "kontext"],
+    category="image",
+    version="1.0.0",
+)
+class FluxKontextConcatenateImagesInvocation(BaseInvocation, WithMetadata, WithBoard):
+    """Prepares an image or images for use with FLUX Kontext. The first/single image is resized to the nearest
+    preferred Kontext resolution. All other images are concatenated horizontally, maintaining their aspect ratio."""
+
+    images: list[ImageField] = InputField(
+        description="The images to concatenate",
+        min_length=1,
+        max_length=10,
+    )
+
+    use_preferred_resolution: bool = InputField(
+        default=True, description="Use FLUX preferred resolutions for the first image"
+    )
+
+    def invoke(self, context: InvocationContext) -> ImageOutput:
+        from invokeai.backend.flux.util import PREFERED_KONTEXT_RESOLUTIONS
+
+        # Step 1: Load all images
+        pil_images = []
+        for image_field in self.images:
+            image = context.images.get_pil(image_field.image_name, mode="RGBA")
+            pil_images.append(image)
+
+        # Step 2: Determine target resolution for the first image
+        first_image = pil_images[0]
+        width, height = first_image.size
+
+        if self.use_preferred_resolution:
+            aspect_ratio = width / height
+
+            # Find the closest preferred resolution for the first image
+            _, target_width, target_height = min(
+                ((abs(aspect_ratio - w / h), w, h) for w, h in PREFERED_KONTEXT_RESOLUTIONS), key=lambda x: x[0]
+            )
+
+            # Apply BFL's scaling formula
+            scaled_height = 2 * int(target_height / 16)
+            final_height = 8 * scaled_height  # This will be consistent for all images
+            scaled_width = 2 * int(target_width / 16)
+            first_width = 8 * scaled_width
+        else:
+            # Use original dimensions of first image, ensuring divisibility by 16
+            final_height = 16 * (height // 16)
+            first_width = 16 * (width // 16)
+            # Ensure minimum dimensions
+            if final_height < 16:
+                final_height = 16
+            if first_width < 16:
+                first_width = 16
+
+        # Step 3: Process and resize all images with consistent height
+        processed_images = []
+        total_width = 0
+
+        for i, image in enumerate(pil_images):
+            if i == 0:
+                # First image uses the calculated dimensions
+                final_width = first_width
+            else:
+                # Subsequent images maintain aspect ratio with the same height
+                img_aspect_ratio = image.width / image.height
+                # Calculate width that maintains aspect ratio at the target height
+                calculated_width = int(final_height * img_aspect_ratio)
+                # Ensure width is divisible by 16 for proper VAE encoding
+                final_width = 16 * (calculated_width // 16)
+                # Ensure minimum width
+                if final_width < 16:
+                    final_width = 16
+
+            # Resize image to calculated dimensions
+            resized_image = image.resize((final_width, final_height), Image.Resampling.LANCZOS)
+            processed_images.append(resized_image)
+            total_width += final_width
+
+        # Step 4: Concatenate images horizontally
+        concatenated_image = Image.new("RGB", (total_width, final_height))
+        x_offset = 0
+        for img in processed_images:
+            concatenated_image.paste(img, (x_offset, 0))
+            x_offset += img.width
+
+        # Save the concatenated image
+        image_dto = context.images.save(image=concatenated_image)
+        return ImageOutput.build(image_dto)

invokeai/app/invocations/image_to_latents.py

@@ -27,6 +27,7 @@ from invokeai.backend.model_manager import LoadedModel
 from invokeai.backend.stable_diffusion.diffusers_pipeline import image_resized_to_grid_as_tensor
 from invokeai.backend.stable_diffusion.vae_tiling import patch_vae_tiling_params
 from invokeai.backend.util.devices import TorchDevice
+from invokeai.backend.util.vae_working_memory import estimate_vae_working_memory_sd15_sdxl
 
 
 @invocation(
@@ -52,11 +53,24 @@ class ImageToLatentsInvocation(BaseInvocation):
     tile_size: int = InputField(default=0, multiple_of=8, description=FieldDescriptions.vae_tile_size)
     fp32: bool = InputField(default=False, description=FieldDescriptions.fp32)
 
-    @staticmethod
+    @classmethod
     def vae_encode(
-        vae_info: LoadedModel, upcast: bool, tiled: bool, image_tensor: torch.Tensor, tile_size: int = 0
+        cls,
+        vae_info: LoadedModel,
+        upcast: bool,
+        tiled: bool,
+        image_tensor: torch.Tensor,
+        tile_size: int = 0,
     ) -> torch.Tensor:
-        with vae_info as vae:
+        assert isinstance(vae_info.model, (AutoencoderKL, AutoencoderTiny))
+        estimated_working_memory = estimate_vae_working_memory_sd15_sdxl(
+            operation="encode",
+            image_tensor=image_tensor,
+            vae=vae_info.model,
+            tile_size=tile_size if tiled else None,
+            fp32=upcast,
+        )
+        with vae_info.model_on_device(working_mem_bytes=estimated_working_memory) as (_, vae):
             assert isinstance(vae, (AutoencoderKL, AutoencoderTiny))
             orig_dtype = vae.dtype
             if upcast:
@@ -113,6 +127,7 @@ class ImageToLatentsInvocation(BaseInvocation):
         image = context.images.get_pil(self.image.image_name)
 
         vae_info = context.models.load(self.vae.vae)
+        assert isinstance(vae_info.model, (AutoencoderKL, AutoencoderTiny))
 
         image_tensor = image_resized_to_grid_as_tensor(image.convert("RGB"))
         if image_tensor.dim() == 3:
@@ -120,7 +135,11 @@ class ImageToLatentsInvocation(BaseInvocation):
 
         context.util.signal_progress("Running VAE encoder")
         latents = self.vae_encode(
-            vae_info=vae_info, upcast=self.fp32, tiled=self.tiled, image_tensor=image_tensor, tile_size=self.tile_size
+            vae_info=vae_info,
+            upcast=self.fp32,
+            tiled=self.tiled or context.config.get().force_tiled_decode,
+            image_tensor=image_tensor,
+            tile_size=self.tile_size,
         )
 
         latents = latents.to("cpu")

invokeai/app/invocations/ip_adapter.py

@@ -5,7 +5,7 @@ from pydantic import BaseModel, Field, field_validator, model_validator
 from typing_extensions import Self
 
 from invokeai.app.invocations.baseinvocation import BaseInvocation, BaseInvocationOutput, invocation, invocation_output
-from invokeai.app.invocations.fields import FieldDescriptions, InputField, OutputField, TensorField, UIType
+from invokeai.app.invocations.fields import FieldDescriptions, InputField, OutputField, TensorField
 from invokeai.app.invocations.model import ModelIdentifierField
 from invokeai.app.invocations.primitives import ImageField
 from invokeai.app.invocations.util import validate_begin_end_step, validate_weights
@@ -85,7 +85,8 @@ class IPAdapterInvocation(BaseInvocation):
         description="The IP-Adapter model.",
         title="IP-Adapter Model",
         ui_order=-1,
-        ui_type=UIType.IPAdapterModel,
+        ui_model_base=[BaseModelType.StableDiffusion1, BaseModelType.StableDiffusionXL],
+        ui_model_type=ModelType.IPAdapter,
     )
     clip_vision_model: Literal["ViT-H", "ViT-G", "ViT-L"] = InputField(
         description="CLIP Vision model to use. Overrides model settings. Mandatory for checkpoint models.",

invokeai/app/invocations/latents_to_image.py

@@ -27,6 +27,7 @@ from invokeai.app.services.shared.invocation_context import InvocationContext
 from invokeai.backend.stable_diffusion.extensions.seamless import SeamlessExt
 from invokeai.backend.stable_diffusion.vae_tiling import patch_vae_tiling_params
 from invokeai.backend.util.devices import TorchDevice
+from invokeai.backend.util.vae_working_memory import estimate_vae_working_memory_sd15_sdxl
 
 
 @invocation(
@@ -53,39 +54,6 @@ class LatentsToImageInvocation(BaseInvocation, WithMetadata, WithBoard):
     tile_size: int = InputField(default=0, multiple_of=8, description=FieldDescriptions.vae_tile_size)
     fp32: bool = InputField(default=False, description=FieldDescriptions.fp32)
 
-    def _estimate_working_memory(
-        self, latents: torch.Tensor, use_tiling: bool, vae: AutoencoderKL | AutoencoderTiny
-    ) -> int:
-        """Estimate the working memory required by the invocation in bytes."""
-        # It was found experimentally that the peak working memory scales linearly with the number of pixels and the
-        # element size (precision). This estimate is accurate for both SD1 and SDXL.
-        element_size = 4 if self.fp32 else 2
-        scaling_constant = 2200  # Determined experimentally.
-
-        if use_tiling:
-            tile_size = self.tile_size
-            if tile_size == 0:
-                tile_size = vae.tile_sample_min_size
-                assert isinstance(tile_size, int)
-            out_h = tile_size
-            out_w = tile_size
-            working_memory = out_h * out_w * element_size * scaling_constant
-
-            # We add 25% to the working memory estimate when tiling is enabled to account for factors like tile overlap
-            # and number of tiles. We could make this more precise in the future, but this should be good enough for
-            # most use cases.
-            working_memory = working_memory * 1.25
-        else:
-            out_h = LATENT_SCALE_FACTOR * latents.shape[-2]
-            out_w = LATENT_SCALE_FACTOR * latents.shape[-1]
-            working_memory = out_h * out_w * element_size * scaling_constant
-
-        if self.fp32:
-            # If we are running in FP32, then we should account for the likely increase in model size (~250MB).
-            working_memory += 250 * 2**20
-
-        return int(working_memory)
-
     @torch.no_grad()
     def invoke(self, context: InvocationContext) -> ImageOutput:
         latents = context.tensors.load(self.latents.latents_name)
@@ -94,8 +62,13 @@ class LatentsToImageInvocation(BaseInvocation, WithMetadata, WithBoard):
 
         vae_info = context.models.load(self.vae.vae)
         assert isinstance(vae_info.model, (AutoencoderKL, AutoencoderTiny))
-
-        estimated_working_memory = self._estimate_working_memory(latents, use_tiling, vae_info.model)
+        estimated_working_memory = estimate_vae_working_memory_sd15_sdxl(
+            operation="decode",
+            image_tensor=latents,
+            vae=vae_info.model,
+            tile_size=self.tile_size if use_tiling else None,
+            fp32=self.fp32,
+        )
         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),

invokeai/app/invocations/llava_onevision_vllm.py

@@ -6,11 +6,12 @@ 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.fields import FieldDescriptions, ImageField, InputField, UIComponent
 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.model_manager.taxonomy import ModelType
 from invokeai.backend.util.devices import TorchDevice
 
 
@@ -34,7 +35,7 @@ class LlavaOnevisionVllmInvocation(BaseInvocation):
     vllm_model: ModelIdentifierField = InputField(
         title="LLaVA Model Type",
         description=FieldDescriptions.vllm_model,
-        ui_type=UIType.LlavaOnevisionModel,
+        ui_model_type=ModelType.LlavaOnevision,
     )
 
     @field_validator("images", mode="before")

invokeai/app/invocations/metadata_linked.py

@@ -53,7 +53,7 @@ from invokeai.app.invocations.primitives import (
 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.taxonomy import ModelType, SubModelType
+from invokeai.backend.model_manager.taxonomy import BaseModelType, ModelType, SubModelType
 from invokeai.backend.stable_diffusion.schedulers.schedulers import SCHEDULER_NAME_VALUES
 from invokeai.version import __version__
 
@@ -473,7 +473,6 @@ class MetadataToModelOutput(BaseInvocationOutput):
     model: ModelIdentifierField = OutputField(
         description=FieldDescriptions.main_model,
         title="Model",
-        ui_type=UIType.MainModel,
     )
     name: str = OutputField(description="Model Name", title="Name")
     unet: UNetField = OutputField(description=FieldDescriptions.unet, title="UNet")
@@ -488,7 +487,6 @@ class MetadataToSDXLModelOutput(BaseInvocationOutput):
     model: ModelIdentifierField = OutputField(
         description=FieldDescriptions.main_model,
         title="Model",
-        ui_type=UIType.SDXLMainModel,
     )
     name: str = OutputField(description="Model Name", title="Name")
     unet: UNetField = OutputField(description=FieldDescriptions.unet, title="UNet")
@@ -519,8 +517,7 @@ class MetadataToModelInvocation(BaseInvocation, WithMetadata):
         input=Input.Direct,
     )
     default_value: ModelIdentifierField = InputField(
-        description="The default model to use if not found in the metadata",
-        ui_type=UIType.MainModel,
+        description="The default model to use if not found in the metadata", ui_model_type=ModelType.Main
     )
 
     _validate_custom_label = model_validator(mode="after")(validate_custom_label)
@@ -575,7 +572,8 @@ class MetadataToSDXLModelInvocation(BaseInvocation, WithMetadata):
     )
     default_value: ModelIdentifierField = InputField(
         description="The default SDXL Model to use if not found in the metadata",
-        ui_type=UIType.SDXLMainModel,
+        ui_model_type=ModelType.Main,
+        ui_model_base=BaseModelType.StableDiffusionXL,
     )
 
     _validate_custom_label = model_validator(mode="after")(validate_custom_label)

invokeai/app/invocations/model.py

@@ -9,7 +9,7 @@ from invokeai.app.invocations.baseinvocation import (
     invocation,
     invocation_output,
 )
-from invokeai.app.invocations.fields import FieldDescriptions, ImageField, Input, InputField, OutputField, UIType
+from invokeai.app.invocations.fields import FieldDescriptions, ImageField, Input, InputField, OutputField
 from invokeai.app.services.shared.invocation_context import InvocationContext
 from invokeai.app.shared.models import FreeUConfig
 from invokeai.backend.model_manager.config import (
@@ -145,7 +145,7 @@ class ModelIdentifierInvocation(BaseInvocation):
 
 @invocation(
     "main_model_loader",
-    title="Main Model - SD1.5",
+    title="Main Model - SD1.5, SD2",
     tags=["model"],
     category="model",
     version="1.0.4",
@@ -153,7 +153,11 @@ class ModelIdentifierInvocation(BaseInvocation):
 class MainModelLoaderInvocation(BaseInvocation):
     """Loads a main model, outputting its submodels."""
 
-    model: ModelIdentifierField = InputField(description=FieldDescriptions.main_model, ui_type=UIType.MainModel)
+    model: ModelIdentifierField = InputField(
+        description=FieldDescriptions.main_model,
+        ui_model_base=[BaseModelType.StableDiffusion1, BaseModelType.StableDiffusion2],
+        ui_model_type=ModelType.Main,
+    )
     # TODO: precision?
 
     def invoke(self, context: InvocationContext) -> ModelLoaderOutput:
@@ -187,7 +191,10 @@ class LoRALoaderInvocation(BaseInvocation):
     """Apply selected lora to unet and text_encoder."""
 
     lora: ModelIdentifierField = InputField(
-        description=FieldDescriptions.lora_model, title="LoRA", ui_type=UIType.LoRAModel
+        description=FieldDescriptions.lora_model,
+        title="LoRA",
+        ui_model_base=BaseModelType.StableDiffusion1,
+        ui_model_type=ModelType.LoRA,
     )
     weight: float = InputField(default=0.75, description=FieldDescriptions.lora_weight)
     unet: Optional[UNetField] = InputField(
@@ -250,7 +257,9 @@ class LoRASelectorInvocation(BaseInvocation):
     """Selects a LoRA model and weight."""
 
     lora: ModelIdentifierField = InputField(
-        description=FieldDescriptions.lora_model, title="LoRA", ui_type=UIType.LoRAModel
+        description=FieldDescriptions.lora_model,
+        title="LoRA",
+        ui_model_type=ModelType.LoRA,
     )
     weight: float = InputField(default=0.75, description=FieldDescriptions.lora_weight)
 
@@ -332,7 +341,10 @@ class SDXLLoRALoaderInvocation(BaseInvocation):
     """Apply selected lora to unet and text_encoder."""
 
     lora: ModelIdentifierField = InputField(
-        description=FieldDescriptions.lora_model, title="LoRA", ui_type=UIType.LoRAModel
+        description=FieldDescriptions.lora_model,
+        title="LoRA",
+        ui_model_base=BaseModelType.StableDiffusionXL,
+        ui_model_type=ModelType.LoRA,
     )
     weight: float = InputField(default=0.75, description=FieldDescriptions.lora_weight)
     unet: Optional[UNetField] = InputField(
@@ -473,13 +485,26 @@ class SDXLLoRACollectionLoader(BaseInvocation):
 
 
 @invocation(
-    "vae_loader", title="VAE Model - SD1.5, SDXL, SD3, FLUX", tags=["vae", "model"], category="model", version="1.0.4"
+    "vae_loader",
+    title="VAE Model - SD1.5, SD2, SDXL, SD3, FLUX",
+    tags=["vae", "model"],
+    category="model",
+    version="1.0.4",
 )
 class VAELoaderInvocation(BaseInvocation):
     """Loads a VAE model, outputting a VaeLoaderOutput"""
 
     vae_model: ModelIdentifierField = InputField(
-        description=FieldDescriptions.vae_model, title="VAE", ui_type=UIType.VAEModel
+        description=FieldDescriptions.vae_model,
+        title="VAE",
+        ui_model_base=[
+            BaseModelType.StableDiffusion1,
+            BaseModelType.StableDiffusion2,
+            BaseModelType.StableDiffusionXL,
+            BaseModelType.StableDiffusion3,
+            BaseModelType.Flux,
+        ],
+        ui_model_type=ModelType.VAE,
     )
 
     def invoke(self, context: InvocationContext) -> VAEOutput:

invokeai/app/invocations/primitives.py

@@ -27,6 +27,7 @@ from invokeai.app.invocations.fields import (
     SD3ConditioningField,
     TensorField,
     UIComponent,
+    VideoField,
 )
 from invokeai.app.services.images.images_common import ImageDTO
 from invokeai.app.services.shared.invocation_context import InvocationContext
@@ -287,6 +288,30 @@ class ImageCollectionInvocation(BaseInvocation):
         return ImageCollectionOutput(collection=self.collection)
 
 
+# endregion
+
+# region Video
+
+
+@invocation_output("video_output")
+class VideoOutput(BaseInvocationOutput):
+    """Base class for nodes that output a video"""
+
+    video: VideoField = OutputField(description="The output video")
+    width: int = OutputField(description="The width of the video in pixels")
+    height: int = OutputField(description="The height of the video in pixels")
+    duration_seconds: float = OutputField(description="The duration of the video in seconds")
+
+    @classmethod
+    def build(cls, video_id: str, width: int, height: int, duration_seconds: float) -> "VideoOutput":
+        return cls(
+            video=VideoField(video_id=video_id),
+            width=width,
+            height=height,
+            duration_seconds=duration_seconds,
+        )
+
+
 # endregion
 
 # region DenoiseMask
@@ -430,6 +455,15 @@ class FluxConditioningOutput(BaseInvocationOutput):
         return cls(conditioning=FluxConditioningField(conditioning_name=conditioning_name))
 
 
+@invocation_output("flux_conditioning_collection_output")
+class FluxConditioningCollectionOutput(BaseInvocationOutput):
+    """Base class for nodes that output a collection of conditioning tensors"""
+
+    collection: list[FluxConditioningField] = OutputField(
+        description="The output conditioning tensors",
+    )
+
+
 @invocation_output("sd3_conditioning_output")
 class SD3ConditioningOutput(BaseInvocationOutput):
     """Base class for nodes that output a single SD3 conditioning tensor"""

invokeai/app/invocations/sd3_image_to_latents.py

@@ -17,6 +17,7 @@ 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
+from invokeai.backend.util.vae_working_memory import estimate_vae_working_memory_sd3
 
 
 @invocation(
@@ -34,7 +35,11 @@ class SD3ImageToLatentsInvocation(BaseInvocation, WithMetadata, WithBoard):
 
     @staticmethod
     def vae_encode(vae_info: LoadedModel, image_tensor: torch.Tensor) -> torch.Tensor:
-        with vae_info as vae:
+        assert isinstance(vae_info.model, AutoencoderKL)
+        estimated_working_memory = estimate_vae_working_memory_sd3(
+            operation="encode", image_tensor=image_tensor, vae=vae_info.model
+        )
+        with vae_info.model_on_device(working_mem_bytes=estimated_working_memory) as (_, vae):
             assert isinstance(vae, AutoencoderKL)
 
             vae.disable_tiling()
@@ -58,6 +63,8 @@ class SD3ImageToLatentsInvocation(BaseInvocation, WithMetadata, WithBoard):
             image_tensor = einops.rearrange(image_tensor, "c h w -> 1 c h w")
 
         vae_info = context.models.load(self.vae.vae)
+        assert isinstance(vae_info.model, AutoencoderKL)
+
         latents = self.vae_encode(vae_info=vae_info, image_tensor=image_tensor)
 
         latents = latents.to("cpu")

invokeai/app/invocations/sd3_latents_to_image.py

@@ -6,7 +6,6 @@ from einops import rearrange
 from PIL import Image
 
 from invokeai.app.invocations.baseinvocation import BaseInvocation, invocation
-from invokeai.app.invocations.constants import LATENT_SCALE_FACTOR
 from invokeai.app.invocations.fields import (
     FieldDescriptions,
     Input,
@@ -20,6 +19,7 @@ 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
+from invokeai.backend.util.vae_working_memory import estimate_vae_working_memory_sd3
 
 
 @invocation(
@@ -41,22 +41,15 @@ class SD3LatentsToImageInvocation(BaseInvocation, WithMetadata, WithBoard):
         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)
+        estimated_working_memory = estimate_vae_working_memory_sd3(
+            operation="decode", image_tensor=latents, vae=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),

invokeai/app/invocations/sd3_model_loader.py

@@ -6,14 +6,14 @@ from invokeai.app.invocations.baseinvocation import (
     invocation,
     invocation_output,
 )
-from invokeai.app.invocations.fields import FieldDescriptions, Input, InputField, OutputField, UIType
+from invokeai.app.invocations.fields import FieldDescriptions, Input, InputField, OutputField
 from invokeai.app.invocations.model import CLIPField, ModelIdentifierField, T5EncoderField, TransformerField, VAEField
 from invokeai.app.services.shared.invocation_context import InvocationContext
 from invokeai.app.util.t5_model_identifier import (
     preprocess_t5_encoder_model_identifier,
     preprocess_t5_tokenizer_model_identifier,
 )
-from invokeai.backend.model_manager.taxonomy import SubModelType
+from invokeai.backend.model_manager.taxonomy import BaseModelType, ClipVariantType, ModelType, SubModelType
 
 
 @invocation_output("sd3_model_loader_output")
@@ -39,36 +39,43 @@ class Sd3ModelLoaderInvocation(BaseInvocation):
 
     model: ModelIdentifierField = InputField(
         description=FieldDescriptions.sd3_model,
-        ui_type=UIType.SD3MainModel,
         input=Input.Direct,
+        ui_model_base=BaseModelType.StableDiffusion3,
+        ui_model_type=ModelType.Main,
     )
 
     t5_encoder_model: Optional[ModelIdentifierField] = InputField(
         description=FieldDescriptions.t5_encoder,
-        ui_type=UIType.T5EncoderModel,
         input=Input.Direct,
         title="T5 Encoder",
         default=None,
+        ui_model_type=ModelType.T5Encoder,
     )
 
     clip_l_model: Optional[ModelIdentifierField] = InputField(
         description=FieldDescriptions.clip_embed_model,
-        ui_type=UIType.CLIPLEmbedModel,
         input=Input.Direct,
         title="CLIP L Encoder",
         default=None,
+        ui_model_type=ModelType.CLIPEmbed,
+        ui_model_variant=ClipVariantType.L,
     )
 
     clip_g_model: Optional[ModelIdentifierField] = InputField(
         description=FieldDescriptions.clip_g_model,
-        ui_type=UIType.CLIPGEmbedModel,
         input=Input.Direct,
         title="CLIP G Encoder",
         default=None,
+        ui_model_type=ModelType.CLIPEmbed,
+        ui_model_variant=ClipVariantType.G,
     )
 
     vae_model: Optional[ModelIdentifierField] = InputField(
-        description=FieldDescriptions.vae_model, ui_type=UIType.VAEModel, title="VAE", default=None
+        description=FieldDescriptions.vae_model,
+        title="VAE",
+        default=None,
+        ui_model_base=BaseModelType.StableDiffusion3,
+        ui_model_type=ModelType.VAE,
     )
 
     def invoke(self, context: InvocationContext) -> Sd3ModelLoaderOutput:

invokeai/app/invocations/sdxl.py

@@ -1,8 +1,8 @@
 from invokeai.app.invocations.baseinvocation import BaseInvocation, BaseInvocationOutput, invocation, invocation_output
-from invokeai.app.invocations.fields import FieldDescriptions, InputField, OutputField, UIType
+from invokeai.app.invocations.fields import FieldDescriptions, InputField, OutputField
 from invokeai.app.invocations.model import CLIPField, ModelIdentifierField, UNetField, VAEField
 from invokeai.app.services.shared.invocation_context import InvocationContext
-from invokeai.backend.model_manager.taxonomy import SubModelType
+from invokeai.backend.model_manager.taxonomy import BaseModelType, ModelType, SubModelType
 
 
 @invocation_output("sdxl_model_loader_output")
@@ -29,7 +29,9 @@ class SDXLModelLoaderInvocation(BaseInvocation):
     """Loads an sdxl base model, outputting its submodels."""
 
     model: ModelIdentifierField = InputField(
-        description=FieldDescriptions.sdxl_main_model, ui_type=UIType.SDXLMainModel
+        description=FieldDescriptions.sdxl_main_model,
+        ui_model_base=BaseModelType.StableDiffusionXL,
+        ui_model_type=ModelType.Main,
     )
     # TODO: precision?
 
@@ -67,7 +69,9 @@ class SDXLRefinerModelLoaderInvocation(BaseInvocation):
     """Loads an sdxl refiner model, outputting its submodels."""
 
     model: ModelIdentifierField = InputField(
-        description=FieldDescriptions.sdxl_refiner_model, ui_type=UIType.SDXLRefinerModel
+        description=FieldDescriptions.sdxl_refiner_model,
+        ui_model_base=BaseModelType.StableDiffusionXLRefiner,
+        ui_model_type=ModelType.Main,
     )
     # TODO: precision?
 

invokeai/app/invocations/segment_anything.py

@@ -1,72 +1,75 @@
-from enum import Enum
+from itertools import zip_longest
 from pathlib import Path
 from typing import Literal
 
 import numpy as np
 import torch
 from PIL import Image
-from pydantic import BaseModel, Field
-from transformers import AutoProcessor
+from pydantic import BaseModel, Field, model_validator
 from transformers.models.sam import SamModel
 from transformers.models.sam.processing_sam import SamProcessor
+from transformers.models.sam2 import Sam2Model
+from transformers.models.sam2.processing_sam2 import Sam2Processor
 
 from invokeai.app.invocations.baseinvocation import BaseInvocation, invocation
 from invokeai.app.invocations.fields import BoundingBoxField, ImageField, InputField, TensorField
 from invokeai.app.invocations.primitives import MaskOutput
 from invokeai.app.services.shared.invocation_context import InvocationContext
 from invokeai.backend.image_util.segment_anything.mask_refinement import mask_to_polygon, polygon_to_mask
+from invokeai.backend.image_util.segment_anything.segment_anything_2_pipeline import SegmentAnything2Pipeline
 from invokeai.backend.image_util.segment_anything.segment_anything_pipeline import SegmentAnythingPipeline
+from invokeai.backend.image_util.segment_anything.shared import SAMInput, SAMPoint
 
-SegmentAnythingModelKey = Literal["segment-anything-base", "segment-anything-large", "segment-anything-huge"]
+SegmentAnythingModelKey = Literal[
+    "segment-anything-base",
+    "segment-anything-large",
+    "segment-anything-huge",
+    "segment-anything-2-tiny",
+    "segment-anything-2-small",
+    "segment-anything-2-base",
+    "segment-anything-2-large",
+]
 SEGMENT_ANYTHING_MODEL_IDS: dict[SegmentAnythingModelKey, str] = {
     "segment-anything-base": "facebook/sam-vit-base",
     "segment-anything-large": "facebook/sam-vit-large",
     "segment-anything-huge": "facebook/sam-vit-huge",
+    "segment-anything-2-tiny": "facebook/sam2.1-hiera-tiny",
+    "segment-anything-2-small": "facebook/sam2.1-hiera-small",
+    "segment-anything-2-base": "facebook/sam2.1-hiera-base-plus",
+    "segment-anything-2-large": "facebook/sam2.1-hiera-large",
 }
 
 
-class SAMPointLabel(Enum):
-    negative = -1
-    neutral = 0
-    positive = 1
-
-
-class SAMPoint(BaseModel):
-    x: int = Field(..., description="The x-coordinate of the point")
-    y: int = Field(..., description="The y-coordinate of the point")
-    label: SAMPointLabel = Field(..., description="The label of the point")
-
-
 class SAMPointsField(BaseModel):
-    points: list[SAMPoint] = Field(..., description="The points of the object")
+    points: list[SAMPoint] = Field(..., description="The points of the object", min_length=1)
 
-    def to_list(self) -> list[list[int]]:
+    def to_list(self) -> list[list[float]]:
         return [[point.x, point.y, point.label.value] for point in self.points]
 
 
 @invocation(
     "segment_anything",
     title="Segment Anything",
-    tags=["prompt", "segmentation"],
+    tags=["prompt", "segmentation", "sam", "sam2"],
     category="segmentation",
-    version="1.2.0",
+    version="1.3.0",
 )
 class SegmentAnythingInvocation(BaseInvocation):
-    """Runs a Segment Anything Model."""
+    """Runs a Segment Anything Model (SAM or SAM2)."""
 
     # Reference:
     # - https://arxiv.org/pdf/2304.02643
     # - https://huggingface.co/docs/transformers/v4.43.3/en/model_doc/grounding-dino#grounded-sam
     # - https://github.com/NielsRogge/Transformers-Tutorials/blob/a39f33ac1557b02ebfb191ea7753e332b5ca933f/Grounding%20DINO/GroundingDINO_with_Segment_Anything.ipynb
 
-    model: SegmentAnythingModelKey = InputField(description="The Segment Anything model to use.")
+    model: SegmentAnythingModelKey = InputField(description="The Segment Anything model to use (SAM or SAM2).")
     image: ImageField = InputField(description="The image to segment.")
     bounding_boxes: list[BoundingBoxField] | None = InputField(
-        default=None, description="The bounding boxes to prompt the SAM model with."
+        default=None, description="The bounding boxes to prompt the model with."
     )
     point_lists: list[SAMPointsField] | None = InputField(
         default=None,
-        description="The list of point lists to prompt the SAM model with. Each list of points represents a single object.",
+        description="The list of point lists to prompt the model with. Each list of points represents a single object.",
     )
     apply_polygon_refinement: bool = InputField(
         description="Whether to apply polygon refinement to the masks. This will smooth the edges of the masks slightly and ensure that each mask consists of a single closed polygon (before merging).",
@@ -77,14 +80,18 @@ class SegmentAnythingInvocation(BaseInvocation):
         default="all",
     )
 
+    @model_validator(mode="after")
+    def validate_points_and_boxes_len(self):
+        if self.point_lists is not None and self.bounding_boxes is not None:
+            if len(self.point_lists) != len(self.bounding_boxes):
+                raise ValueError("If both point_lists and bounding_boxes are provided, they must have the same length.")
+        return self
+
     @torch.no_grad()
     def invoke(self, context: InvocationContext) -> MaskOutput:
         # The models expect a 3-channel RGB image.
         image_pil = context.images.get_pil(self.image.image_name, mode="RGB")
 
-        if self.point_lists is not None and self.bounding_boxes is not None:
-            raise ValueError("Only one of point_lists or bounding_box can be provided.")
-
         if (not self.bounding_boxes or len(self.bounding_boxes) == 0) and (
             not self.point_lists or len(self.point_lists) == 0
         ):
@@ -111,26 +118,38 @@ class SegmentAnythingInvocation(BaseInvocation):
             # model, and figure out how to make it work in the pipeline.
             # torch_dtype=TorchDevice.choose_torch_dtype(),
         )
-
-        sam_processor = AutoProcessor.from_pretrained(model_path, local_files_only=True)
-        assert isinstance(sam_processor, SamProcessor)
+        sam_processor = SamProcessor.from_pretrained(model_path, local_files_only=True)
         return SegmentAnythingPipeline(sam_model=sam_model, sam_processor=sam_processor)
 
-    def _segment(self, context: InvocationContext, image: Image.Image) -> list[torch.Tensor]:
-        """Use Segment Anything (SAM) to generate masks given an image + a set of bounding boxes."""
-        # Convert the bounding boxes to the SAM input format.
-        sam_bounding_boxes = (
-            [[bb.x_min, bb.y_min, bb.x_max, bb.y_max] for bb in self.bounding_boxes] if self.bounding_boxes else None
-        )
-        sam_points = [p.to_list() for p in self.point_lists] if self.point_lists else None
+    @staticmethod
+    def _load_sam_2_model(model_path: Path):
+        sam2_model = Sam2Model.from_pretrained(model_path, local_files_only=True)
+        sam2_processor = Sam2Processor.from_pretrained(model_path, local_files_only=True)
+        return SegmentAnything2Pipeline(sam2_model=sam2_model, sam2_processor=sam2_processor)
 
-        with (
-            context.models.load_remote_model(
-                source=SEGMENT_ANYTHING_MODEL_IDS[self.model], loader=SegmentAnythingInvocation._load_sam_model
-            ) as sam_pipeline,
-        ):
-            assert isinstance(sam_pipeline, SegmentAnythingPipeline)
-            masks = sam_pipeline.segment(image=image, bounding_boxes=sam_bounding_boxes, point_lists=sam_points)
+    def _segment(self, context: InvocationContext, image: Image.Image) -> list[torch.Tensor]:
+        """Use Segment Anything (SAM or SAM2) to generate masks given an image + a set of bounding boxes."""
+
+        source = SEGMENT_ANYTHING_MODEL_IDS[self.model]
+        inputs: list[SAMInput] = []
+        for bbox_field, point_field in zip_longest(self.bounding_boxes or [], self.point_lists or [], fillvalue=None):
+            inputs.append(
+                SAMInput(
+                    bounding_box=bbox_field,
+                    points=point_field.points if point_field else None,
+                )
+            )
+
+        if "sam2" in source:
+            loader = SegmentAnythingInvocation._load_sam_2_model
+            with context.models.load_remote_model(source=source, loader=loader) as pipeline:
+                assert isinstance(pipeline, SegmentAnything2Pipeline)
+                masks = pipeline.segment(image=image, inputs=inputs)
+        else:
+            loader = SegmentAnythingInvocation._load_sam_model
+            with context.models.load_remote_model(source=source, loader=loader) as pipeline:
+                assert isinstance(pipeline, SegmentAnythingPipeline)
+                masks = pipeline.segment(image=image, inputs=inputs)
 
         masks = self._process_masks(masks)
         if self.apply_polygon_refinement:

invokeai/app/invocations/spandrel_image_to_image.py

@@ -11,7 +11,6 @@ from invokeai.app.invocations.fields import (
     FieldDescriptions,
     ImageField,
     InputField,
-    UIType,
     WithBoard,
     WithMetadata,
 )
@@ -19,6 +18,7 @@ from invokeai.app.invocations.model import ModelIdentifierField
 from invokeai.app.invocations.primitives import ImageOutput
 from invokeai.app.services.session_processor.session_processor_common import CanceledException
 from invokeai.app.services.shared.invocation_context import InvocationContext
+from invokeai.backend.model_manager.taxonomy import ModelType
 from invokeai.backend.spandrel_image_to_image_model import SpandrelImageToImageModel
 from invokeai.backend.tiles.tiles import calc_tiles_min_overlap
 from invokeai.backend.tiles.utils import TBLR, Tile
@@ -33,7 +33,7 @@ class SpandrelImageToImageInvocation(BaseInvocation, WithMetadata, WithBoard):
     image_to_image_model: ModelIdentifierField = InputField(
         title="Image-to-Image Model",
         description=FieldDescriptions.spandrel_image_to_image_model,
-        ui_type=UIType.SpandrelImageToImageModel,
+        ui_model_type=ModelType.SpandrelImageToImage,
     )
     tile_size: int = InputField(
         default=512, description="The tile size for tiled image-to-image. Set to 0 to disable tiling."

invokeai/app/invocations/t2i_adapter.py

@@ -8,11 +8,12 @@ from invokeai.app.invocations.baseinvocation import (
     invocation,
     invocation_output,
 )
-from invokeai.app.invocations.fields import FieldDescriptions, ImageField, InputField, OutputField, UIType
+from invokeai.app.invocations.fields import FieldDescriptions, ImageField, InputField, OutputField
 from invokeai.app.invocations.model import ModelIdentifierField
 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_RESIZE_VALUES
+from invokeai.backend.model_manager.taxonomy import BaseModelType, ModelType
 
 
 class T2IAdapterField(BaseModel):
@@ -60,7 +61,8 @@ class T2IAdapterInvocation(BaseInvocation):
         description="The T2I-Adapter model.",
         title="T2I-Adapter Model",
         ui_order=-1,
-        ui_type=UIType.T2IAdapterModel,
+        ui_model_base=[BaseModelType.StableDiffusion1, BaseModelType.StableDiffusionXL],
+        ui_model_type=ModelType.T2IAdapter,
     )
     weight: Union[float, list[float]] = InputField(
         default=1, ge=0, description="The weight given to the T2I-Adapter", title="Weight"

invokeai/app/services/board_image_records/board_image_records_base.py

@@ -49,3 +49,11 @@ class BoardImageRecordStorageBase(ABC):
     ) -> int:
         """Gets the number of images for a board."""
         pass
+
+    @abstractmethod
+    def get_asset_count_for_board(
+        self,
+        board_id: str,
+    ) -> int:
+        """Gets the number of assets for a board."""
+        pass

invokeai/app/services/board_image_records/board_image_records_sqlite.py

@@ -3,6 +3,8 @@ from typing import Optional, cast
 
 from invokeai.app.services.board_image_records.board_image_records_base import BoardImageRecordStorageBase
 from invokeai.app.services.image_records.image_records_common import (
+    ASSETS_CATEGORIES,
+    IMAGE_CATEGORIES,
     ImageCategory,
     ImageRecord,
     deserialize_image_record,
@@ -151,15 +153,38 @@ class SqliteBoardImageRecordStorage(BoardImageRecordStorageBase):
 
     def get_image_count_for_board(self, board_id: str) -> int:
         with self._db.transaction() as cursor:
+            # Convert the enum values to unique list of strings
+            category_strings = [c.value for c in set(IMAGE_CATEGORIES)]
+            # Create the correct length of placeholders
+            placeholders = ",".join("?" * len(category_strings))
             cursor.execute(
-                """--sql
+                f"""--sql
                     SELECT COUNT(*)
                     FROM board_images
                     INNER JOIN images ON board_images.image_name = images.image_name
-                    WHERE images.is_intermediate = FALSE
+                    WHERE images.is_intermediate = FALSE AND images.image_category IN ( {placeholders} )
                     AND board_images.board_id = ?;
                     """,
-                (board_id,),
+                (*category_strings, board_id),
+            )
+            count = cast(int, cursor.fetchone()[0])
+        return count
+
+    def get_asset_count_for_board(self, board_id: str) -> int:
+        with self._db.transaction() as cursor:
+            # Convert the enum values to unique list of strings
+            category_strings = [c.value for c in set(ASSETS_CATEGORIES)]
+            # Create the correct length of placeholders
+            placeholders = ",".join("?" * len(category_strings))
+            cursor.execute(
+                f"""--sql
+                    SELECT COUNT(*)
+                    FROM board_images
+                    INNER JOIN images ON board_images.image_name = images.image_name
+                    WHERE images.is_intermediate = FALSE AND images.image_category IN ( {placeholders} )
+                    AND board_images.board_id = ?;
+                    """,
+                (*category_strings, board_id),
             )
             count = cast(int, cursor.fetchone()[0])
         return count

invokeai/app/services/boards/boards_common.py

@@ -12,12 +12,20 @@ class BoardDTO(BoardRecord):
     """The URL of the thumbnail of the most recent image in the board."""
     image_count: int = Field(description="The number of images in the board.")
     """The number of images in the board."""
+    asset_count: int = Field(description="The number of assets in the board.")
+    """The number of assets in the board."""
+    video_count: int = Field(description="The number of videos in the board.")
+    """The number of videos in the board."""
 
 
-def board_record_to_dto(board_record: BoardRecord, cover_image_name: Optional[str], image_count: int) -> BoardDTO:
+def board_record_to_dto(
+    board_record: BoardRecord, cover_image_name: Optional[str], image_count: int, asset_count: int, video_count: int
+) -> BoardDTO:
     """Converts a board record to a board DTO."""
     return BoardDTO(
         **board_record.model_dump(exclude={"cover_image_name"}),
         cover_image_name=cover_image_name,
         image_count=image_count,
+        asset_count=asset_count,
+        video_count=video_count,
     )

invokeai/app/services/boards/boards_default.py

@@ -17,7 +17,7 @@ class BoardService(BoardServiceABC):
         board_name: str,
     ) -> BoardDTO:
         board_record = self.__invoker.services.board_records.save(board_name)
-        return board_record_to_dto(board_record, None, 0)
+        return board_record_to_dto(board_record, None, 0, 0, 0)
 
     def get_dto(self, board_id: str) -> BoardDTO:
         board_record = self.__invoker.services.board_records.get(board_id)
@@ -27,7 +27,9 @@ class BoardService(BoardServiceABC):
         else:
             cover_image_name = None
         image_count = self.__invoker.services.board_image_records.get_image_count_for_board(board_id)
-        return board_record_to_dto(board_record, cover_image_name, image_count)
+        asset_count = self.__invoker.services.board_image_records.get_asset_count_for_board(board_id)
+        video_count = 0  # noop for OSS
+        return board_record_to_dto(board_record, cover_image_name, image_count, asset_count, video_count)
 
     def update(
         self,
@@ -42,7 +44,9 @@ class BoardService(BoardServiceABC):
             cover_image_name = None
 
         image_count = self.__invoker.services.board_image_records.get_image_count_for_board(board_id)
-        return board_record_to_dto(board_record, cover_image_name, image_count)
+        asset_count = self.__invoker.services.board_image_records.get_asset_count_for_board(board_id)
+        video_count = 0  # noop for OSS
+        return board_record_to_dto(board_record, cover_image_name, image_count, asset_count, video_count)
 
     def delete(self, board_id: str) -> None:
         self.__invoker.services.board_records.delete(board_id)
@@ -67,7 +71,9 @@ class BoardService(BoardServiceABC):
                 cover_image_name = None
 
             image_count = self.__invoker.services.board_image_records.get_image_count_for_board(r.board_id)
-            board_dtos.append(board_record_to_dto(r, cover_image_name, image_count))
+            asset_count = self.__invoker.services.board_image_records.get_asset_count_for_board(r.board_id)
+            video_count = 0  # noop for OSS
+            board_dtos.append(board_record_to_dto(r, cover_image_name, image_count, asset_count, video_count))
 
         return OffsetPaginatedResults[BoardDTO](items=board_dtos, offset=offset, limit=limit, total=len(board_dtos))
 
@@ -84,6 +90,8 @@ class BoardService(BoardServiceABC):
                 cover_image_name = None
 
             image_count = self.__invoker.services.board_image_records.get_image_count_for_board(r.board_id)
-            board_dtos.append(board_record_to_dto(r, cover_image_name, image_count))
+            asset_count = self.__invoker.services.board_image_records.get_asset_count_for_board(r.board_id)
+            video_count = 0  # noop for OSS
+            board_dtos.append(board_record_to_dto(r, cover_image_name, image_count, asset_count, video_count))
 
         return board_dtos

invokeai/app/services/bulk_download/bulk_download_default.py

@@ -150,4 +150,15 @@ class BulkDownloadService(BulkDownloadBase):
     def _is_valid_path(self, path: Union[str, Path]) -> bool:
         """Validates the path given for a bulk download."""
         path = path if isinstance(path, Path) else Path(path)
-        return path.exists()
+
+        # Resolve the path to handle any path traversal attempts (e.g., ../)
+        resolved_path = path.resolve()
+
+        # The path may not traverse out of the bulk downloads folder or its subfolders
+        does_not_traverse = resolved_path.parent == self._bulk_downloads_folder.resolve()
+
+        # The path must exist and be a .zip file
+        does_exist = resolved_path.exists()
+        is_zip_file = resolved_path.suffix == ".zip"
+
+        return does_exist and is_zip_file and does_not_traverse

invokeai/app/services/client_state_persistence/client_state_persistence_base.py

@@ -0,0 +1,42 @@
+from abc import ABC, abstractmethod
+
+
+class ClientStatePersistenceABC(ABC):
+    """
+    Base class for client persistence implementations.
+    This class defines the interface for persisting client data.
+    """
+
+    @abstractmethod
+    def set_by_key(self, queue_id: str, key: str, value: str) -> str:
+        """
+        Set a key-value pair for the client.
+
+        Args:
+            key (str): The key to set.
+            value (str): The value to set for the key.
+
+        Returns:
+            str: The value that was set.
+        """
+        pass
+
+    @abstractmethod
+    def get_by_key(self, queue_id: str, key: str) -> str | None:
+        """
+        Get the value for a specific key of the client.
+
+        Args:
+            key (str): The key to retrieve the value for.
+
+        Returns:
+            str | None: The value associated with the key, or None if the key does not exist.
+        """
+        pass
+
+    @abstractmethod
+    def delete(self, queue_id: str) -> None:
+        """
+        Delete all client state.
+        """
+        pass

invokeai/app/services/client_state_persistence/client_state_persistence_sqlite.py

@@ -0,0 +1,65 @@
+import json
+
+from invokeai.app.services.client_state_persistence.client_state_persistence_base import ClientStatePersistenceABC
+from invokeai.app.services.invoker import Invoker
+from invokeai.app.services.shared.sqlite.sqlite_database import SqliteDatabase
+
+
+class ClientStatePersistenceSqlite(ClientStatePersistenceABC):
+    """
+    Base class for client persistence implementations.
+    This class defines the interface for persisting client data.
+    """
+
+    def __init__(self, db: SqliteDatabase) -> None:
+        super().__init__()
+        self._db = db
+        self._default_row_id = 1
+
+    def start(self, invoker: Invoker) -> None:
+        self._invoker = invoker
+
+    def _get(self) -> dict[str, str] | None:
+        with self._db.transaction() as cursor:
+            cursor.execute(
+                f"""
+                SELECT data FROM client_state
+                WHERE id = {self._default_row_id}
+                """
+            )
+            row = cursor.fetchone()
+            if row is None:
+                return None
+            return json.loads(row[0])
+
+    def set_by_key(self, queue_id: str, key: str, value: str) -> str:
+        state = self._get() or {}
+        state.update({key: value})
+
+        with self._db.transaction() as cursor:
+            cursor.execute(
+                f"""
+                INSERT INTO client_state (id, data)
+                VALUES ({self._default_row_id}, ?)
+                ON CONFLICT(id) DO UPDATE
+                  SET data = excluded.data;
+                """,
+                (json.dumps(state),),
+            )
+
+        return value
+
+    def get_by_key(self, queue_id: str, key: str) -> str | None:
+        state = self._get()
+        if state is None:
+            return None
+        return state.get(key, None)
+
+    def delete(self, queue_id: str) -> None:
+        with self._db.transaction() as cursor:
+            cursor.execute(
+                f"""
+                DELETE FROM client_state
+                WHERE id = {self._default_row_id}
+                """
+            )

invokeai/app/services/config/config_default.py

@@ -107,6 +107,7 @@ class InvokeAIAppConfig(BaseSettings):
         hashing_algorithm: Model hashing algorthim for model installs. 'blake3_multi' is best for SSDs. 'blake3_single' is best for spinning disk HDDs. 'random' disables hashing, instead assigning a UUID to models. Useful when using a memory db to reduce model installation time, or if you don't care about storing stable hashes for models. Alternatively, any other hashlib algorithm is accepted, though these are not nearly as performant as blake3.<br>Valid values: `blake3_multi`, `blake3_single`, `random`, `md5`, `sha1`, `sha224`, `sha256`, `sha384`, `sha512`, `blake2b`, `blake2s`, `sha3_224`, `sha3_256`, `sha3_384`, `sha3_512`, `shake_128`, `shake_256`
         remote_api_tokens: List of regular expression and token pairs used when downloading models from URLs. The download URL is tested against the regex, and if it matches, the token is provided in as a Bearer token.
         scan_models_on_startup: Scan the models directory on startup, registering orphaned models. This is typically only used in conjunction with `use_memory_db` for testing purposes.
+        unsafe_disable_picklescan: UNSAFE. Disable the picklescan security check during model installation. Recommended only for development and testing purposes. This will allow arbitrary code execution during model installation, so should never be used in production.
     """
 
     _root: Optional[Path] = PrivateAttr(default=None)
@@ -196,6 +197,7 @@ class InvokeAIAppConfig(BaseSettings):
     hashing_algorithm: HASHING_ALGORITHMS = Field(default="blake3_single",  description="Model hashing algorthim for model installs. 'blake3_multi' is best for SSDs. 'blake3_single' is best for spinning disk HDDs. 'random' disables hashing, instead assigning a UUID to models. Useful when using a memory db to reduce model installation time, or if you don't care about storing stable hashes for models. Alternatively, any other hashlib algorithm is accepted, though these are not nearly as performant as blake3.")
     remote_api_tokens: Optional[list[URLRegexTokenPair]] = Field(default=None, description="List of regular expression and token pairs used when downloading models from URLs. The download URL is tested against the regex, and if it matches, the token is provided in as a Bearer token.")
     scan_models_on_startup:        bool = Field(default=False,              description="Scan the models directory on startup, registering orphaned models. This is typically only used in conjunction with `use_memory_db` for testing purposes.")
+    unsafe_disable_picklescan:     bool = Field(default=False,              description="UNSAFE. Disable the picklescan security check during model installation. Recommended only for development and testing purposes. This will allow arbitrary code execution during model installation, so should never be used in production.")
 
     # fmt: on
 

invokeai/app/services/events/events_common.py

@@ -234,8 +234,8 @@ class QueueItemStatusChangedEvent(QueueItemEventBase):
     error_type: Optional[str] = Field(default=None, description="The error type, if any")
     error_message: Optional[str] = Field(default=None, description="The error message, if any")
     error_traceback: Optional[str] = Field(default=None, description="The error traceback, if any")
-    created_at: Optional[str] = Field(default=None, description="The timestamp when the queue item was created")
-    updated_at: Optional[str] = Field(default=None, description="The timestamp when the queue item was last updated")
+    created_at: str = Field(description="The timestamp when the queue item was created")
+    updated_at: str = Field(description="The timestamp when the queue item was last updated")
     started_at: Optional[str] = Field(default=None, description="The timestamp when the queue item was started")
     completed_at: Optional[str] = Field(default=None, description="The timestamp when the queue item was completed")
     batch_status: BatchStatus = Field(description="The status of the batch")
@@ -258,8 +258,8 @@ class QueueItemStatusChangedEvent(QueueItemEventBase):
             error_type=queue_item.error_type,
             error_message=queue_item.error_message,
             error_traceback=queue_item.error_traceback,
-            created_at=str(queue_item.created_at) if queue_item.created_at else None,
-            updated_at=str(queue_item.updated_at) if queue_item.updated_at else None,
+            created_at=str(queue_item.created_at),
+            updated_at=str(queue_item.updated_at),
             started_at=str(queue_item.started_at) if queue_item.started_at else None,
             completed_at=str(queue_item.completed_at) if queue_item.completed_at else None,
             batch_status=batch_status,

invokeai/app/services/image_records/image_records_common.py

@@ -58,6 +58,15 @@ class ImageCategory(str, Enum, metaclass=MetaEnum):
     """OTHER: The image is some other type of image with a specialized purpose. To be used by external nodes."""
 
 
+IMAGE_CATEGORIES: list[ImageCategory] = [ImageCategory.GENERAL]
+ASSETS_CATEGORIES: list[ImageCategory] = [
+    ImageCategory.CONTROL,
+    ImageCategory.MASK,
+    ImageCategory.USER,
+    ImageCategory.OTHER,
+]
+
+
 class InvalidImageCategoryException(ValueError):
     """Raised when a provided value is not a valid ImageCategory.
 

invokeai/app/services/invocation_services.py

@@ -17,6 +17,7 @@ if TYPE_CHECKING:
     from invokeai.app.services.board_records.board_records_base import BoardRecordStorageBase
     from invokeai.app.services.boards.boards_base import BoardServiceABC
     from invokeai.app.services.bulk_download.bulk_download_base import BulkDownloadBase
+    from invokeai.app.services.client_state_persistence.client_state_persistence_base import ClientStatePersistenceABC
     from invokeai.app.services.config import InvokeAIAppConfig
     from invokeai.app.services.download import DownloadQueueServiceBase
     from invokeai.app.services.events.events_base import EventServiceBase
@@ -73,6 +74,7 @@ class InvocationServices:
         style_preset_records: "StylePresetRecordsStorageBase",
         style_preset_image_files: "StylePresetImageFileStorageBase",
         workflow_thumbnails: "WorkflowThumbnailServiceBase",
+        client_state_persistence: "ClientStatePersistenceABC",
     ):
         self.board_images = board_images
         self.board_image_records = board_image_records
@@ -102,3 +104,4 @@ class InvocationServices:
         self.style_preset_records = style_preset_records
         self.style_preset_image_files = style_preset_image_files
         self.workflow_thumbnails = workflow_thumbnails
+        self.client_state_persistence = client_state_persistence

invokeai/app/services/model_install/model_install_default.py

@@ -7,7 +7,7 @@ import threading
 import time
 from pathlib import Path
 from queue import Empty, Queue
-from shutil import copyfile, copytree, move, rmtree
+from shutil import move, rmtree
 from tempfile import mkdtemp
 from typing import TYPE_CHECKING, Any, Dict, List, Optional, Tuple, Type, Union
 
@@ -51,6 +51,7 @@ from invokeai.backend.model_manager.metadata import (
 from invokeai.backend.model_manager.metadata.metadata_base import HuggingFaceMetadata
 from invokeai.backend.model_manager.search import ModelSearch
 from invokeai.backend.model_manager.taxonomy import ModelRepoVariant, ModelSourceType
+from invokeai.backend.model_manager.util.lora_metadata_extractor import apply_lora_metadata
 from invokeai.backend.util import InvokeAILogger
 from invokeai.backend.util.catch_sigint import catch_sigint
 from invokeai.backend.util.devices import TorchDevice
@@ -185,13 +186,15 @@ class ModelInstallService(ModelInstallServiceBase):
         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)
+            if Path(model_path).is_file():
+                # Careful! Don't use pathlib.Path(...).with_suffix - it can will strip everything after the first dot.
+                preferred_name = f"{preferred_name}{model_path.suffix}"
 
         dest_path = (
             self.app_config.models_path / info.base.value / info.type.value / (preferred_name or model_path.name)
         )
         try:
-            new_path = self._copy_model(model_path, dest_path)
+            new_path = self._move_model(model_path, dest_path)
         except FileExistsError as excp:
             raise DuplicateModelException(
                 f"A model named {model_path.name} is already installed at {dest_path.as_posix()}"
@@ -616,30 +619,17 @@ class ModelInstallService(ModelInstallServiceBase):
         self.record_store.update_model(key, ModelRecordChanges(path=model.path))
         return model
 
-    def _copy_model(self, old_path: Path, new_path: Path) -> Path:
-        if old_path == new_path:
-            return old_path
-        new_path.parent.mkdir(parents=True, exist_ok=True)
-        if old_path.is_dir():
-            copytree(old_path, new_path)
-        else:
-            copyfile(old_path, new_path)
-        return new_path
-
     def _move_model(self, old_path: Path, new_path: Path) -> Path:
         if old_path == new_path:
             return old_path
 
+        if new_path.exists():
+            raise FileExistsError(f"Cannot move {old_path} to {new_path}: destination already exists")
+
         new_path.parent.mkdir(parents=True, exist_ok=True)
 
-        # if path already exists then we jigger the name to make it unique
-        counter: int = 1
-        while new_path.exists():
-            path = new_path.with_stem(new_path.stem + f"_{counter:02d}")
-            if not path.exists():
-                new_path = path
-            counter += 1
         move(old_path, new_path)
+
         return new_path
 
     def _probe(self, model_path: Path, config: Optional[ModelRecordChanges] = None):
@@ -667,6 +657,10 @@ class ModelInstallService(ModelInstallServiceBase):
 
         info = info or self._probe(model_path, config)
 
+        # Apply LoRA metadata if applicable
+        model_images_path = self.app_config.models_path / "model_images"
+        apply_lora_metadata(info, model_path.resolve(), model_images_path)
+
         model_path = model_path.resolve()
 
         # Models in the Invoke-managed models dir should use relative paths.

invokeai/app/services/model_load/model_load_default.py

@@ -87,9 +87,21 @@ class ModelLoadService(ModelLoadServiceBase):
         def torch_load_file(checkpoint: Path) -> AnyModel:
             scan_result = scan_file_path(checkpoint)
             if scan_result.infected_files != 0:
-                raise Exception(f"The model at {checkpoint} is potentially infected by malware. Aborting load.")
+                if self._app_config.unsafe_disable_picklescan:
+                    self._logger.warning(
+                        f"Model at {checkpoint} is potentially infected by malware, but picklescan is disabled. "
+                        "Proceeding with caution."
+                    )
+                else:
+                    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.")
+                if self._app_config.unsafe_disable_picklescan:
+                    self._logger.warning(
+                        f"Error scanning model at {checkpoint} for malware, but picklescan is disabled. "
+                        "Proceeding with caution."
+                    )
+                else:
+                    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_records/model_records_base.py

@@ -15,6 +15,7 @@ from invokeai.app.util.model_exclude_null import BaseModelExcludeNull
 from invokeai.backend.model_manager.config import (
     AnyModelConfig,
     ControlAdapterDefaultSettings,
+    LoraModelDefaultSettings,
     MainModelDefaultSettings,
 )
 from invokeai.backend.model_manager.taxonomy import (
@@ -83,8 +84,8 @@ class ModelRecordChanges(BaseModelExcludeNull):
     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(
-        description="Default settings for this model", default=None
+    default_settings: Optional[MainModelDefaultSettings | LoraModelDefaultSettings | ControlAdapterDefaultSettings] = (
+        Field(description="Default settings for this model", default=None)
     )
 
     # Checkpoint-specific changes

invokeai/app/services/session_queue/session_queue_base.py

@@ -15,6 +15,7 @@ from invokeai.app.services.session_queue.session_queue_common import (
     EnqueueBatchResult,
     IsEmptyResult,
     IsFullResult,
+    ItemIdsResult,
     PruneResult,
     RetryItemsResult,
     SessionQueueCountsByDestination,
@@ -23,6 +24,7 @@ from invokeai.app.services.session_queue.session_queue_common import (
 )
 from invokeai.app.services.shared.graph import GraphExecutionState
 from invokeai.app.services.shared.pagination import CursorPaginatedResults
+from invokeai.app.services.shared.sqlite.sqlite_common import SQLiteDirection
 
 
 class SessionQueueBase(ABC):
@@ -145,7 +147,7 @@ class SessionQueueBase(ABC):
         status: Optional[QUEUE_ITEM_STATUS] = None,
         destination: Optional[str] = None,
     ) -> CursorPaginatedResults[SessionQueueItem]:
-        """Gets a page of session queue items"""
+        """Gets a page of session queue items. Do not remove."""
         pass
 
     @abstractmethod
@@ -157,9 +159,18 @@ class SessionQueueBase(ABC):
         """Gets all queue items that match the given parameters"""
         pass
 
+    @abstractmethod
+    def get_queue_item_ids(
+        self,
+        queue_id: str,
+        order_dir: SQLiteDirection = SQLiteDirection.Descending,
+    ) -> ItemIdsResult:
+        """Gets all queue item ids that match the given parameters"""
+        pass
+
     @abstractmethod
     def get_queue_item(self, item_id: int) -> SessionQueueItem:
-        """Gets a session queue item by ID"""
+        """Gets a session queue item by ID for a given queue"""
         pass
 
     @abstractmethod

invokeai/app/services/session_queue/session_queue_common.py

@@ -176,6 +176,14 @@ DEFAULT_QUEUE_ID = "default"
 
 QUEUE_ITEM_STATUS = Literal["pending", "in_progress", "completed", "failed", "canceled"]
 
+
+class ItemIdsResult(BaseModel):
+    """Response containing ordered item ids with metadata for optimistic updates."""
+
+    item_ids: list[int] = Field(description="Ordered list of item ids")
+    total_count: int = Field(description="Total number of queue items matching the query")
+
+
 NodeFieldValueValidator = TypeAdapter(list[NodeFieldValue])
 
 

invokeai/app/services/session_queue/session_queue_sqlite.py

@@ -22,6 +22,7 @@ from invokeai.app.services.session_queue.session_queue_common import (
     EnqueueBatchResult,
     IsEmptyResult,
     IsFullResult,
+    ItemIdsResult,
     PruneResult,
     RetryItemsResult,
     SessionQueueCountsByDestination,
@@ -34,6 +35,7 @@ from invokeai.app.services.session_queue.session_queue_common import (
 )
 from invokeai.app.services.shared.graph import GraphExecutionState
 from invokeai.app.services.shared.pagination import CursorPaginatedResults
+from invokeai.app.services.shared.sqlite.sqlite_common import SQLiteDirection
 from invokeai.app.services.shared.sqlite.sqlite_database import SqliteDatabase
 
 
@@ -671,6 +673,26 @@ class SqliteSessionQueue(SessionQueueBase):
         items = [SessionQueueItem.queue_item_from_dict(dict(result)) for result in results]
         return items
 
+    def get_queue_item_ids(
+        self,
+        queue_id: str,
+        order_dir: SQLiteDirection = SQLiteDirection.Descending,
+    ) -> ItemIdsResult:
+        with self._db.transaction() as cursor_:
+            query = f"""--sql
+                SELECT item_id
+                FROM session_queue
+                WHERE queue_id = ?
+                ORDER BY created_at {order_dir.value}
+                """
+            query_params = [queue_id]
+
+            cursor_.execute(query, query_params)
+            result = cast(list[sqlite3.Row], cursor_.fetchall())
+        item_ids = [row[0] for row in result]
+
+        return ItemIdsResult(item_ids=item_ids, total_count=len(item_ids))
+
     def get_queue_status(self, queue_id: str) -> SessionQueueStatus:
         with self._db.transaction() as cursor:
             cursor.execute(

invokeai/app/services/shared/sqlite/sqlite_util.py

@@ -23,6 +23,7 @@ from invokeai.app.services.shared.sqlite_migrator.migrations.migration_17 import
 from invokeai.app.services.shared.sqlite_migrator.migrations.migration_18 import build_migration_18
 from invokeai.app.services.shared.sqlite_migrator.migrations.migration_19 import build_migration_19
 from invokeai.app.services.shared.sqlite_migrator.migrations.migration_20 import build_migration_20
+from invokeai.app.services.shared.sqlite_migrator.migrations.migration_21 import build_migration_21
 from invokeai.app.services.shared.sqlite_migrator.sqlite_migrator_impl import SqliteMigrator
 
 
@@ -63,6 +64,7 @@ def init_db(config: InvokeAIAppConfig, logger: Logger, image_files: ImageFileSto
     migrator.register_migration(build_migration_18())
     migrator.register_migration(build_migration_19(app_config=config))
     migrator.register_migration(build_migration_20())
+    migrator.register_migration(build_migration_21())
     migrator.run_migrations()
 
     return db

invokeai/app/services/shared/sqlite_migrator/migrations/migration_21.py

@@ -0,0 +1,40 @@
+import sqlite3
+
+from invokeai.app.services.shared.sqlite_migrator.sqlite_migrator_common import Migration
+
+
+class Migration21Callback:
+    def __call__(self, cursor: sqlite3.Cursor) -> None:
+        cursor.execute(
+            """
+            CREATE TABLE client_state (
+              id          INTEGER PRIMARY KEY CHECK(id = 1),
+              data        TEXT    NOT NULL, -- Frontend will handle the shape of this data
+              updated_at  DATETIME NOT NULL DEFAULT (CURRENT_TIMESTAMP)
+            );
+            """
+        )
+        cursor.execute(
+            """
+            CREATE TRIGGER tg_client_state_updated_at
+            AFTER UPDATE ON client_state
+            FOR EACH ROW
+            BEGIN
+              UPDATE client_state
+                SET updated_at = CURRENT_TIMESTAMP
+              WHERE id = OLD.id;
+            END;
+            """
+        )
+
+
+def build_migration_21() -> Migration:
+    """Builds the migration object for migrating from version 20 to version 21. This includes:
+    - Creating the `client_state` table.
+    - Adding a trigger to update the `updated_at` field on updates.
+    """
+    return Migration(
+        from_version=20,
+        to_version=21,
+        callback=Migration21Callback(),
+    )

invokeai/app/services/videos_common.py

@@ -0,0 +1,179 @@
+import datetime
+from typing import Optional, Union
+
+from pydantic import BaseModel, Field, StrictBool, StrictStr
+
+from invokeai.app.util.misc import get_iso_timestamp
+from invokeai.app.util.model_exclude_null import BaseModelExcludeNull
+
+VIDEO_DTO_COLS = ", ".join(
+    [
+        "videos." + c
+        for c in [
+            "video_id",
+            "width",
+            "height",
+            "session_id",
+            "node_id",
+            "is_intermediate",
+            "created_at",
+            "updated_at",
+            "deleted_at",
+            "starred",
+        ]
+    ]
+)
+
+
+class VideoRecord(BaseModelExcludeNull):
+    """Deserialized video record without metadata."""
+
+    video_id: str = Field(description="The unique id of the video.")
+    """The unique id of the video."""
+    width: int = Field(description="The width of the video in px.")
+    """The actual width of the video in px. This may be different from the width in metadata."""
+    height: int = Field(description="The height of the video in px.")
+    """The actual height of the video in px. This may be different from the height in metadata."""
+    created_at: Union[datetime.datetime, str] = Field(description="The created timestamp of the video.")
+    """The created timestamp of the video."""
+    updated_at: Union[datetime.datetime, str] = Field(description="The updated timestamp of the video.")
+    """The updated timestamp of the video."""
+    deleted_at: Optional[Union[datetime.datetime, str]] = Field(
+        default=None, description="The deleted timestamp of the video."
+    )
+    """The deleted timestamp of the video."""
+    is_intermediate: bool = Field(description="Whether this is an intermediate video.")
+    """Whether this is an intermediate video."""
+    session_id: Optional[str] = Field(
+        default=None,
+        description="The session ID that generated this video, if it is a generated video.",
+    )
+    """The session ID that generated this video, if it is a generated video."""
+    node_id: Optional[str] = Field(
+        default=None,
+        description="The node ID that generated this video, if it is a generated video.",
+    )
+    """The node ID that generated this video, if it is a generated video."""
+    starred: bool = Field(description="Whether this video is starred.")
+    """Whether this video is starred."""
+
+
+class VideoRecordChanges(BaseModelExcludeNull):
+    """A set of changes to apply to a video record.
+
+    Only limited changes are valid:
+      - `session_id`: change the session associated with a video
+      - `is_intermediate`: change the video's `is_intermediate` flag
+      - `starred`: change whether the video is starred
+    """
+
+    session_id: Optional[StrictStr] = Field(
+        default=None,
+        description="The video's new session ID.",
+    )
+    """The video's new session ID."""
+    is_intermediate: Optional[StrictBool] = Field(default=None, description="The video's new `is_intermediate` flag.")
+    """The video's new `is_intermediate` flag."""
+    starred: Optional[StrictBool] = Field(default=None, description="The video's new `starred` state")
+    """The video's new `starred` state."""
+
+
+def deserialize_video_record(video_dict: dict) -> VideoRecord:
+    """Deserializes a video record."""
+
+    # Retrieve all the values, setting "reasonable" defaults if they are not present.
+    video_id = video_dict.get("video_id", "unknown")
+    width = video_dict.get("width", 0)
+    height = video_dict.get("height", 0)
+    session_id = video_dict.get("session_id", None)
+    node_id = video_dict.get("node_id", None)
+    created_at = video_dict.get("created_at", get_iso_timestamp())
+    updated_at = video_dict.get("updated_at", get_iso_timestamp())
+    deleted_at = video_dict.get("deleted_at", get_iso_timestamp())
+    is_intermediate = video_dict.get("is_intermediate", False)
+    starred = video_dict.get("starred", False)
+
+    return VideoRecord(
+        video_id=video_id,
+        width=width,
+        height=height,
+        session_id=session_id,
+        node_id=node_id,
+        created_at=created_at,
+        updated_at=updated_at,
+        deleted_at=deleted_at,
+        is_intermediate=is_intermediate,
+        starred=starred,
+    )
+
+
+class VideoCollectionCounts(BaseModel):
+    starred_count: int = Field(description="The number of starred videos in the collection.")
+    unstarred_count: int = Field(description="The number of unstarred videos in the collection.")
+
+
+class VideoIdsResult(BaseModel):
+    """Response containing ordered video ids with metadata for optimistic updates."""
+
+    video_ids: list[str] = Field(description="Ordered list of video ids")
+    starred_count: int = Field(description="Number of starred videos (when starred_first=True)")
+    total_count: int = Field(description="Total number of videos matching the query")
+
+
+class VideoUrlsDTO(BaseModelExcludeNull):
+    """The URLs for an image and its thumbnail."""
+
+    video_id: str = Field(description="The unique id of the video.")
+    """The unique id of the video."""
+    video_url: str = Field(description="The URL of the video.")
+    """The URL of the video."""
+    thumbnail_url: str = Field(description="The URL of the video's thumbnail.")
+    """The URL of the video's thumbnail."""
+
+
+class VideoDTO(VideoRecord, VideoUrlsDTO):
+    """Deserialized video record, enriched for the frontend."""
+
+    board_id: Optional[str] = Field(
+        default=None, description="The id of the board the image belongs to, if one exists."
+    )
+    """The id of the board the image belongs to, if one exists."""
+
+
+def video_record_to_dto(
+    video_record: VideoRecord,
+    video_url: str,
+    thumbnail_url: str,
+    board_id: Optional[str],
+) -> VideoDTO:
+    """Converts a video record to a video DTO."""
+    return VideoDTO(
+        **video_record.model_dump(),
+        video_url=video_url,
+        thumbnail_url=thumbnail_url,
+        board_id=board_id,
+    )
+
+
+class ResultWithAffectedBoards(BaseModel):
+    affected_boards: list[str] = Field(description="The ids of boards affected by the delete operation")
+
+
+class DeleteVideosResult(ResultWithAffectedBoards):
+    deleted_videos: list[str] = Field(description="The ids of the videos that were deleted")
+
+
+class StarredVideosResult(ResultWithAffectedBoards):
+    starred_videos: list[str] = Field(description="The ids of the videos that were starred")
+
+
+class UnstarredVideosResult(ResultWithAffectedBoards):
+    unstarred_videos: list[str] = Field(description="The ids of the videos that were unstarred")
+
+
+class AddVideosToBoardResult(ResultWithAffectedBoards):
+    added_videos: list[str] = Field(description="The video ids that were added to the board")
+
+
+class RemoveVideosFromBoardResult(ResultWithAffectedBoards):
+    removed_videos: list[str] = Field(description="The video ids that were removed from their board")

invokeai/backend/flux/denoise.py

@@ -112,7 +112,7 @@ def denoise(
         )
 
         # Slice prediction to only include the main image tokens
-        if img_input_ids is not None:
+        if img_cond_seq is not None:
             pred = pred[:, :original_seq_len]
 
         step_cfg_scale = cfg_scale[step_index]
@@ -125,9 +125,26 @@ def denoise(
             if neg_regional_prompting_extension is None:
                 raise ValueError("Negative text conditioning is required when cfg_scale is not 1.0.")
 
+            # For negative prediction with Kontext, we need to include the reference images
+            # to maintain consistency between positive and negative passes. Without this,
+            # CFG would create artifacts as the attention mechanism would see different
+            # spatial structures in each pass
+            neg_img_input = img
+            neg_img_input_ids = img_ids
+
+            # Add channel-wise conditioning for negative pass if present
+            if img_cond is not None:
+                neg_img_input = torch.cat((neg_img_input, img_cond), dim=-1)
+
+            # Add sequence-wise conditioning (Kontext) for negative pass
+            # This ensures reference images are processed consistently
+            if img_cond_seq is not None:
+                neg_img_input = torch.cat((neg_img_input, img_cond_seq), dim=1)
+                neg_img_input_ids = torch.cat((neg_img_input_ids, img_cond_seq_ids), dim=1)
+
             neg_pred = model(
-                img=img,
-                img_ids=img_ids,
+                img=neg_img_input,
+                img_ids=neg_img_input_ids,
                 txt=neg_regional_prompting_extension.regional_text_conditioning.t5_embeddings,
                 txt_ids=neg_regional_prompting_extension.regional_text_conditioning.t5_txt_ids,
                 y=neg_regional_prompting_extension.regional_text_conditioning.clip_embeddings,
@@ -140,6 +157,10 @@ def denoise(
                 ip_adapter_extensions=neg_ip_adapter_extensions,
                 regional_prompting_extension=neg_regional_prompting_extension,
             )
+
+            # Slice negative prediction to match main image tokens
+            if img_cond_seq is not None:
+                neg_pred = neg_pred[:, :original_seq_len]
             pred = neg_pred + step_cfg_scale * (pred - neg_pred)
 
         preview_img = img - t_curr * pred

invokeai/backend/flux/extensions/kontext_extension.py

@@ -1,15 +1,14 @@
-import einops
-import numpy as np
 import torch
+import torch.nn.functional as F
+import torchvision.transforms as T
 from einops import repeat
-from PIL import Image
 
 from invokeai.app.invocations.fields import FluxKontextConditioningField
-from invokeai.app.invocations.flux_vae_encode import FluxVaeEncodeInvocation
 from invokeai.app.invocations.model import VAEField
 from invokeai.app.services.shared.invocation_context import InvocationContext
+from invokeai.backend.flux.modules.autoencoder import AutoEncoder
 from invokeai.backend.flux.sampling_utils import pack
-from invokeai.backend.flux.util import PREFERED_KONTEXT_RESOLUTIONS
+from invokeai.backend.util.devices import TorchDevice
 
 
 def generate_img_ids_with_offset(
@@ -19,8 +18,10 @@ def generate_img_ids_with_offset(
     device: torch.device,
     dtype: torch.dtype,
     idx_offset: int = 0,
+    h_offset: int = 0,
+    w_offset: int = 0,
 ) -> torch.Tensor:
-    """Generate tensor of image position ids with an optional offset.
+    """Generate tensor of image position ids with optional index and spatial offsets.
 
     Args:
         latent_height (int): Height of image in latent space (after packing, this becomes h//2).
@@ -28,7 +29,9 @@ def generate_img_ids_with_offset(
         batch_size (int): Number of images in the batch.
         device (torch.device): Device to create tensors on.
         dtype (torch.dtype): Data type for the tensors.
-        idx_offset (int): Offset to add to the first dimension of the image ids.
+        idx_offset (int): Offset to add to the first dimension of the image ids (default: 0).
+        h_offset (int): Spatial offset for height/y-coordinates in latent space (default: 0).
+        w_offset (int): Spatial offset for width/x-coordinates in latent space (default: 0).
 
     Returns:
         torch.Tensor: Image position ids with shape [batch_size, (latent_height//2 * latent_width//2), 3].
@@ -42,6 +45,10 @@ def generate_img_ids_with_offset(
     packed_height = latent_height // 2
     packed_width = latent_width // 2
 
+    # Convert spatial offsets from latent space to packed space
+    packed_h_offset = h_offset // 2
+    packed_w_offset = w_offset // 2
+
     # Create base tensor for position IDs with shape [packed_height, packed_width, 3]
     # The 3 channels represent: [batch_offset, y_position, x_position]
     img_ids = torch.zeros(packed_height, packed_width, 3, device=device, dtype=dtype)
@@ -49,13 +56,13 @@ def generate_img_ids_with_offset(
     # Set the batch offset for all positions
     img_ids[..., 0] = idx_offset
 
-    # Create y-coordinate indices (vertical positions)
-    y_indices = torch.arange(packed_height, device=device, dtype=dtype)
+    # Create y-coordinate indices (vertical positions) with spatial offset
+    y_indices = torch.arange(packed_height, device=device, dtype=dtype) + packed_h_offset
     # Broadcast y_indices to match the spatial dimensions [packed_height, 1]
     img_ids[..., 1] = y_indices[:, None]
 
-    # Create x-coordinate indices (horizontal positions)
-    x_indices = torch.arange(packed_width, device=device, dtype=dtype)
+    # Create x-coordinate indices (horizontal positions) with spatial offset
+    x_indices = torch.arange(packed_width, device=device, dtype=dtype) + packed_w_offset
     # Broadcast x_indices to match the spatial dimensions [1, packed_width]
     img_ids[..., 2] = x_indices[None, :]
 
@@ -73,14 +80,14 @@ class KontextExtension:
 
     def __init__(
         self,
-        kontext_conditioning: FluxKontextConditioningField,
+        kontext_conditioning: list[FluxKontextConditioningField],
         context: InvocationContext,
         vae_field: VAEField,
         device: torch.device,
         dtype: torch.dtype,
     ):
         """
-        Initializes the KontextExtension, pre-processing the reference image
+        Initializes the KontextExtension, pre-processing the reference images
         into latents and positional IDs.
         """
         self._context = context
@@ -93,54 +100,116 @@ class KontextExtension:
         self.kontext_latents, self.kontext_ids = self._prepare_kontext()
 
     def _prepare_kontext(self) -> tuple[torch.Tensor, torch.Tensor]:
-        """Encodes the reference image and prepares its latents and IDs."""
-        image = self._context.images.get_pil(self.kontext_conditioning.image.image_name)
+        """Encodes the reference images and prepares their concatenated latents and IDs with spatial tiling."""
+        all_latents = []
+        all_ids = []
 
-        # Calculate aspect ratio of input image
-        width, height = image.size
-        aspect_ratio = width / height
+        # Track cumulative dimensions for spatial tiling
+        # These track the running extent of the virtual canvas in latent space
+        canvas_h = 0  # Running canvas height
+        canvas_w = 0  # Running canvas width
 
-        # Find the closest preferred resolution by aspect ratio
-        _, target_width, target_height = min(
-            ((abs(aspect_ratio - w / h), w, h) for w, h in PREFERED_KONTEXT_RESOLUTIONS), key=lambda x: x[0]
-        )
-
-        # Apply BFL's scaling formula
-        # This ensures compatibility with the model's training
-        scaled_width = 2 * int(target_width / 16)
-        scaled_height = 2 * int(target_height / 16)
-
-        # Resize to the exact resolution used during training
-        image = image.convert("RGB")
-        final_width = 8 * scaled_width
-        final_height = 8 * scaled_height
-        image = image.resize((final_width, final_height), Image.Resampling.LANCZOS)
-
-        # Convert to tensor with same normalization as BFL
-        image_np = np.array(image)
-        image_tensor = torch.from_numpy(image_np).float() / 127.5 - 1.0
-        image_tensor = einops.rearrange(image_tensor, "h w c -> 1 c h w")
-        image_tensor = image_tensor.to(self._device)
-
-        # Continue with VAE encoding
         vae_info = self._context.models.load(self._vae_field.vae)
-        kontext_latents_unpacked = FluxVaeEncodeInvocation.vae_encode(vae_info=vae_info, image_tensor=image_tensor)
 
-        # Extract tensor dimensions
-        batch_size, _, latent_height, latent_width = kontext_latents_unpacked.shape
+        for idx, kontext_field in enumerate(self.kontext_conditioning):
+            image = self._context.images.get_pil(kontext_field.image.image_name)
 
-        # Pack the latents and generate IDs
-        kontext_latents_packed = pack(kontext_latents_unpacked).to(self._device, self._dtype)
-        kontext_ids = generate_img_ids_with_offset(
-            latent_height=latent_height,
-            latent_width=latent_width,
-            batch_size=batch_size,
-            device=self._device,
-            dtype=self._dtype,
-            idx_offset=1,
-        )
+            # Convert to RGB
+            image = image.convert("RGB")
 
-        return kontext_latents_packed, kontext_ids
+            # Convert to tensor using torchvision transforms for consistency
+            transformation = T.Compose(
+                [
+                    T.ToTensor(),  # Converts PIL image to tensor and scales to [0, 1]
+                ]
+            )
+            image_tensor = transformation(image)
+            # Convert from [0, 1] to [-1, 1] range expected by VAE
+            image_tensor = image_tensor * 2.0 - 1.0
+            image_tensor = image_tensor.unsqueeze(0)  # Add batch dimension
+            image_tensor = image_tensor.to(self._device)
+
+            # Continue with VAE encoding
+            # Don't sample from the distribution for reference images - use the mean (matching ComfyUI)
+            # Estimate working memory for encode operation (50% of decode memory requirements)
+            img_h = image_tensor.shape[-2]
+            img_w = image_tensor.shape[-1]
+            element_size = next(vae_info.model.parameters()).element_size()
+            scaling_constant = 1100  # 50% of decode scaling constant (2200)
+            estimated_working_memory = int(img_h * img_w * element_size * scaling_constant)
+
+            with vae_info.model_on_device(working_mem_bytes=estimated_working_memory) as (_, vae):
+                assert isinstance(vae, AutoEncoder)
+                vae_dtype = next(iter(vae.parameters())).dtype
+                image_tensor = image_tensor.to(device=TorchDevice.choose_torch_device(), dtype=vae_dtype)
+                # Use sample=False to get the distribution mean without noise
+                kontext_latents_unpacked = vae.encode(image_tensor, sample=False)
+                TorchDevice.empty_cache()
+
+            # Extract tensor dimensions
+            batch_size, _, latent_height, latent_width = kontext_latents_unpacked.shape
+
+            # Pad latents to be compatible with patch_size=2
+            # This ensures dimensions are even for the pack() function
+            pad_h = (2 - latent_height % 2) % 2
+            pad_w = (2 - latent_width % 2) % 2
+            if pad_h > 0 or pad_w > 0:
+                kontext_latents_unpacked = F.pad(kontext_latents_unpacked, (0, pad_w, 0, pad_h), mode="circular")
+                # Update dimensions after padding
+                _, _, latent_height, latent_width = kontext_latents_unpacked.shape
+
+            # Pack the latents
+            kontext_latents_packed = pack(kontext_latents_unpacked).to(self._device, self._dtype)
+
+            # Determine spatial offsets for this reference image
+            h_offset = 0
+            w_offset = 0
+
+            if idx > 0:  # First image starts at (0, 0)
+                # Calculate potential canvas dimensions for each tiling option
+                # Option 1: Tile vertically (below existing content)
+                potential_h_vertical = canvas_h + latent_height
+
+                # Option 2: Tile horizontally (to the right of existing content)
+                potential_w_horizontal = canvas_w + latent_width
+
+                # Choose arrangement that minimizes the maximum dimension
+                # This keeps the canvas closer to square, optimizing attention computation
+                if potential_h_vertical > potential_w_horizontal:
+                    # Tile horizontally (to the right of existing images)
+                    w_offset = canvas_w
+                    canvas_w = canvas_w + latent_width
+                    canvas_h = max(canvas_h, latent_height)
+                else:
+                    # Tile vertically (below existing images)
+                    h_offset = canvas_h
+                    canvas_h = canvas_h + latent_height
+                    canvas_w = max(canvas_w, latent_width)
+            else:
+                # First image - just set canvas dimensions
+                canvas_h = latent_height
+                canvas_w = latent_width
+
+            # Generate IDs with both index offset and spatial offsets
+            kontext_ids = generate_img_ids_with_offset(
+                latent_height=latent_height,
+                latent_width=latent_width,
+                batch_size=batch_size,
+                device=self._device,
+                dtype=self._dtype,
+                idx_offset=1,  # All reference images use index=1 (matching ComfyUI implementation)
+                h_offset=h_offset,
+                w_offset=w_offset,
+            )
+
+            all_latents.append(kontext_latents_packed)
+            all_ids.append(kontext_ids)
+
+        # Concatenate all latents and IDs along the sequence dimension
+        concatenated_latents = torch.cat(all_latents, dim=1)  # Concatenate along sequence dimension
+        concatenated_ids = torch.cat(all_ids, dim=1)  # Concatenate along sequence dimension
+
+        return concatenated_latents, concatenated_ids
 
     def ensure_batch_size(self, target_batch_size: int) -> None:
         """Ensures the kontext latents and IDs match the target batch size by repeating if necessary."""

invokeai/backend/image_util/imwatermark/vendor.py

@@ -0,0 +1,304 @@
+# This file is vendored from https://github.com/ShieldMnt/invisible-watermark
+#
+# `invisible-watermark` is MIT licensed as of August 23, 2025, when the code was copied into this repo.
+#
+# Why we vendored it in:
+# `invisible-watermark` has a dependency on `opencv-python`, which conflicts with Invoke's dependency on
+# `opencv-contrib-python`. It's easier to copy the code over than complicate the installation process by
+# requiring an extra post-install step of removing `opencv-python` and installing `opencv-contrib-python`.
+
+import struct
+import uuid
+import base64
+import cv2
+import numpy as np
+import pywt
+
+
+class WatermarkEncoder(object):
+    def __init__(self, content=b""):
+        seq = np.array([n for n in content], dtype=np.uint8)
+        self._watermarks = list(np.unpackbits(seq))
+        self._wmLen = len(self._watermarks)
+        self._wmType = "bytes"
+
+    def set_by_ipv4(self, addr):
+        bits = []
+        ips = addr.split(".")
+        for ip in ips:
+            bits += list(np.unpackbits(np.array([ip % 255], dtype=np.uint8)))
+        self._watermarks = bits
+        self._wmLen = len(self._watermarks)
+        self._wmType = "ipv4"
+        assert self._wmLen == 32
+
+    def set_by_uuid(self, uid):
+        u = uuid.UUID(uid)
+        self._wmType = "uuid"
+        seq = np.array([n for n in u.bytes], dtype=np.uint8)
+        self._watermarks = list(np.unpackbits(seq))
+        self._wmLen = len(self._watermarks)
+
+    def set_by_bytes(self, content):
+        self._wmType = "bytes"
+        seq = np.array([n for n in content], dtype=np.uint8)
+        self._watermarks = list(np.unpackbits(seq))
+        self._wmLen = len(self._watermarks)
+
+    def set_by_b16(self, b16):
+        content = base64.b16decode(b16)
+        self.set_by_bytes(content)
+        self._wmType = "b16"
+
+    def set_by_bits(self, bits=[]):
+        self._watermarks = [int(bit) % 2 for bit in bits]
+        self._wmLen = len(self._watermarks)
+        self._wmType = "bits"
+
+    def set_watermark(self, wmType="bytes", content=""):
+        if wmType == "ipv4":
+            self.set_by_ipv4(content)
+        elif wmType == "uuid":
+            self.set_by_uuid(content)
+        elif wmType == "bits":
+            self.set_by_bits(content)
+        elif wmType == "bytes":
+            self.set_by_bytes(content)
+        elif wmType == "b16":
+            self.set_by_b16(content)
+        else:
+            raise NameError("%s is not supported" % wmType)
+
+    def get_length(self):
+        return self._wmLen
+
+    # @classmethod
+    # def loadModel(cls):
+    #     RivaWatermark.loadModel()
+
+    def encode(self, cv2Image, method="dwtDct", **configs):
+        (r, c, channels) = cv2Image.shape
+        if r * c < 256 * 256:
+            raise RuntimeError("image too small, should be larger than 256x256")
+
+        if method == "dwtDct":
+            embed = EmbedMaxDct(self._watermarks, wmLen=self._wmLen, **configs)
+            return embed.encode(cv2Image)
+        # elif method == 'dwtDctSvd':
+        #     embed = EmbedDwtDctSvd(self._watermarks, wmLen=self._wmLen, **configs)
+        #     return embed.encode(cv2Image)
+        # elif method == 'rivaGan':
+        #     embed = RivaWatermark(self._watermarks, self._wmLen)
+        #     return embed.encode(cv2Image)
+        else:
+            raise NameError("%s is not supported" % method)
+
+
+class WatermarkDecoder(object):
+    def __init__(self, wm_type="bytes", length=0):
+        self._wmType = wm_type
+        if wm_type == "ipv4":
+            self._wmLen = 32
+        elif wm_type == "uuid":
+            self._wmLen = 128
+        elif wm_type == "bytes":
+            self._wmLen = length
+        elif wm_type == "bits":
+            self._wmLen = length
+        elif wm_type == "b16":
+            self._wmLen = length
+        else:
+            raise NameError("%s is unsupported" % wm_type)
+
+    def reconstruct_ipv4(self, bits):
+        ips = [str(ip) for ip in list(np.packbits(bits))]
+        return ".".join(ips)
+
+    def reconstruct_uuid(self, bits):
+        nums = np.packbits(bits)
+        bstr = b""
+        for i in range(16):
+            bstr += struct.pack(">B", nums[i])
+
+        return str(uuid.UUID(bytes=bstr))
+
+    def reconstruct_bits(self, bits):
+        # return ''.join([str(b) for b in bits])
+        return bits
+
+    def reconstruct_b16(self, bits):
+        bstr = self.reconstruct_bytes(bits)
+        return base64.b16encode(bstr)
+
+    def reconstruct_bytes(self, bits):
+        nums = np.packbits(bits)
+        bstr = b""
+        for i in range(self._wmLen // 8):
+            bstr += struct.pack(">B", nums[i])
+        return bstr
+
+    def reconstruct(self, bits):
+        if len(bits) != self._wmLen:
+            raise RuntimeError("bits are not matched with watermark length")
+
+        if self._wmType == "ipv4":
+            return self.reconstruct_ipv4(bits)
+        elif self._wmType == "uuid":
+            return self.reconstruct_uuid(bits)
+        elif self._wmType == "bits":
+            return self.reconstruct_bits(bits)
+        elif self._wmType == "b16":
+            return self.reconstruct_b16(bits)
+        else:
+            return self.reconstruct_bytes(bits)
+
+    def decode(self, cv2Image, method="dwtDct", **configs):
+        (r, c, channels) = cv2Image.shape
+        if r * c < 256 * 256:
+            raise RuntimeError("image too small, should be larger than 256x256")
+
+        bits = []
+        if method == "dwtDct":
+            embed = EmbedMaxDct(watermarks=[], wmLen=self._wmLen, **configs)
+            bits = embed.decode(cv2Image)
+        # elif method == 'dwtDctSvd':
+        #     embed = EmbedDwtDctSvd(watermarks=[], wmLen=self._wmLen, **configs)
+        #     bits = embed.decode(cv2Image)
+        # elif method == 'rivaGan':
+        #     embed = RivaWatermark(watermarks=[], wmLen=self._wmLen, **configs)
+        #     bits = embed.decode(cv2Image)
+        else:
+            raise NameError("%s is not supported" % method)
+        return self.reconstruct(bits)
+
+    # @classmethod
+    # def loadModel(cls):
+    #     RivaWatermark.loadModel()
+
+
+class EmbedMaxDct(object):
+    def __init__(self, watermarks=[], wmLen=8, scales=[0, 36, 36], block=4):
+        self._watermarks = watermarks
+        self._wmLen = wmLen
+        self._scales = scales
+        self._block = block
+
+    def encode(self, bgr):
+        (row, col, channels) = bgr.shape
+
+        yuv = cv2.cvtColor(bgr, cv2.COLOR_BGR2YUV)
+
+        for channel in range(2):
+            if self._scales[channel] <= 0:
+                continue
+
+            ca1, (h1, v1, d1) = pywt.dwt2(yuv[: row // 4 * 4, : col // 4 * 4, channel], "haar")
+            self.encode_frame(ca1, self._scales[channel])
+
+            yuv[: row // 4 * 4, : col // 4 * 4, channel] = pywt.idwt2((ca1, (v1, h1, d1)), "haar")
+
+        bgr_encoded = cv2.cvtColor(yuv, cv2.COLOR_YUV2BGR)
+        return bgr_encoded
+
+    def decode(self, bgr):
+        (row, col, channels) = bgr.shape
+
+        yuv = cv2.cvtColor(bgr, cv2.COLOR_BGR2YUV)
+
+        scores = [[] for i in range(self._wmLen)]
+        for channel in range(2):
+            if self._scales[channel] <= 0:
+                continue
+
+            ca1, (h1, v1, d1) = pywt.dwt2(yuv[: row // 4 * 4, : col // 4 * 4, channel], "haar")
+
+            scores = self.decode_frame(ca1, self._scales[channel], scores)
+
+        avgScores = list(map(lambda l: np.array(l).mean(), scores))
+
+        bits = np.array(avgScores) * 255 > 127
+        return bits
+
+    def decode_frame(self, frame, scale, scores):
+        (row, col) = frame.shape
+        num = 0
+
+        for i in range(row // self._block):
+            for j in range(col // self._block):
+                block = frame[
+                    i * self._block : i * self._block + self._block, j * self._block : j * self._block + self._block
+                ]
+
+                score = self.infer_dct_matrix(block, scale)
+                # score = self.infer_dct_svd(block, scale)
+                wmBit = num % self._wmLen
+                scores[wmBit].append(score)
+                num = num + 1
+
+        return scores
+
+    def diffuse_dct_svd(self, block, wmBit, scale):
+        u, s, v = np.linalg.svd(cv2.dct(block))
+
+        s[0] = (s[0] // scale + 0.25 + 0.5 * wmBit) * scale
+        return cv2.idct(np.dot(u, np.dot(np.diag(s), v)))
+
+    def infer_dct_svd(self, block, scale):
+        u, s, v = np.linalg.svd(cv2.dct(block))
+
+        score = 0
+        score = int((s[0] % scale) > scale * 0.5)
+        return score
+        if score >= 0.5:
+            return 1.0
+        else:
+            return 0.0
+
+    def diffuse_dct_matrix(self, block, wmBit, scale):
+        pos = np.argmax(abs(block.flatten()[1:])) + 1
+        i, j = pos // self._block, pos % self._block
+        val = block[i][j]
+        if val >= 0.0:
+            block[i][j] = (val // scale + 0.25 + 0.5 * wmBit) * scale
+        else:
+            val = abs(val)
+            block[i][j] = -1.0 * (val // scale + 0.25 + 0.5 * wmBit) * scale
+        return block
+
+    def infer_dct_matrix(self, block, scale):
+        pos = np.argmax(abs(block.flatten()[1:])) + 1
+        i, j = pos // self._block, pos % self._block
+
+        val = block[i][j]
+        if val < 0:
+            val = abs(val)
+
+        if (val % scale) > 0.5 * scale:
+            return 1
+        else:
+            return 0
+
+    def encode_frame(self, frame, scale):
+        """
+        frame is a matrix (M, N)
+
+        we get K (watermark bits size) blocks (self._block x self._block)
+
+        For i-th block, we encode watermark[i] bit into it
+        """
+        (row, col) = frame.shape
+        num = 0
+        for i in range(row // self._block):
+            for j in range(col // self._block):
+                block = frame[
+                    i * self._block : i * self._block + self._block, j * self._block : j * self._block + self._block
+                ]
+                wmBit = self._watermarks[(num % self._wmLen)]
+
+                diffusedBlock = self.diffuse_dct_matrix(block, wmBit, scale)
+                # diffusedBlock = self.diffuse_dct_svd(block, wmBit, scale)
+                frame[
+                    i * self._block : i * self._block + self._block, j * self._block : j * self._block + self._block
+                ] = diffusedBlock
+
+                num = num + 1

invokeai/backend/image_util/invisible_watermark.py

@@ -6,13 +6,10 @@ configuration variable, that allows the watermarking to be supressed.
 
 import cv2
 import numpy as np
-from imwatermark import WatermarkEncoder
 from PIL import Image
 
 import invokeai.backend.util.logging as logger
-from invokeai.app.services.config.config_default import get_config
-
-config = get_config()
+from invokeai.backend.image_util.imwatermark.vendor import WatermarkEncoder
 
 
 class InvisibleWatermark:

invokeai/backend/image_util/segment_anything/segment_anything_2_pipeline.py

@@ -0,0 +1,109 @@
+from typing import Optional
+
+import torch
+from PIL import Image
+
+# Import SAM2 components - these should be available in transformers 4.56.0+
+from transformers.models.sam2 import Sam2Model
+from transformers.models.sam2.processing_sam2 import Sam2Processor
+
+from invokeai.backend.image_util.segment_anything.shared import SAMInput
+from invokeai.backend.raw_model import RawModel
+
+
+class SegmentAnything2Pipeline(RawModel):
+    """A wrapper class for the transformers SAM2 model and processor that makes it compatible with the model manager."""
+
+    def __init__(self, sam2_model: Sam2Model, sam2_processor: Sam2Processor):
+        """Initialize the SAM2 pipeline.
+
+        Args:
+            sam2_model: The SAM2 model
+            sam2_processor: The SAM2 processor (can be Sam2Processor or Sam2VideoProcessor)
+        """
+        self._sam2_model = sam2_model
+        self._sam2_processor = sam2_processor
+
+    def to(self, device: Optional[torch.device] = None, dtype: Optional[torch.dtype] = None):
+        # HACK: The SAM2 pipeline may not work on MPS devices. We only allow it to be moved to CPU or CUDA.
+        if device is not None and device.type not in {"cpu", "cuda"}:
+            device = None
+        self._sam2_model.to(device=device, dtype=dtype)
+
+    def calc_size(self) -> int:
+        # HACK: Fix the circular import issue.
+        from invokeai.backend.model_manager.load.model_util import calc_module_size
+
+        return calc_module_size(self._sam2_model)
+
+    def segment(
+        self,
+        image: Image.Image,
+        inputs: list[SAMInput],
+    ) -> torch.Tensor:
+        """Segment the image using the provided inputs.
+
+        Args:
+            image: The image to segment.
+            inputs: A list of SAMInput objects containing bounding boxes and/or point lists.
+
+        Returns:
+            torch.Tensor: The segmentation masks. dtype: torch.bool. shape: [num_masks, channels, height, width].
+        """
+
+        input_boxes: list[list[float]] = []
+        input_points: list[list[list[float]]] = []
+        input_labels: list[list[int]] = []
+
+        for i in inputs:
+            box: list[float] | None = None
+            points: list[list[float]] | None = None
+            labels: list[int] | None = None
+
+            if i.bounding_box is not None:
+                box: list[float] | None = [
+                    i.bounding_box.x_min,
+                    i.bounding_box.y_min,
+                    i.bounding_box.x_max,
+                    i.bounding_box.y_max,
+                ]
+
+            if i.points is not None:
+                points = []
+                labels = []
+                for point in i.points:
+                    points.append([point.x, point.y])
+                    labels.append(point.label.value)
+
+            if box is not None:
+                input_boxes.append(box)
+            if points is not None:
+                input_points.append(points)
+            if labels is not None:
+                input_labels.append(labels)
+
+        batched_input_boxes = [input_boxes] if input_boxes else None
+        batched_input_points = [input_points] if input_points else None
+        batched_input_labels = [input_labels] if input_labels else None
+
+        processed_inputs = self._sam2_processor(
+            images=image,
+            input_boxes=batched_input_boxes,
+            input_points=batched_input_points,
+            input_labels=batched_input_labels,
+            return_tensors="pt",
+        ).to(self._sam2_model.device)
+
+        # Generate masks using the SAM2 model
+        outputs = self._sam2_model(**processed_inputs)
+
+        # Post-process the masks to get the final segmentation
+        masks = self._sam2_processor.post_process_masks(
+            masks=outputs.pred_masks,
+            original_sizes=processed_inputs.original_sizes,
+            reshaped_input_sizes=processed_inputs.reshaped_input_sizes,
+        )
+
+        # There should be only one batch.
+        assert len(masks) == 1
+        return masks[0]

invokeai/backend/image_util/segment_anything/segment_anything_pipeline.py

@@ -1,20 +1,13 @@
-from typing import Optional, TypeAlias
+from typing import Optional
 
 import torch
 from PIL import Image
 from transformers.models.sam import SamModel
 from transformers.models.sam.processing_sam import SamProcessor
 
+from invokeai.backend.image_util.segment_anything.shared import SAMInput
 from invokeai.backend.raw_model import RawModel
 
-# Type aliases for the inputs to the SAM model.
-ListOfBoundingBoxes: TypeAlias = list[list[int]]
-"""A list of bounding boxes. Each bounding box is in the format [xmin, ymin, xmax, ymax]."""
-ListOfPoints: TypeAlias = list[list[int]]
-"""A list of points. Each point is in the format [x, y]."""
-ListOfPointLabels: TypeAlias = list[int]
-"""A list of SAM point labels. Each label is an integer where -1 is background, 0 is neutral, and 1 is foreground."""
-
 
 class SegmentAnythingPipeline(RawModel):
     """A wrapper class for the transformers SAM model and processor that makes it compatible with the model manager."""
@@ -38,55 +31,65 @@ class SegmentAnythingPipeline(RawModel):
     def segment(
         self,
         image: Image.Image,
-        bounding_boxes: list[list[int]] | None = None,
-        point_lists: list[list[list[int]]] | None = None,
+        inputs: list[SAMInput],
     ) -> torch.Tensor:
-        """Run the SAM model.
-
-        Either bounding_boxes or point_lists must be provided. If both are provided, bounding_boxes will be used and
-        point_lists will be ignored.
+        """Segment the image using the provided inputs.
 
         Args:
-            image (Image.Image): The image to segment.
-            bounding_boxes (list[list[int]]): The bounding box prompts. Each bounding box is in the format
-                [xmin, ymin, xmax, ymax].
-            point_lists (list[list[list[int]]]): The points prompts. Each point is in the format [x, y, label].
-                `label` is an integer where -1 is background, 0 is neutral, and 1 is foreground.
+            image: The image to segment.
+            inputs: A list of SAMInput objects containing bounding boxes and/or point lists.
 
         Returns:
             torch.Tensor: The segmentation masks. dtype: torch.bool. shape: [num_masks, channels, height, width].
         """
 
-        # Prep the inputs:
-        # - Create a list of bounding boxes or points and labels.
-        # - Add a batch dimension of 1 to the inputs.
-        if bounding_boxes:
-            input_boxes: list[ListOfBoundingBoxes] | None = [bounding_boxes]
-            input_points: list[ListOfPoints] | None = None
-            input_labels: list[ListOfPointLabels] | None = None
-        elif point_lists:
-            input_boxes: list[ListOfBoundingBoxes] | None = None
-            input_points: list[ListOfPoints] | None = []
-            input_labels: list[ListOfPointLabels] | None = []
-            for point_list in point_lists:
-                input_points.append([[p[0], p[1]] for p in point_list])
-                input_labels.append([p[2] for p in point_list])
+        input_boxes: list[list[float]] = []
+        input_points: list[list[list[float]]] = []
+        input_labels: list[list[int]] = []
 
-        else:
-            raise ValueError("Either bounding_boxes or points and labels must be provided.")
+        for i in inputs:
+            box: list[float] | None = None
+            points: list[list[float]] | None = None
+            labels: list[int] | None = None
 
-        inputs = self._sam_processor(
+            if i.bounding_box is not None:
+                box: list[float] | None = [
+                    i.bounding_box.x_min,
+                    i.bounding_box.y_min,
+                    i.bounding_box.x_max,
+                    i.bounding_box.y_max,
+                ]
+
+            if i.points is not None:
+                points = []
+                labels = []
+                for point in i.points:
+                    points.append([point.x, point.y])
+                    labels.append(point.label.value)
+
+            if box is not None:
+                input_boxes.append(box)
+            if points is not None:
+                input_points.append(points)
+            if labels is not None:
+                input_labels.append(labels)
+
+        batched_input_boxes = [input_boxes] if input_boxes else None
+        batched_input_points = input_points if input_points else None
+        batched_input_labels = input_labels if input_labels else None
+
+        processed_inputs = self._sam_processor(
             images=image,
-            input_boxes=input_boxes,
-            input_points=input_points,
-            input_labels=input_labels,
+            input_boxes=batched_input_boxes,
+            input_points=batched_input_points,
+            input_labels=batched_input_labels,
             return_tensors="pt",
         ).to(self._sam_model.device)
-        outputs = self._sam_model(**inputs)
+        outputs = self._sam_model(**processed_inputs)
         masks = self._sam_processor.post_process_masks(
             masks=outputs.pred_masks,
-            original_sizes=inputs.original_sizes,
-            reshaped_input_sizes=inputs.reshaped_input_sizes,
+            original_sizes=processed_inputs.original_sizes,
+            reshaped_input_sizes=processed_inputs.reshaped_input_sizes,
         )
 
         # There should be only one batch.

invokeai/backend/image_util/segment_anything/shared.py

@@ -0,0 +1,49 @@
+from enum import Enum
+
+from pydantic import BaseModel, model_validator
+from pydantic.fields import Field
+
+
+class BoundingBox(BaseModel):
+    x_min: int = Field(..., description="The minimum x-coordinate of the bounding box (inclusive).")
+    x_max: int = Field(..., description="The maximum x-coordinate of the bounding box (exclusive).")
+    y_min: int = Field(..., description="The minimum y-coordinate of the bounding box (inclusive).")
+    y_max: int = Field(..., description="The maximum y-coordinate of the bounding box (exclusive).")
+
+    @model_validator(mode="after")
+    def check_coords(self):
+        if self.x_min > self.x_max:
+            raise ValueError(f"x_min ({self.x_min}) is greater than x_max ({self.x_max}).")
+        if self.y_min > self.y_max:
+            raise ValueError(f"y_min ({self.y_min}) is greater than y_max ({self.y_max}).")
+        return self
+
+    def tuple(self) -> tuple[int, int, int, int]:
+        """
+        Returns the bounding box as a tuple suitable for use with PIL's `Image.crop()` method.
+        This method returns a tuple of the form (left, upper, right, lower) == (x_min, y_min, x_max, y_max).
+        """
+        return (self.x_min, self.y_min, self.x_max, self.y_max)
+
+
+class SAMPointLabel(Enum):
+    negative = -1
+    neutral = 0
+    positive = 1
+
+
+class SAMPoint(BaseModel):
+    x: int = Field(..., description="The x-coordinate of the point")
+    y: int = Field(..., description="The y-coordinate of the point")
+    label: SAMPointLabel = Field(..., description="The label of the point")
+
+
+class SAMInput(BaseModel):
+    bounding_box: BoundingBox | None = Field(None, description="The bounding box to use for segmentation")
+    points: list[SAMPoint] | None = Field(None, description="The points to use for segmentation")
+
+    @model_validator(mode="after")
+    def check_input(self):
+        if not self.bounding_box and not self.points:
+            raise ValueError("Either bounding_box or points must be provided")
+        return self

invokeai/backend/ip_adapter/ip_adapter.py

@@ -207,15 +207,24 @@ class IPAdapterPlusXL(IPAdapterPlus):
 
 
 def load_ip_adapter_tensors(ip_adapter_ckpt_path: pathlib.Path, device: str) -> IPAdapterStateDict:
-    state_dict: IPAdapterStateDict = {"ip_adapter": {}, "image_proj": {}}
+    state_dict: IPAdapterStateDict = {
+        "ip_adapter": {},
+        "image_proj": {},
+        "adapter_modules": {},  # added for noobai-mark-ipa
+        "image_proj_model": {},  # added for noobai-mark-ipa
+    }
 
     if ip_adapter_ckpt_path.suffix == ".safetensors":
         model = safetensors.torch.load_file(ip_adapter_ckpt_path, device=device)
         for key in model.keys():
-            if key.startswith("image_proj."):
-                state_dict["image_proj"][key.replace("image_proj.", "")] = model[key]
-            elif key.startswith("ip_adapter."):
+            if key.startswith("ip_adapter."):
                 state_dict["ip_adapter"][key.replace("ip_adapter.", "")] = model[key]
+            elif key.startswith("image_proj_model."):
+                state_dict["image_proj_model"][key.replace("image_proj_model.", "")] = model[key]
+            elif key.startswith("image_proj."):
+                state_dict["image_proj"][key.replace("image_proj.", "")] = model[key]
+            elif key.startswith("adapter_modules."):
+                state_dict["adapter_modules"][key.replace("adapter_modules.", "")] = model[key]
             else:
                 raise RuntimeError(f"Encountered unexpected IP Adapter state dict key: '{key}'.")
     else:

invokeai/backend/model_manager/config.py

@@ -90,6 +90,11 @@ class MainModelDefaultSettings(BaseModel):
     model_config = ConfigDict(extra="forbid")
 
 
+class LoraModelDefaultSettings(BaseModel):
+    weight: float | None = Field(default=None, ge=-1, le=2, description="Default weight for this model")
+    model_config = ConfigDict(extra="forbid")
+
+
 class ControlAdapterDefaultSettings(BaseModel):
     # This could be narrowed to controlnet processor nodes, but they change. Leaving this a string is safer.
     preprocessor: str | None
@@ -287,6 +292,9 @@ class LoRAConfigBase(ABC, BaseModel):
 
     type: Literal[ModelType.LoRA] = ModelType.LoRA
     trigger_phrases: Optional[set[str]] = Field(description="Set of trigger phrases for this model", default=None)
+    default_settings: Optional[LoraModelDefaultSettings] = Field(
+        description="Default settings for this model", default=None
+    )
 
     @classmethod
     def flux_lora_format(cls, mod: ModelOnDisk):
@@ -492,6 +500,15 @@ class MainConfigBase(ABC, BaseModel):
     variant: AnyVariant = ModelVariantType.Normal
 
 
+class VideoConfigBase(ABC, BaseModel):
+    type: Literal[ModelType.Video] = ModelType.Video
+    trigger_phrases: Optional[set[str]] = Field(description="Set of trigger phrases for this model", default=None)
+    default_settings: Optional[MainModelDefaultSettings] = Field(
+        description="Default settings for this model", default=None
+    )
+    variant: AnyVariant = ModelVariantType.Normal
+
+
 class MainCheckpointConfig(CheckpointConfigBase, MainConfigBase, LegacyProbeMixin, ModelConfigBase):
     """Model config for main checkpoint models."""
 
@@ -649,6 +666,21 @@ class ApiModelConfig(MainConfigBase, ModelConfigBase):
         raise NotImplementedError("API models are not parsed from disk.")
 
 
+class VideoApiModelConfig(VideoConfigBase, ModelConfigBase):
+    """Model config for API-based video models."""
+
+    format: Literal[ModelFormat.Api] = ModelFormat.Api
+
+    @classmethod
+    def matches(cls, mod: ModelOnDisk) -> bool:
+        # API models are not stored on disk, so we can't match them.
+        return False
+
+    @classmethod
+    def parse(cls, mod: ModelOnDisk) -> dict[str, Any]:
+        raise NotImplementedError("API models are not parsed from disk.")
+
+
 def get_model_discriminator_value(v: Any) -> str:
     """
     Computes the discriminator value for a model config.
@@ -718,12 +750,13 @@ AnyModelConfig = Annotated[
         Annotated[FluxReduxConfig, FluxReduxConfig.get_tag()],
         Annotated[LlavaOnevisionConfig, LlavaOnevisionConfig.get_tag()],
         Annotated[ApiModelConfig, ApiModelConfig.get_tag()],
+        Annotated[VideoApiModelConfig, VideoApiModelConfig.get_tag()],
     ],
     Discriminator(get_model_discriminator_value),
 ]
 
 AnyModelConfigValidator = TypeAdapter(AnyModelConfig)
-AnyDefaultSettings: TypeAlias = Union[MainModelDefaultSettings, ControlAdapterDefaultSettings]
+AnyDefaultSettings: TypeAlias = Union[MainModelDefaultSettings, LoraModelDefaultSettings, ControlAdapterDefaultSettings]
 
 
 class ModelConfigFactory:

invokeai/backend/model_manager/legacy_probe.py

@@ -9,6 +9,7 @@ import spandrel
 import torch
 
 import invokeai.backend.util.logging as logger
+from invokeai.app.services.config.config_default import get_config
 from invokeai.app.util.misc import uuid_string
 from invokeai.backend.flux.controlnet.state_dict_utils import (
     is_state_dict_instantx_controlnet,
@@ -22,6 +23,7 @@ from invokeai.backend.model_manager.config import (
     AnyModelConfig,
     ControlAdapterDefaultSettings,
     InvalidModelConfigException,
+    LoraModelDefaultSettings,
     MainModelDefaultSettings,
     ModelConfigFactory,
     SubmodelDefinition,
@@ -216,6 +218,8 @@ class ModelProbe(object):
         if not fields["default_settings"]:
             if fields["type"] in {ModelType.ControlNet, ModelType.T2IAdapter, ModelType.ControlLoRa}:
                 fields["default_settings"] = get_default_settings_control_adapters(fields["name"])
+            if fields["type"] in {ModelType.LoRA}:
+                fields["default_settings"] = get_default_settings_lora()
             elif fields["type"] is ModelType.Main:
                 fields["default_settings"] = get_default_settings_main(fields["base"])
 
@@ -493,9 +497,21 @@ class ModelProbe(object):
         # scan model
         scan_result = pscan.scan_file_path(checkpoint)
         if scan_result.infected_files != 0:
-            raise Exception(f"The model {model_name} is potentially infected by malware. Aborting import.")
+            if get_config().unsafe_disable_picklescan:
+                logger.warning(
+                    f"The model {model_name} is potentially infected by malware, but picklescan is disabled. "
+                    "Proceeding with caution."
+                )
+            else:
+                raise RuntimeError(f"The model {model_name} is potentially infected by malware. Aborting import.")
         if scan_result.scan_err:
-            raise Exception(f"Error scanning model {model_name} for malware. Aborting import.")
+            if get_config().unsafe_disable_picklescan:
+                logger.warning(
+                    f"Error scanning the model at {model_name} for malware, but picklescan is disabled. "
+                    "Proceeding with caution."
+                )
+            else:
+                raise RuntimeError(f"Error scanning the model at {model_name} for malware. Aborting import.")
 
 
 # Probing utilities
@@ -530,6 +546,10 @@ def get_default_settings_control_adapters(model_name: str) -> Optional[ControlAd
     return None
 
 
+def get_default_settings_lora() -> LoraModelDefaultSettings:
+    return LoraModelDefaultSettings()
+
+
 def get_default_settings_main(model_base: BaseModelType) -> Optional[MainModelDefaultSettings]:
     if model_base is BaseModelType.StableDiffusion1 or model_base is BaseModelType.StableDiffusion2:
         return MainModelDefaultSettings(width=512, height=512)

invokeai/backend/model_manager/model_on_disk.py

@@ -6,13 +6,17 @@ import torch
 from picklescan.scanner import scan_file_path
 from safetensors import safe_open
 
+from invokeai.app.services.config.config_default import get_config
 from invokeai.backend.model_hash.model_hash import HASHING_ALGORITHMS, ModelHash
 from invokeai.backend.model_manager.taxonomy import ModelRepoVariant
 from invokeai.backend.quantization.gguf.loaders import gguf_sd_loader
+from invokeai.backend.util.logging import InvokeAILogger
 from invokeai.backend.util.silence_warnings import SilenceWarnings
 
 StateDict: TypeAlias = dict[str | int, Any]  # When are the keys int?
 
+logger = InvokeAILogger.get_logger()
+
 
 class ModelOnDisk:
     """A utility class representing a model stored on disk."""
@@ -79,8 +83,24 @@ class ModelOnDisk:
         with SilenceWarnings():
             if path.suffix.endswith((".ckpt", ".pt", ".pth", ".bin")):
                 scan_result = scan_file_path(path)
-                if scan_result.infected_files != 0 or scan_result.scan_err:
-                    raise RuntimeError(f"The model {path.stem} is potentially infected by malware. Aborting import.")
+                if scan_result.infected_files != 0:
+                    if get_config().unsafe_disable_picklescan:
+                        logger.warning(
+                            f"The model {path.stem} is potentially infected by malware, but picklescan is disabled. "
+                            "Proceeding with caution."
+                        )
+                    else:
+                        raise RuntimeError(
+                            f"The model {path.stem} is potentially infected by malware. Aborting import."
+                        )
+                if scan_result.scan_err:
+                    if get_config().unsafe_disable_picklescan:
+                        logger.warning(
+                            f"Error scanning the model at {path.stem} for malware, but picklescan is disabled. "
+                            "Proceeding with caution."
+                        )
+                    else:
+                        raise RuntimeError(f"Error scanning the model at {path.stem} for malware. Aborting import.")
                 checkpoint = torch.load(path, map_location="cpu")
                 assert isinstance(checkpoint, dict)
             elif path.suffix.endswith(".gguf"):

invokeai/backend/model_manager/starter_models.py

@@ -149,13 +149,29 @@ flux_kontext = StarterModel(
     dependencies=[t5_base_encoder, flux_vae, clip_l_encoder],
 )
 flux_kontext_quantized = StarterModel(
-    name="FLUX.1 Kontext dev (Quantized)",
+    name="FLUX.1 Kontext dev (quantized)",
     base=BaseModelType.Flux,
     source="https://huggingface.co/unsloth/FLUX.1-Kontext-dev-GGUF/resolve/main/flux1-kontext-dev-Q4_K_M.gguf",
     description="FLUX.1 Kontext dev quantized (q4_k_m). Total size with dependencies: ~14GB",
     type=ModelType.Main,
     dependencies=[t5_8b_quantized_encoder, flux_vae, clip_l_encoder],
 )
+flux_krea = StarterModel(
+    name="FLUX.1 Krea dev",
+    base=BaseModelType.Flux,
+    source="https://huggingface.co/InvokeAI/FLUX.1-Krea-dev/resolve/main/flux1-krea-dev.safetensors",
+    description="FLUX.1 Krea dev. Total size with dependencies: ~33GB",
+    type=ModelType.Main,
+    dependencies=[t5_8b_quantized_encoder, flux_vae, clip_l_encoder],
+)
+flux_krea_quantized = StarterModel(
+    name="FLUX.1 Krea dev (quantized)",
+    base=BaseModelType.Flux,
+    source="https://huggingface.co/InvokeAI/FLUX.1-Krea-dev-GGUF/resolve/main/flux1-krea-dev-Q4_K_M.gguf",
+    description="FLUX.1 Krea dev quantized (q4_k_m). Total size with dependencies: ~14GB",
+    type=ModelType.Main,
+    dependencies=[t5_8b_quantized_encoder, flux_vae, clip_l_encoder],
+)
 sd35_medium = StarterModel(
     name="SD3.5 Medium",
     base=BaseModelType.StableDiffusion3,
@@ -580,13 +596,14 @@ t2i_sketch_sdxl = StarterModel(
 )
 # endregion
 # region SpandrelImageToImage
-realesrgan_anime = StarterModel(
-    name="RealESRGAN_x4plus_anime_6B",
+animesharp_v4_rcan = StarterModel(
+    name="2x-AnimeSharpV4_RCAN",
     base=BaseModelType.Any,
-    source="https://github.com/xinntao/Real-ESRGAN/releases/download/v0.2.2.4/RealESRGAN_x4plus_anime_6B.pth",
-    description="A Real-ESRGAN 4x upscaling model (optimized for anime images).",
+    source="https://github.com/Kim2091/Kim2091-Models/releases/download/2x-AnimeSharpV4/2x-AnimeSharpV4_RCAN.safetensors",
+    description="A 2x upscaling model (optimized for anime images).",
     type=ModelType.SpandrelImageToImage,
 )
+
 realesrgan_x4 = StarterModel(
     name="RealESRGAN_x4plus",
     base=BaseModelType.Any,
@@ -732,7 +749,7 @@ STARTER_MODELS: list[StarterModel] = [
     t2i_lineart_sdxl,
     t2i_sketch_sdxl,
     realesrgan_x4,
-    realesrgan_anime,
+    animesharp_v4_rcan,
     realesrgan_x2,
     swinir,
     t5_base_encoder,
@@ -743,6 +760,8 @@ STARTER_MODELS: list[StarterModel] = [
     llava_onevision,
     flux_fill,
     cogview4,
+    flux_krea,
+    flux_krea_quantized,
 ]
 
 sd1_bundle: list[StarterModel] = [
@@ -794,6 +813,7 @@ flux_bundle: list[StarterModel] = [
     flux_redux,
     flux_fill,
     flux_kontext_quantized,
+    flux_krea_quantized,
 ]
 
 STARTER_BUNDLES: dict[str, StarterModelBundle] = {

invokeai/backend/model_manager/taxonomy.py

@@ -28,8 +28,11 @@ class BaseModelType(str, Enum):
     CogView4 = "cogview4"
     Imagen3 = "imagen3"
     Imagen4 = "imagen4"
+    Gemini2_5 = "gemini-2.5"
     ChatGPT4o = "chatgpt-4o"
     FluxKontext = "flux-kontext"
+    Veo3 = "veo3"
+    Runway = "runway"
 
 
 class ModelType(str, Enum):
@@ -51,6 +54,7 @@ class ModelType(str, Enum):
     SigLIP = "siglip"
     FluxRedux = "flux_redux"
     LlavaOnevision = "llava_onevision"
+    Video = "video"
 
 
 class SubModelType(str, Enum):

invokeai/backend/model_manager/util/lora_metadata_extractor.py

@@ -0,0 +1,145 @@
+"""Utility functions for extracting metadata from LoRA model files."""
+
+import json
+import logging
+from pathlib import Path
+from typing import Any, Dict, Optional, Set, Tuple
+
+from PIL import Image
+
+from invokeai.app.util.thumbnails import make_thumbnail
+from invokeai.backend.model_manager.config import AnyModelConfig, ModelType
+
+logger = logging.getLogger(__name__)
+
+
+def extract_lora_metadata(
+    model_path: Path, model_key: str, model_images_path: Path
+) -> Tuple[Optional[str], Optional[Set[str]]]:
+    """
+    Extract metadata for a LoRA model from associated JSON and image files.
+
+    Args:
+        model_path: Path to the LoRA model file
+        model_key: Unique key for the model
+        model_images_path: Path to the model images directory
+
+    Returns:
+        Tuple of (description, trigger_phrases)
+    """
+    model_stem = model_path.stem
+    model_dir = model_path.parent
+
+    # Find and process preview image
+    _process_preview_image(model_stem, model_dir, model_key, model_images_path)
+
+    # Extract metadata from JSON
+    description, trigger_phrases = _extract_json_metadata(model_stem, model_dir)
+
+    return description, trigger_phrases
+
+
+def _process_preview_image(model_stem: str, model_dir: Path, model_key: str, model_images_path: Path) -> bool:
+    """Find and process a preview image for the model, saving it to the model images store."""
+    image_extensions = [".png", ".jpg", ".jpeg", ".webp"]
+
+    for ext in image_extensions:
+        image_path = model_dir / f"{model_stem}{ext}"
+        if image_path.exists():
+            try:
+                # Open the image
+                with Image.open(image_path) as img:
+                    # Create thumbnail and save to model images directory
+                    thumbnail = make_thumbnail(img, 256)
+                    thumbnail_path = model_images_path / f"{model_key}.webp"
+                    thumbnail.save(thumbnail_path, format="webp")
+
+                logger.info(f"Processed preview image {image_path.name} for model {model_key}")
+                return True
+
+            except Exception as e:
+                logger.warning(f"Failed to process preview image {image_path.name}: {e}")
+                return False
+
+    return False
+
+
+def _extract_json_metadata(model_stem: str, model_dir: Path) -> Tuple[Optional[str], Optional[Set[str]]]:
+    """Extract metadata from a JSON file with the same name as the model."""
+    json_path = model_dir / f"{model_stem}.json"
+
+    if not json_path.exists():
+        return None, None
+
+    try:
+        with open(json_path, "r", encoding="utf-8") as f:
+            metadata = json.load(f)
+
+        # Extract description
+        description = _build_description(metadata)
+
+        # Extract trigger phrases
+        trigger_phrases = _extract_trigger_phrases(metadata)
+
+        if description or trigger_phrases:
+            logger.info(f"Applied metadata from {json_path.name}")
+
+        return description, trigger_phrases
+
+    except (json.JSONDecodeError, IOError, Exception) as e:
+        logger.warning(f"Failed to read metadata from {json_path}: {e}")
+        return None, None
+
+
+def _build_description(metadata: Dict[str, Any]) -> Optional[str]:
+    """Build a description from metadata fields."""
+    description_parts = []
+
+    if description := metadata.get("description"):
+        description_parts.append(str(description).strip())
+
+    if notes := metadata.get("notes"):
+        description_parts.append(str(notes).strip())
+
+    return " | ".join(description_parts) if description_parts else None
+
+
+def _extract_trigger_phrases(metadata: Dict[str, Any]) -> Optional[Set[str]]:
+    """Extract trigger phrases from metadata."""
+    if not (activation_text := metadata.get("activation text")):
+        return None
+
+    activation_text = str(activation_text).strip()
+    if not activation_text:
+        return None
+
+    # Split on commas and clean up each phrase
+    phrases = [phrase.strip() for phrase in activation_text.split(",") if phrase.strip()]
+
+    return set(phrases) if phrases else None
+
+
+def apply_lora_metadata(info: AnyModelConfig, model_path: Path, model_images_path: Path) -> None:
+    """
+    Apply extracted metadata to a LoRA model configuration.
+
+    Args:
+        info: The model configuration to update
+        model_path: Path to the LoRA model file
+        model_images_path: Path to the model images directory
+    """
+    # Only process LoRA models
+    if info.type != ModelType.LoRA:
+        return
+
+    # Extract and apply metadata
+    description, trigger_phrases = extract_lora_metadata(model_path, info.key, model_images_path)
+
+    # We don't set cover_image path in the config anymore since images are stored
+    # separately in the model images store by model key
+
+    if description:
+        info.description = description
+
+    if trigger_phrases:
+        info.trigger_phrases = trigger_phrases

invokeai/backend/model_manager/util/model_util.py

@@ -8,8 +8,12 @@ import picklescan.scanner as pscan
 import safetensors
 import torch
 
+from invokeai.app.services.config.config_default import get_config
 from invokeai.backend.model_manager.taxonomy import ClipVariantType
 from invokeai.backend.quantization.gguf.loaders import gguf_sd_loader
+from invokeai.backend.util.logging import InvokeAILogger
+
+logger = InvokeAILogger.get_logger()
 
 
 def _fast_safetensors_reader(path: str) -> Dict[str, torch.Tensor]:
@@ -59,9 +63,21 @@ def read_checkpoint_meta(path: Union[str, Path], scan: bool = True) -> Dict[str,
         if scan:
             scan_result = pscan.scan_file_path(path)
             if scan_result.infected_files != 0:
-                raise Exception(f"The model at {path} is potentially infected by malware. Aborting import.")
+                if get_config().unsafe_disable_picklescan:
+                    logger.warning(
+                        f"The model {path} is potentially infected by malware, but picklescan is disabled. "
+                        "Proceeding with caution."
+                    )
+                else:
+                    raise RuntimeError(f"The model {path} is potentially infected by malware. Aborting import.")
             if scan_result.scan_err:
-                raise Exception(f"Error scanning model at {path} for malware. Aborting import.")
+                if get_config().unsafe_disable_picklescan:
+                    logger.warning(
+                        f"Error scanning the model at {path} for malware, but picklescan is disabled. "
+                        "Proceeding with caution."
+                    )
+                else:
+                    raise RuntimeError(f"Error scanning the model at {path} for malware. Aborting import.")
 
         checkpoint = torch.load(path, map_location=torch.device("meta"))
     return checkpoint

invokeai/backend/patches/lora_conversions/flux_diffusers_lora_conversion_utils.py

@@ -18,16 +18,25 @@ def is_state_dict_likely_in_flux_diffusers_format(state_dict: Dict[str, torch.Te
     # First, check that all keys end in "lora_A.weight" or "lora_B.weight" (i.e. are in PEFT format).
     all_keys_in_peft_format = all(k.endswith(("lora_A.weight", "lora_B.weight")) for k in state_dict.keys())
 
-    # Next, check that this is likely a FLUX model by spot-checking a few keys.
-    expected_keys = [
+    # Check if keys use transformer prefix
+    transformer_prefix_keys = [
         "transformer.single_transformer_blocks.0.attn.to_q.lora_A.weight",
         "transformer.single_transformer_blocks.0.attn.to_q.lora_B.weight",
         "transformer.transformer_blocks.0.attn.add_q_proj.lora_A.weight",
         "transformer.transformer_blocks.0.attn.add_q_proj.lora_B.weight",
     ]
-    all_expected_keys_present = all(k in state_dict for k in expected_keys)
+    transformer_keys_present = all(k in state_dict for k in transformer_prefix_keys)
 
-    return all_keys_in_peft_format and all_expected_keys_present
+    # Check if keys use base_model.model prefix
+    base_model_prefix_keys = [
+        "base_model.model.single_transformer_blocks.0.attn.to_q.lora_A.weight",
+        "base_model.model.single_transformer_blocks.0.attn.to_q.lora_B.weight",
+        "base_model.model.transformer_blocks.0.attn.add_q_proj.lora_A.weight",
+        "base_model.model.transformer_blocks.0.attn.add_q_proj.lora_B.weight",
+    ]
+    base_model_keys_present = all(k in state_dict for k in base_model_prefix_keys)
+
+    return all_keys_in_peft_format and (transformer_keys_present or base_model_keys_present)
 
 
 def lora_model_from_flux_diffusers_state_dict(
@@ -49,8 +58,16 @@ def lora_layers_from_flux_diffusers_grouped_state_dict(
     https://github.com/huggingface/diffusers/blob/55ac421f7bb12fd00ccbef727be4dc2f3f920abb/scripts/convert_flux_to_diffusers.py
     """
 
-    # Remove the "transformer." prefix from all keys.
-    grouped_state_dict = {k.replace("transformer.", ""): v for k, v in grouped_state_dict.items()}
+    # Determine which prefix is used and remove it from all keys.
+    # Check if any key starts with "base_model.model." prefix
+    has_base_model_prefix = any(k.startswith("base_model.model.") for k in grouped_state_dict.keys())
+
+    if has_base_model_prefix:
+        # Remove the "base_model.model." prefix from all keys.
+        grouped_state_dict = {k.replace("base_model.model.", ""): v for k, v in grouped_state_dict.items()}
+    else:
+        # Remove the "transformer." prefix from all keys.
+        grouped_state_dict = {k.replace("transformer.", ""): v for k, v in grouped_state_dict.items()}
 
     # Constants for FLUX.1
     num_double_layers = 19

invokeai/backend/quantization/scripts/load_flux_model_bnb_llm_int8.py

@@ -20,7 +20,7 @@ def main():
         "/data/invokeai/models/.download_cache/https__huggingface.co_black-forest-labs_flux.1-schnell_resolve_main_flux1-schnell.safetensors/flux1-schnell.safetensors"
     )
 
-    with log_time("Intialize FLUX transformer on meta device"):
+    with log_time("Initialize FLUX transformer on meta device"):
         # TODO(ryand): Determine if this is a schnell model or a dev model and load the appropriate config.
         p = params["flux-schnell"]
 

invokeai/backend/quantization/scripts/load_flux_model_bnb_nf4.py

@@ -33,7 +33,7 @@ def main():
     )
 
     # inference_dtype = torch.bfloat16
-    with log_time("Intialize FLUX transformer on meta device"):
+    with log_time("Initialize FLUX transformer on meta device"):
         # TODO(ryand): Determine if this is a schnell model or a dev model and load the appropriate config.
         p = params["flux-schnell"]
 

invokeai/backend/quantization/scripts/quantize_t5_xxl_bnb_llm_int8.py

@@ -27,7 +27,7 @@ def main():
     """
     model_path = Path("/data/misc/text_encoder_2")
 
-    with log_time("Intialize T5 on meta device"):
+    with log_time("Initialize T5 on meta device"):
         model_config = AutoConfig.from_pretrained(model_path)
         with accelerate.init_empty_weights():
             model = AutoModelForTextEncoding.from_config(model_config)

invokeai/backend/util/vae_working_memory.py

@@ -0,0 +1,117 @@
+from typing import Literal
+
+import torch
+from diffusers.models.autoencoders.autoencoder_kl import AutoencoderKL
+from diffusers.models.autoencoders.autoencoder_tiny import AutoencoderTiny
+
+from invokeai.app.invocations.constants import LATENT_SCALE_FACTOR
+from invokeai.backend.flux.modules.autoencoder import AutoEncoder
+
+
+def estimate_vae_working_memory_sd15_sdxl(
+    operation: Literal["encode", "decode"],
+    image_tensor: torch.Tensor,
+    vae: AutoencoderKL | AutoencoderTiny,
+    tile_size: int | None,
+    fp32: bool,
+) -> int:
+    """Estimate the working memory required to encode or decode the given tensor."""
+    # It was found experimentally that the peak working memory scales linearly with the number of pixels and the
+    # element size (precision). This estimate is accurate for both SD1 and SDXL.
+    element_size = 4 if fp32 else 2
+
+    # This constant is determined experimentally and takes into consideration both allocated and reserved memory. See #8414
+    # Encoding uses ~45% the working memory as decoding.
+    scaling_constant = 2200 if operation == "decode" else 1100
+
+    latent_scale_factor_for_operation = LATENT_SCALE_FACTOR if operation == "decode" else 1
+
+    if tile_size is not None:
+        if tile_size == 0:
+            tile_size = vae.tile_sample_min_size
+            assert isinstance(tile_size, int)
+        h = tile_size
+        w = tile_size
+        working_memory = h * w * element_size * scaling_constant
+
+        # We add 25% to the working memory estimate when tiling is enabled to account for factors like tile overlap
+        # and number of tiles. We could make this more precise in the future, but this should be good enough for
+        # most use cases.
+        working_memory = working_memory * 1.25
+    else:
+        h = latent_scale_factor_for_operation * image_tensor.shape[-2]
+        w = latent_scale_factor_for_operation * image_tensor.shape[-1]
+        working_memory = h * w * element_size * scaling_constant
+
+    if fp32:
+        # If we are running in FP32, then we should account for the likely increase in model size (~250MB).
+        working_memory += 250 * 2**20
+
+    print(f"estimate_vae_working_memory_sd15_sdxl: {int(working_memory)}")
+
+    return int(working_memory)
+
+
+def estimate_vae_working_memory_cogview4(
+    operation: Literal["encode", "decode"], image_tensor: torch.Tensor, vae: AutoencoderKL
+) -> int:
+    """Estimate the working memory required by the invocation in bytes."""
+    latent_scale_factor_for_operation = LATENT_SCALE_FACTOR if operation == "decode" else 1
+
+    h = latent_scale_factor_for_operation * image_tensor.shape[-2]
+    w = latent_scale_factor_for_operation * image_tensor.shape[-1]
+    element_size = next(vae.parameters()).element_size()
+
+    # This constant is determined experimentally and takes into consideration both allocated and reserved memory. See #8414
+    # Encoding uses ~45% the working memory as decoding.
+    scaling_constant = 2200 if operation == "decode" else 1100
+    working_memory = h * w * element_size * scaling_constant
+
+    print(f"estimate_vae_working_memory_cogview4: {int(working_memory)}")
+
+    return int(working_memory)
+
+
+def estimate_vae_working_memory_flux(
+    operation: Literal["encode", "decode"], image_tensor: torch.Tensor, vae: AutoEncoder
+) -> int:
+    """Estimate the working memory required by the invocation in bytes."""
+
+    latent_scale_factor_for_operation = LATENT_SCALE_FACTOR if operation == "decode" else 1
+
+    out_h = latent_scale_factor_for_operation * image_tensor.shape[-2]
+    out_w = latent_scale_factor_for_operation * image_tensor.shape[-1]
+    element_size = next(vae.parameters()).element_size()
+
+    # This constant is determined experimentally and takes into consideration both allocated and reserved memory. See #8414
+    # Encoding uses ~45% the working memory as decoding.
+    scaling_constant = 2200 if operation == "decode" else 1100
+
+    working_memory = out_h * out_w * element_size * scaling_constant
+
+    print(f"estimate_vae_working_memory_flux: {int(working_memory)}")
+
+    return int(working_memory)
+
+
+def estimate_vae_working_memory_sd3(
+    operation: Literal["encode", "decode"], image_tensor: torch.Tensor, vae: AutoencoderKL
+) -> int:
+    """Estimate the working memory required by the invocation in bytes."""
+    # Encode operations use approximately 50% of the memory required for decode operations
+
+    latent_scale_factor_for_operation = LATENT_SCALE_FACTOR if operation == "decode" else 1
+
+    h = latent_scale_factor_for_operation * image_tensor.shape[-2]
+    w = latent_scale_factor_for_operation * image_tensor.shape[-1]
+    element_size = next(vae.parameters()).element_size()
+
+    # This constant is determined experimentally and takes into consideration both allocated and reserved memory. See #8414
+    # Encoding uses ~45% the working memory as decoding.
+    scaling_constant = 2200 if operation == "decode" else 1100
+
+    working_memory = h * w * element_size * scaling_constant
+
+    print(f"estimate_vae_working_memory_sd3: {int(working_memory)}")
+
+    return int(working_memory)

invokeai/frontend/web/.eslintignore

@@ -1,10 +0,0 @@
-dist/
-static/
-.husky/
-node_modules/
-patches/
-stats.html
-index.html
-.yarn/
-*.scss
-src/services/api/schema.ts

invokeai/frontend/web/.eslintrc.js

@@ -1,88 +0,0 @@
-module.exports = {
-  extends: ['@invoke-ai/eslint-config-react'],
-  plugins: ['path', 'i18next'],
-  rules: {
-    // TODO(psyche): Enable this rule. Requires no default exports in components - many changes.
-    'react-refresh/only-export-components': 'off',
-    // TODO(psyche): Enable this rule. Requires a lot of eslint-disable-next-line comments.
-    '@typescript-eslint/consistent-type-assertions': 'off',
-    // https://github.com/qdanik/eslint-plugin-path
-    'path/no-relative-imports': ['error', { maxDepth: 0 }],
-    // https://github.com/edvardchen/eslint-plugin-i18next/blob/HEAD/docs/rules/no-literal-string.md
-    // TODO: ENABLE THIS RULE BEFORE v6.0.0
-    // 'i18next/no-literal-string': 'error',
-    // https://eslint.org/docs/latest/rules/no-console
-    'no-console': 'warn',
-    // https://eslint.org/docs/latest/rules/no-promise-executor-return
-    'no-promise-executor-return': 'error',
-    // https://eslint.org/docs/latest/rules/require-await
-    'require-await': 'error',
-    // Restrict setActiveTab calls to only use-navigation-api.tsx
-    'no-restricted-syntax': [
-      'error',
-      {
-        selector: 'CallExpression[callee.name="setActiveTab"]',
-        message:
-          'setActiveTab() can only be called from use-navigation-api.tsx. Use navigationApi.switchToTab() instead.',
-      },
-    ],
-    // TODO: ENABLE THIS RULE BEFORE v6.0.0
-    'react/display-name': 'off',
-    'no-restricted-properties': [
-      'error',
-      {
-        object: 'crypto',
-        property: 'randomUUID',
-        message: 'Use of crypto.randomUUID is not allowed as it is not available in all browsers.',
-      },
-      {
-        object: 'navigator',
-        property: 'clipboard',
-        message:
-          'The Clipboard API is not available by default in Firefox. Use the `useClipboard` hook instead, which wraps clipboard access to prevent errors.',
-      },
-    ],
-    'no-restricted-imports': [
-      'error',
-      {
-        paths: [
-          {
-            name: 'lodash-es',
-            importNames: ['isEqual'],
-            message: 'Please use objectEquals from @observ33r/object-equals instead.',
-          },
-          {
-            name: 'lodash-es',
-            message: 'Please use es-toolkit instead.',
-          },
-          {
-            name: 'es-toolkit',
-            importNames: ['isEqual'],
-            message: 'Please use objectEquals from @observ33r/object-equals instead.',
-          },
-        ],
-      },
-    ],
-  },
-  overrides: [
-    /**
-     * Allow setActiveTab calls only in use-navigation-api.tsx
-     */
-    {
-      files: ['**/use-navigation-api.tsx'],
-      rules: {
-        'no-restricted-syntax': 'off',
-      },
-    },
-    /**
-     * Overrides for stories
-     */
-    {
-      files: ['*.stories.tsx'],
-      rules: {
-        // We may not have i18n available in stories.
-        'i18next/no-literal-string': 'off',
-      },
-    },
-  ],
-};

invokeai/frontend/web/.gitignore

@@ -44,4 +44,5 @@ yalc.lock
 
 # vitest
 tsconfig.vitest-temp.json
-coverage/
+coverage/
+*.tgz

invokeai/frontend/web/.prettierignore

@@ -14,3 +14,4 @@ static/
 src/theme/css/overlayscrollbars.css
 src/theme_/css/overlayscrollbars.css
 pnpm-lock.yaml
+.claude

invokeai/frontend/web/.prettierrc.js

@@ -1,11 +0,0 @@
-module.exports = {
-  ...require('@invoke-ai/prettier-config-react'),
-  overrides: [
-    {
-      files: ['public/locales/*.json'],
-      options: {
-        tabWidth: 4,
-      },
-    },
-  ],
-};