fix(ui): typo

Currently translated at 100.0% (1802 of 1802 strings)

Co-authored-by: Linos <linos.coding@gmail.com>
Translate-URL: https://hosted.weblate.org/projects/invokeai/web-ui/vi/
Translation: InvokeAI/Web UI

.github/workflows/build-container.yml

@@ -76,9 +76,6 @@ jobs:
             latest=${{ matrix.gpu-driver == 'cuda' && github.ref == 'refs/heads/main' }}
             suffix=-${{ matrix.gpu-driver }},onlatest=false
 
-      - name: Set up QEMU
-        uses: docker/setup-qemu-action@v3
-
       - name: Set up Docker Buildx
         uses: docker/setup-buildx-action@v3
         with:
@@ -103,7 +100,7 @@ jobs:
           push: ${{ github.ref == 'refs/heads/main' || github.ref_type == 'tag' || github.event.inputs.push-to-registry }}
           tags: ${{ steps.meta.outputs.tags }}
           labels: ${{ steps.meta.outputs.labels }}
-          cache-from: |
-            type=gha,scope=${{ github.ref_name }}-${{ matrix.gpu-driver }}
-            type=gha,scope=main-${{ matrix.gpu-driver }}
-          cache-to: type=gha,mode=max,scope=${{ github.ref_name }}-${{ matrix.gpu-driver }}
+          # cache-from: |
+          #   type=gha,scope=${{ github.ref_name }}-${{ matrix.gpu-driver }}
+          #   type=gha,scope=main-${{ matrix.gpu-driver }}
+          # cache-to: type=gha,mode=max,scope=${{ github.ref_name }}-${{ matrix.gpu-driver }}

.nvmrc

@@ -0,0 +1 @@
+v22.12.0

docker/Dockerfile

@@ -13,52 +13,67 @@ RUN --mount=type=cache,target=/var/cache/apt,sharing=locked \
         git
 
 # Install `uv` for package management
-COPY --from=ghcr.io/astral-sh/uv:0.5.5 /uv /uvx /bin/
+COPY --from=ghcr.io/astral-sh/uv:0.6.0 /uv /uvx /bin/
 
 ENV VIRTUAL_ENV=/opt/venv
 ENV PATH="$VIRTUAL_ENV/bin:$PATH"
 ENV INVOKEAI_SRC=/opt/invokeai
 ENV PYTHON_VERSION=3.11
+ENV UV_PYTHON=3.11
 ENV UV_COMPILE_BYTECODE=1
 ENV UV_LINK_MODE=copy
+ENV UV_PROJECT_ENVIRONMENT="$VIRTUAL_ENV"
+ENV UV_INDEX="https://download.pytorch.org/whl/cu124"
 
 ARG GPU_DRIVER=cuda
-ARG TARGETPLATFORM="linux/amd64"
 # unused but available
 ARG BUILDPLATFORM
 
 # Switch to the `ubuntu` user to work around dependency issues with uv-installed python
 RUN mkdir -p ${VIRTUAL_ENV} && \
     mkdir -p ${INVOKEAI_SRC} && \
-    chmod -R a+w /opt
+    chmod -R a+w /opt && \
+    mkdir ~ubuntu/.cache && chown ubuntu: ~ubuntu/.cache
 USER ubuntu
 
-# Install python and create the venv
-RUN uv python install ${PYTHON_VERSION} && \
-    uv venv --relocatable --prompt "invoke" --python ${PYTHON_VERSION} ${VIRTUAL_ENV}
+# Install python
+RUN --mount=type=cache,target=/home/ubuntu/.cache/uv,uid=1000,gid=1000 \
+  uv python install ${PYTHON_VERSION}
 
 WORKDIR ${INVOKEAI_SRC}
-COPY invokeai ./invokeai
-COPY pyproject.toml ./
 
-# Editable mode helps use the same image for development:
-# the local working copy can be bind-mounted into the image
-# at path defined by ${INVOKEAI_SRC}
+# Install project's dependencies as a separate layer so they aren't rebuilt every commit.
+# bind-mount instead of copy to defer adding sources to the image until next layer.
+#
 # NOTE: there are no pytorch builds for arm64 + cuda, only cpu
 # x86_64/CUDA is the default
 RUN --mount=type=cache,target=/home/ubuntu/.cache/uv,uid=1000,gid=1000 \
+    --mount=type=bind,source=pyproject.toml,target=pyproject.toml \
+    --mount=type=bind,source=invokeai/version,target=invokeai/version \
     if [ "$TARGETPLATFORM" = "linux/arm64" ] || [ "$GPU_DRIVER" = "cpu" ]; then \
-        extra_index_url_arg="--extra-index-url https://download.pytorch.org/whl/cpu"; \
+      UV_INDEX="https://download.pytorch.org/whl/cpu"; \
     elif [ "$GPU_DRIVER" = "rocm" ]; then \
-        extra_index_url_arg="--extra-index-url https://download.pytorch.org/whl/rocm6.1"; \
-    else \
-        extra_index_url_arg="--extra-index-url https://download.pytorch.org/whl/cu124"; \
+      UV_INDEX="https://download.pytorch.org/whl/rocm6.1"; \
     fi && \
-    uv pip install --python ${PYTHON_VERSION} $extra_index_url_arg -e "."
+    uv sync --no-install-project
+
+# Now that the bulk of the dependencies have been installed, copy in the project files that change more frequently.
+COPY invokeai invokeai
+COPY pyproject.toml .
+
+RUN --mount=type=cache,target=/home/ubuntu/.cache/uv,uid=1000,gid=1000 \
+    --mount=type=bind,source=pyproject.toml,target=pyproject.toml \
+    if [ "$TARGETPLATFORM" = "linux/arm64" ] || [ "$GPU_DRIVER" = "cpu" ]; then \
+      UV_INDEX="https://download.pytorch.org/whl/cpu"; \
+    elif [ "$GPU_DRIVER" = "rocm" ]; then \
+      UV_INDEX="https://download.pytorch.org/whl/rocm6.1"; \
+    fi && \
+    uv sync
+
 
 #### Build the Web UI ------------------------------------
 
-FROM node:20-slim AS web-builder
+FROM docker.io/node:22-slim AS web-builder
 ENV PNPM_HOME="/pnpm"
 ENV PATH="$PNPM_HOME:$PATH"
 RUN corepack use pnpm@8.x
@@ -98,6 +113,7 @@ RUN apt update && apt install -y --no-install-recommends \
 
 ENV INVOKEAI_SRC=/opt/invokeai
 ENV VIRTUAL_ENV=/opt/venv
+ENV UV_PROJECT_ENVIRONMENT="$VIRTUAL_ENV"
 ENV PYTHON_VERSION=3.11
 ENV INVOKEAI_ROOT=/invokeai
 ENV INVOKEAI_HOST=0.0.0.0
@@ -109,7 +125,7 @@ ENV CONTAINER_GID=${CONTAINER_GID:-1000}
 # Install `uv` for package management
 # and install python for the ubuntu user (expected to exist on ubuntu >=24.x)
 # this is too tiny to optimize with multi-stage builds, but maybe we'll come back to it
-COPY --from=ghcr.io/astral-sh/uv:0.5.5 /uv /uvx /bin/
+COPY --from=ghcr.io/astral-sh/uv:0.6.0 /uv /uvx /bin/
 USER ubuntu
 RUN uv python install ${PYTHON_VERSION}
 USER root

docs/RELEASE.md

@@ -1,41 +1,50 @@
 # Release Process
 
-The app is published in twice, in different build formats.
+The Invoke application is published as a python package on [PyPI]. This includes both a source distribution and built distribution (a wheel).
 
-- A [PyPI] distribution. This includes both a source distribution and built distribution (a wheel). Users install with `pip install invokeai`. The updater uses this build.
-- An installer on the [InvokeAI Releases Page]. This is a zip file with install scripts and a wheel. This is only used for new installs.
+Most users install it with the [Launcher](https://github.com/invoke-ai/launcher/), others with `pip`.
+
+The launcher uses GitHub as the source of truth for available releases.
+
+## Broad Strokes
+
+- Merge all changes and bump the version in the codebase.
+- Tag the release commit.
+- Wait for the release workflow to complete.
+- Approve the PyPI publish jobs.
+- Write GH release notes.
 
 ## General Prep
 
-Make a developer call-out for PRs to merge. Merge and test things out.
-
-While the release workflow does not include end-to-end tests, it does pause before publishing so you can download and test the final build.
+Make a developer call-out for PRs to merge. Merge and test things out. Bump the version by editing `invokeai/version/invokeai_version.py`.
 
 ## Release Workflow
 
 The `release.yml` workflow runs a number of jobs to handle code checks, tests, build and publish on PyPI.
 
-It is triggered on **tag push**, when the tag matches `v*`. It doesn't matter if you've prepped a release branch like `release/v3.5.0` or are releasing from `main` - it works the same.
-
-> Because commits are reference-counted, it is safe to create a release branch, tag it, let the workflow run, then delete the branch. So long as the tag exists, that commit will exist.
+It is triggered on **tag push**, when the tag matches `v*`.
 
 ### Triggering the Workflow
 
-Run `make tag-release` to tag the current commit and kick off the workflow.
+Ensure all commits that should be in the release are merged, and you have pulled them locally.
 
-The release may also be dispatched [manually].
+Double-check that you have checked out the commit that will represent the release (typically the latest commit on `main`).
+
+Run `make tag-release` to tag the current commit and kick off the workflow. You will be prompted to provide a message - use the version specifier.
+
+If this version's tag already exists for some reason (maybe you had to make a last minute change), the script will overwrite it.
+
+> In case you cannot use the Make target, the release may also be dispatched [manually] via GH.
 
 ### Workflow Jobs and Process
 
-The workflow consists of a number of concurrently-run jobs, and two final publish jobs.
+The workflow consists of a number of concurrently-run checks and tests, then two final publish jobs.
 
 The publish jobs require manual approval and are only run if the other jobs succeed.
 
 #### `check-version` Job
 
-This job checks that the git ref matches the app version. It matches the ref against the `__version__` variable in `invokeai/version/invokeai_version.py`.
-
-When the workflow is triggered by tag push, the ref is the tag. If the workflow is run manually, the ref is the target selected from the **Use workflow from** dropdown.
+This job ensures that the `invokeai` python package version specifier matches the tag for the release. The version specifier is pulled from the `__version__` variable in `invokeai/version/invokeai_version.py`.
 
 This job uses [samuelcolvin/check-python-version].
 
@@ -43,62 +52,52 @@ This job uses [samuelcolvin/check-python-version].
 
 #### Check and Test Jobs
 
+Next, these jobs run and must pass. They are the same jobs that are run for every PR.
+
 - **`python-tests`**: runs `pytest` on matrix of platforms
 - **`python-checks`**: runs `ruff` (format and lint)
 - **`frontend-tests`**: runs `vitest`
 - **`frontend-checks`**: runs `prettier` (format), `eslint` (lint), `dpdm` (circular refs), `tsc` (static type check) and `knip` (unused imports)
-
-> **TODO** We should add `mypy` or `pyright` to the **`check-python`** job.
-
-> **TODO** We should add an end-to-end test job that generates an image.
+- **`typegen-checks`**: ensures the frontend and backend types are synced
 
 #### `build-installer` Job
 
 This sets up both python and frontend dependencies and builds the python package. Internally, this runs `installer/create_installer.sh` and uploads two artifacts:
 
 - **`dist`**: the python distribution, to be published on PyPI
-- **`InvokeAI-installer-${VERSION}.zip`**: the installer to be included in the GitHub release
+- **`InvokeAI-installer-${VERSION}.zip`**: the legacy install scripts
+
+You don't need to download either of these files.
+
+> The legacy install scripts are no longer used, but we haven't updated the workflow to skip building them.
 
 #### Sanity Check & Smoke Test
 
-At this point, the release workflow pauses as the remaining publish jobs require approval. Time to test the installer.
+At this point, the release workflow pauses as the remaining publish jobs require approval.
 
-Because the installer pulls from PyPI, and we haven't published to PyPI yet, you will need to install from the wheel:
+It's possible to test the python package before it gets published to PyPI. We've never had problems with it, so it's not necessary to do this.
 
-- Download and unzip `dist.zip` and the installer from the **Summary** tab of the workflow
-- Run the installer script using the `--wheel` CLI arg, pointing at the wheel:
+But, if you want to be extra-super careful, here's how to test it:
 
-  ```sh
-  ./install.sh --wheel ../InvokeAI-4.0.0rc6-py3-none-any.whl
-  ```
-
-- Install to a temporary directory so you get the new user experience
-- Download a model and generate
-
-> The same wheel file is bundled in the installer and in the `dist` artifact, which is uploaded to PyPI. You should end up with the exactly the same installation as if the installer got the wheel from PyPI.
+- Download the `dist.zip` build artifact from the `build-installer` job
+- Unzip it and find the wheel file
+- Create a fresh Invoke install by following the [manual install guide](https://invoke-ai.github.io/InvokeAI/installation/manual/) - but instead of installing from PyPI, install from the wheel
+- Test the app
 
 ##### Something isn't right
 
-If testing reveals any issues, no worries. Cancel the workflow, which will cancel the pending publish jobs (you didn't approve them prematurely, right?).
-
-Now you can start from the top:
-
-- Fix the issues and PR the fixes per usual
-- Get the PR approved and merged per usual
-- Switch to `main` and pull in the fixes
-- Run `make tag-release` to move the tag to `HEAD` (which has the fixes) and kick off the release workflow again
-- Re-do the sanity check
+If testing reveals any issues, no worries. Cancel the workflow, which will cancel the pending publish jobs (you didn't approve them prematurely, right?) and start over.
 
 #### PyPI Publish Jobs
 
-The publish jobs will run if any of the previous jobs fail.
+The publish jobs will not run if any of the previous jobs fail.
 
 They use [GitHub environments], which are configured as [trusted publishers] on PyPI.
 
-Both jobs require a maintainer to approve them from the workflow's **Summary** tab.
+Both jobs require a @hipsterusername or @psychedelicious to approve them from the workflow's **Summary** tab.
 
 - Click the **Review deployments** button
-- Select the environment (either `testpypi` or `pypi`)
+- Select the environment (either `testpypi` or `pypi` - typically you select both)
 - Click **Approve and deploy**
 
 > **If the version already exists on PyPI, the publish jobs will fail.** PyPI only allows a given version to be published once - you cannot change it. If version published on PyPI has a problem, you'll need to "fail forward" by bumping the app version and publishing a followup release.
@@ -113,46 +112,33 @@ If there are no incidents, contact @hipsterusername or @lstein, who have owner a
 
 Publishes the distribution on the [Test PyPI] index, using the `testpypi` GitHub environment.
 
-This job is not required for the production PyPI publish, but included just in case you want to test the PyPI release.
+This job is not required for the production PyPI publish, but included just in case you want to test the PyPI release for some reason:
 
-If approved and successful, you could try out the test release like this:
-
-```sh
-# Create a new virtual environment
-python -m venv ~/.test-invokeai-dist --prompt test-invokeai-dist
-# Install the distribution from Test PyPI
-pip install --index-url https://test.pypi.org/simple/ invokeai
-# Run and test the app
-invokeai-web
-# Cleanup
-deactivate
-rm -rf ~/.test-invokeai-dist
-```
+- Approve this publish job without approving the prod publish
+- Let it finish
+- Create a fresh Invoke install by following the [manual install guide](https://invoke-ai.github.io/InvokeAI/installation/manual/), making sure to use the Test PyPI index URL: `https://test.pypi.org/simple/`
+- Test the app
 
 #### `publish-pypi` Job
 
 Publishes the distribution on the production PyPI index, using the `pypi` GitHub environment.
 
-## Publish the GitHub Release with installer
+It's a good idea to wait to approve and run this job until you have the release notes ready!
 
-Once the release is published to PyPI, it's time to publish the GitHub release.
+## Prep and publish the GitHub Release
 
 1. [Draft a new release] on GitHub, choosing the tag that triggered the release.
-1. Write the release notes, describing important changes. The **Generate release notes** button automatically inserts the changelog and new contributors, and you can copy/paste the intro from previous releases.
-1. Use `scripts/get_external_contributions.py` to get a list of external contributions to shout out in the release notes.
-1. Upload the zip file created in **`build`** job into the Assets section of the release notes.
-1. Check **Set as a pre-release** if it's a pre-release.
-1. Check **Create a discussion for this release**.
-1. Publish the release.
-1. Announce the release in Discord.
-
-> **TODO** Workflows can create a GitHub release from a template and upload release assets. One popular action to handle this is [ncipollo/release-action]. A future enhancement to the release process could set this up.
-
-## Manual Build
-
-The `build installer` workflow can be dispatched manually. This is useful to test the installer for a given branch or tag.
-
-No checks are run, it just builds.
+2. The **Generate release notes** button automatically inserts the changelog and new contributors. Make sure to select the correct tags for this release and the last stable release. GH often selects the wrong tags - do this manually.
+3. Write the release notes, describing important changes. Contributions from community members should be shouted out. Use the GH-generated changelog to see all contributors. If there are Weblate translation updates, open that PR and shout out every person who contributed a translation.
+4. Check **Set as a pre-release** if it's a pre-release.
+5. Approve and wait for the `publish-pypi` job to finish if you haven't already.
+6. Publish the GH release.
+7. Post the release in Discord in the [releases](https://discord.com/channels/1020123559063990373/1149260708098359327) channel with abbreviated notes. For example:
+   > Invoke v5.7.0 (stable): <https://github.com/invoke-ai/InvokeAI/releases/tag/v5.7.0>
+   >
+   > It's a pretty big one - Form Builder, Metadata Nodes (thanks @SkunkWorxDark!), and much more.
+8. Right click the message in releases and copy the link to it. Then, post that link in the [new-release-discussion](https://discord.com/channels/1020123559063990373/1149506274971631688) channel. For example:
+   > Invoke v5.7.0 (stable): <https://discord.com/channels/1020123559063990373/1149260708098359327/1344521744916021248>
 
 ## Manual Release
 
@@ -160,12 +146,10 @@ The `release` workflow can be dispatched manually. You must dispatch the workflo
 
 This functionality is available as a fallback in case something goes wonky. Typically, releases should be triggered via tag push as described above.
 
-[InvokeAI Releases Page]: https://github.com/invoke-ai/InvokeAI/releases
 [PyPI]: https://pypi.org/
 [Draft a new release]: https://github.com/invoke-ai/InvokeAI/releases/new
 [Test PyPI]: https://test.pypi.org/
 [version specifier]: https://packaging.python.org/en/latest/specifications/version-specifiers/
-[ncipollo/release-action]: https://github.com/ncipollo/release-action
 [GitHub environments]: https://docs.github.com/en/actions/deployment/targeting-different-environments/using-environments-for-deployment
 [trusted publishers]: https://docs.pypi.org/trusted-publishers/
 [samuelcolvin/check-python-version]: https://github.com/samuelcolvin/check-python-version

docs/faq.md

@@ -1,26 +1,18 @@
 # FAQ
 
-!!! info "How to Reinstall"
-
-    Many issues can be resolved by re-installing the application. You won't lose any data by re-installing. We suggest downloading the [latest release](https://github.com/invoke-ai/InvokeAI/releases/latest) and using it to re-install the application. Consult the [installer guide](./installation/installer.md) for more information.
-
-    When you run the installer, you'll have an option to select the version to install. If you aren't ready to upgrade, you choose the current version to fix a broken install.
-
 If the troubleshooting steps on this page don't get you up and running, please either [create an issue] or hop on [discord] for help.
 
 ## How to Install
 
-You can download the latest installers [here](https://github.com/invoke-ai/InvokeAI/releases).
-
-Note that any releases marked as _pre-release_ are in a beta state. You may experience some issues, but we appreciate your help testing those! For stable/reliable installations, please install the [latest release].
+Follow the [Quick Start guide](./installation/quick_start.md) to install Invoke.
 
 ## Downloading models and using existing models
 
 The Model Manager tab in the UI provides a few ways to install models, including using your already-downloaded models. You'll see a popup directing you there on first startup. For more information, see the [model install docs].
 
-## Missing models after updating to v4
+## Missing models after updating from v3
 
-If you find some models are missing after updating to v4, it's likely they weren't correctly registered before the update and didn't get picked up in the migration.
+If you find some models are missing after updating from v3, it's likely they weren't correctly registered before the update and didn't get picked up in the migration.
 
 You can use the `Scan Folder` tab in the Model Manager UI to fix this. The models will either be in the old, now-unused `autoimport` folder, or your `models` folder.
 
@@ -37,115 +29,27 @@ Follow the same steps to scan and import the missing models.
 ## Slow generation
 
 - Check the [system requirements] to ensure that your system is capable of generating images.
-- Check the `ram` setting in `invokeai.yaml`. This setting tells Invoke how much of your system RAM can be used to cache models. Having this too high or too low can slow things down. That said, it's generally safest to not set this at all and instead let Invoke manage it.
-- Check the `vram` setting in `invokeai.yaml`. This setting tells Invoke how much of your GPU VRAM can be used to cache models. Counter-intuitively, if this setting is too high, Invoke will need to do a lot of shuffling of models as it juggles the VRAM cache and the currently-loaded model. The default value of 0.25 is generally works well for GPUs without 16GB or more VRAM. Even on a 24GB card, the default works well.
-- Check that your generations are happening on your GPU (if you have one). InvokeAI will log what is being used for generation upon startup. If your GPU isn't used, re-install to ensure the correct versions of torch get installed.
-- If you are on Windows, you may have exceeded your GPU's VRAM capacity and are using slower [shared GPU memory](#shared-gpu-memory-windows). There's a guide to opt out of this behaviour in the linked FAQ entry.
-
-## Shared GPU Memory (Windows)
-
-!!! tip "Nvidia GPUs with driver 536.40"
-
-    This only applies to current Nvidia cards with driver 536.40 or later, released in June 2023.
-
-When the GPU doesn't have enough VRAM for a task, Windows is able to allocate some of its CPU RAM to the GPU. This is much slower than VRAM, but it does allow the system to generate when it otherwise might no have enough VRAM.
-
-When shared GPU memory is used, generation slows down dramatically - but at least it doesn't crash.
-
-If you'd like to opt out of this behavior and instead get an error when you exceed your GPU's VRAM, follow [this guide from Nvidia](https://nvidia.custhelp.com/app/answers/detail/a_id/5490).
-
-Here's how to get the python path required in the linked guide:
-
-- Run `invoke.bat`.
-- Select option 2 for developer console.
-- At least one python path will be printed. Copy the path that includes your invoke installation directory (typically the first).
-
-## Installer cannot find python (Windows)
-
-Ensure that you checked **Add python.exe to PATH** when installing Python. This can be found at the bottom of the Python Installer window. If you already have Python installed, you can re-run the python installer, choose the Modify option and check the box.
+- Follow the [Low-VRAM mode guide](./features/low-vram.md) to optimize performance.
+- Check that your generations are happening on your GPU (if you have one). Invoke will log what is being used for generation upon startup. If your GPU isn't used, re-install to and ensure you select the appropriate GPU option.
+- If you are on Windows with an Nvidia GPU, you may have exceeded your GPU's VRAM capacity and are triggering Nvidia's "sysmem fallback". There's a guide to opt out of this behaviour in the [Low-VRAM mode guide](./features/low-vram.md).
 
 ## Triton error on startup
 
-This can be safely ignored. InvokeAI doesn't use Triton, but if you are on Linux and wish to dismiss the error, you can install Triton.
+This can be safely ignored. Invoke doesn't use Triton, but if you are on Linux and wish to dismiss the error, you can install Triton.
 
-## Updated to 3.4.0 and xformers can’t load C++/CUDA
+## Unable to Copy on Firefox
 
-An issue occurred with your PyTorch update. Follow these steps to fix :
+Firefox does not allow Invoke to directly access the clipboard by default. As a result, you may be unable to use certain copy functions. You can fix this by configuring Firefox to allow access to write to the clipboard:
 
-1. Launch your invoke.bat / invoke.sh and select the option to open the developer console
-2. Run:`pip install ".[xformers]" --upgrade --force-reinstall --extra-index-url https://download.pytorch.org/whl/cu121`
-   - If you run into an error with `typing_extensions`, re-open the developer console and run: `pip install -U typing-extensions`
-
-Note that v3.4.0 is an old, unsupported version. Please upgrade to the [latest release].
-
-## Install failed and says `pip` is out of date
-
-An out of date `pip` typically won't cause an installation to fail. The cause of the error can likely be found above the message that says `pip` is out of date.
-
-If you saw that warning but the install went well, don't worry about it (but you can update `pip` afterwards if you'd like).
+- Go to `about:config` and click the Accept button
+- Search for `dom.events.asyncClipboard.clipboardItem`
+- Set it to `true` by clicking the toggle button
+- Restart Firefox
 
 ## Replicate image found online
 
 Most example images with prompts that you'll find on the internet have been generated using different software, so you can't expect to get identical results. In order to reproduce an image, you need to replicate the exact settings and processing steps, including (but not limited to) the model, the positive and negative prompts, the seed, the sampler, the exact image size, any upscaling steps, etc.
 
-## OSErrors on Windows while installing dependencies
-
-During a zip file installation or an update, installation stops with an error like this:
-
-![broken-dependency-screenshot](./assets/troubleshooting/broken-dependency.png){:width="800px"}
-
-To resolve this, re-install the application as described above.
-
-## HuggingFace install failed due to invalid access token
-
-Some HuggingFace models require you to authenticate using an [access token].
-
-Invoke doesn't manage this token for you, but it's easy to set it up:
-
-- Follow the instructions in the link above to create an access token. Copy it.
-- Run the launcher script.
-- Select option 2 (developer console).
-- Paste the following command:
-
-  ```sh
-  python -c "import huggingface_hub; huggingface_hub.login()"
-  ```
-
-- Paste your access token when prompted and press Enter. You won't see anything when you paste it.
-- Type `n` if prompted about git credentials.
-
-If you get an error, try the command again - maybe the token didn't paste correctly.
-
-Once your token is set, start Invoke and try downloading the model again. The installer will automatically use the access token.
-
-If the install still fails, you may not have access to the model.
-
-## Stable Diffusion XL generation fails after trying to load UNet
-
-InvokeAI is working in other respects, but when trying to generate
-images with Stable Diffusion XL you get a "Server Error". The text log
-in the launch window contains this log line above several more lines of
-error messages:
-
-`INFO --> Loading model:D:\LONG\PATH\TO\MODEL, type sdxl:main:unet`
-
-This failure mode occurs when there is a network glitch during
-downloading the very large SDXL model.
-
-To address this, first go to the Model Manager and delete the
-Stable-Diffusion-XL-base-1.X model. Then, click the HuggingFace tab,
-paste the Repo ID stabilityai/stable-diffusion-xl-base-1.0 and install
-the model.
-
-## Package dependency conflicts during installation or update
-
-If you have previously installed InvokeAI or another Stable Diffusion
-package, the installer may occasionally pick up outdated libraries and
-either the installer or `invoke` will fail with complaints about
-library conflicts.
-
-To resolve this, re-install the application as described above.
-
 ## Invalid configuration file
 
 Everything seems to install ok, you get a `ValidationError` when starting up the app.
@@ -154,64 +58,9 @@ This is caused by an invalid setting in the `invokeai.yaml` configuration file.
 
 Check the [configuration docs] for more detail about the settings and how to specify them.
 
-## `ModuleNotFoundError: No module named 'controlnet_aux'`
+## Out of Memory Errors
 
-`controlnet_aux` is a dependency of Invoke and appears to have been packaged or distributed strangely. Sometimes, it doesn't install correctly. This is outside our control.
-
-If you encounter this error, the solution is to remove the package from the `pip` cache and re-run the Invoke installer so a fresh, working version of `controlnet_aux` can be downloaded and installed:
-
-- Run the Invoke launcher
-- Choose the developer console option
-- Run this command: `pip cache remove controlnet_aux`
-- Close the terminal window
-- Download and run the [installer][latest release], selecting your current install location
-
-## Out of Memory Issues
-
-The models are large, VRAM is expensive, and you may find yourself
-faced with Out of Memory errors when generating images. Here are some
-tips to reduce the problem:
-
-!!! info "Optimizing for GPU VRAM"
-
-    === "4GB VRAM GPU"
-
-        This should be adequate for 512x512 pixel images using Stable Diffusion 1.5
-        and derived models, provided that you do not use the NSFW checker. It won't be loaded unless you go into the UI settings and turn it on.
-
-        If you are on a CUDA-enabled GPU, we will automatically use xformers or torch-sdp to reduce VRAM requirements, though you can explicitly configure this. See the [configuration docs].
-
-    === "6GB VRAM GPU"
-
-        This is a border case. Using the SD 1.5 series you should be able to
-        generate images up to 640x640 with the NSFW checker enabled, and up to
-        1024x1024 with it disabled.
-
-        If you run into persistent memory issues there are a series of
-        environment variables that you can set before launching InvokeAI that
-        alter how the PyTorch machine learning library manages memory. See
-        <https://pytorch.org/docs/stable/notes/cuda.html#memory-management> for
-        a list of these tweaks.
-
-    === "12GB VRAM GPU"
-
-        This should be sufficient to generate larger images up to about 1280x1280.
-  
-## Checkpoint Models Load Slowly or Use Too Much RAM
-
-The difference between diffusers models (a folder containing multiple
-subfolders) and checkpoint models (a file ending with .safetensors or
-.ckpt) is that InvokeAI is able to load diffusers models into memory
-incrementally, while checkpoint models must be loaded all at
-once. With very large models, or systems with limited RAM, you may
-experience slowdowns and other memory-related issues when loading
-checkpoint models.
-
-To solve this, go to the Model Manager tab (the cube), select the
-checkpoint model that's giving you trouble, and press the "Convert"
-button in the upper right of your browser window. This will convert the
-checkpoint into a diffusers model, after which loading should be
-faster and less memory-intensive.
+The models are large, VRAM is expensive, and you may find yourself faced with Out of Memory errors when generating images. Follow our [Low-VRAM mode guide](./features/low-vram.md) to configure Invoke to prevent these.
 
 ## Memory Leak (Linux)
 
@@ -253,8 +102,6 @@ Note the differences between memory allocated as chunks in an arena vs. memory a
 
 [model install docs]: ./installation/models.md
 [system requirements]: ./installation/requirements.md
-[latest release]: https://github.com/invoke-ai/InvokeAI/releases/latest
 [create an issue]: https://github.com/invoke-ai/InvokeAI/issues
 [discord]: https://discord.gg/ZmtBAhwWhy
 [configuration docs]: ./configuration.md
-[access token]: https://huggingface.co/docs/hub/security-tokens#how-to-manage-user-access-tokens

docs/features/low-vram.md

@@ -28,11 +28,13 @@ It is possible to fine-tune the settings for best performance or if you still ge
 
 ## Details and fine-tuning
 
-Low-VRAM mode involves 3 features, each of which can be configured or fine-tuned:
+Low-VRAM mode involves 4 features, each of which can be configured or fine-tuned:
 
-- Partial model loading
-- Dynamic RAM and VRAM cache sizes
-- Working memory
+- Partial model loading (`enable_partial_loading`)
+- PyTorch CUDA allocator config (`pytorch_cuda_alloc_conf`)
+- Dynamic RAM and VRAM cache sizes (`max_cache_ram_gb`, `max_cache_vram_gb`)
+- Working memory (`device_working_mem_gb`)
+- Keeping a RAM weight copy (`keep_ram_copy_of_weights`)
 
 Read on to learn about these features and understand how to fine-tune them for your system and use-cases.
 
@@ -50,6 +52,16 @@ As described above, you can enable partial model loading by adding this line to
 enable_partial_loading: true
 ```
 
+### PyTorch CUDA allocator config
+
+The PyTorch CUDA allocator's behavior can be configured using the `pytorch_cuda_alloc_conf` config. Tuning the allocator configuration can help to reduce the peak reserved VRAM. The optimal configuration is dependent on many factors (e.g. device type, VRAM, CUDA driver version, etc.), but switching from PyTorch's native allocator to using CUDA's built-in allocator works well on many systems. To try this, add the following line to your `invokeai.yaml` file:
+
+```yaml
+pytorch_cuda_alloc_conf: "backend:cudaMallocAsync"
+```
+
+A more complete explanation of the available configuration options is [here](https://pytorch.org/docs/stable/notes/cuda.html#optimizing-memory-usage-with-pytorch-cuda-alloc-conf).
+
 ### Dynamic RAM and VRAM cache sizes
 
 Loading models from disk is slow and can be a major bottleneck for performance. Invoke uses two model caches - RAM and VRAM - to reduce loading from disk to a minimum.
@@ -67,23 +79,33 @@ As of v5.6.0, the caches are dynamically sized. The `ram` and `vram` settings ar
 But, if your GPU has enough VRAM to hold models fully, you might get a perf boost by manually setting the cache sizes in `invokeai.yaml`:
 
 ```yaml
-# Set the RAM cache size to as large as possible, leaving a few GB free for the rest of your system and Invoke.
-# For example, if your system has 32GB RAM, 28GB is a good value.
+# The default max cache RAM size is logged on InvokeAI startup. It is determined based on your system RAM / VRAM.
+# You can override the default value by setting `max_cache_ram_gb`.
+# Increasing `max_cache_ram_gb` will increase the amount of RAM used to cache inactive models, resulting in faster model
+# reloads for the cached models.
+# As an example, if your system has 32GB of RAM and no other heavy processes, setting the `max_cache_ram_gb` to 28GB
+# might be a good value to achieve aggressive model caching.
 max_cache_ram_gb: 28
-# Set the VRAM cache size to be as large as possible while leaving enough room for the working memory of the tasks you will be doing.
-# For example, on a 24GB GPU that will be running unquantized FLUX without any auxiliary models,
-# 18GB is a good value.
-max_cache_vram_gb: 18
+
+# The default max cache VRAM size is adjusted dynamically based on the amount of available VRAM (taking into
+# consideration the VRAM used by other processes).
+# You can override the default value by setting `max_cache_vram_gb`.
+# CAUTION: Most users should not manually set this value. See warning below.
+max_cache_vram_gb: 16
 ```
 
-!!! tip "Max safe value for `max_cache_vram_gb`"
+!!! warning "Max safe value for `max_cache_vram_gb`"
 
-    To determine the max safe value for `max_cache_vram_gb`, subtract `device_working_mem_gb` from your GPU's VRAM. As described below, the default for `device_working_mem_gb` is 3GB.
+    Most users should not manually configure the `max_cache_vram_gb`. This configuration value takes precedence over the `device_working_mem_gb` and any operations that explicitly reserve additional working memory (e.g. VAE decode). As such, manually configuring it increases the likelihood of encountering out-of-memory errors.
+
+    For users who wish to configure `max_cache_vram_gb`, the max safe value can be determined by subtracting `device_working_mem_gb` from your GPU's VRAM. As described below, the default for `device_working_mem_gb` is 3GB.
 
     For example, if you have a 12GB GPU, the max safe value for `max_cache_vram_gb` is `12GB - 3GB = 9GB`.
 
     If you had increased `device_working_mem_gb` to 4GB, then the max safe value for `max_cache_vram_gb` is `12GB - 4GB = 8GB`.
 
+    Most users who override `max_cache_vram_gb` are doing so because they wish to use significantly less VRAM, and should be setting `max_cache_vram_gb` to a value significantly less than the 'max safe value'.
+
 ### Working memory
 
 Invoke cannot use _all_ of your VRAM for model caching and loading. It requires some VRAM to use as working memory for various operations.
@@ -109,6 +131,15 @@ device_working_mem_gb: 4
 
     Once decoding completes, the model manager "reclaims" the extra VRAM allocated as working memory for future model loading operations.
 
+### Keeping a RAM weight copy
+
+Invoke has the option of keeping a RAM copy of all model weights, even when they are loaded onto the GPU. This optimization is _on_ by default, and enables faster model switching and LoRA patching. Disabling this feature will reduce the average RAM load while running Invoke (peak RAM likely won't change), at the cost of slower model switching and LoRA patching. If you have limited RAM, you can disable this optimization:
+
+```yaml
+# Set to false to reduce the average RAM usage at the cost of slower model switching and LoRA patching.
+keep_ram_copy_of_weights: false
+```
+
 ### Disabling Nvidia sysmem fallback (Windows only)
 
 On Windows, Nvidia GPUs are able to use system RAM when their VRAM fills up via **sysmem fallback**. While it sounds like a good idea on the surface, in practice it causes massive slowdowns during generation.
@@ -127,3 +158,19 @@ It is strongly suggested to disable this feature:
     If the sysmem fallback feature sounds familiar, that's because Invoke's partial model loading strategy is conceptually very similar - use VRAM when there's room, else fall back to RAM.
 
     Unfortunately, the Nvidia implementation is not optimized for applications like Invoke and does more harm than good.
+
+## Troubleshooting
+
+### Windows page file
+
+Invoke has high virtual memory (a.k.a. 'committed memory') requirements. This can cause issues on Windows if the page file size limits are hit. (See this issue for the technical details on why this happens: https://github.com/invoke-ai/InvokeAI/issues/7563).
+
+If you run out of page file space, InvokeAI may crash. Often, these crashes will happen with one of the following errors:
+
+- InvokeAI exits with Windows error code `3221225477`
+- InvokeAI crashes without an error, but `eventvwr.msc` reveals an error with code `0xc0000005` (the hex equivalent of `3221225477`)
+
+If you are running out of page file space, try the following solutions:
+
+- Make sure that you have sufficient disk space for the page file to grow. Watch your disk usage as Invoke runs. If it climbs near 100% leading up to the crash, then this is very likely the source of the issue. Clear out some disk space to resolve the issue.
+- Make sure that your page file is set to "System managed size" (this is the default) rather than a custom size. Under the "System managed size" policy, the page file will grow dynamically as needed. 

docs/installation/manual.md

@@ -88,13 +88,13 @@ The following commands vary depending on the version of Invoke being installed a
 8. Install the `invokeai` package. Substitute the package specifier and version.
 
     ```sh
-    uv pip install <PACKAGE_SPECIFIER>=<VERSION> --python 3.11 --python-preference only-managed --force-reinstall
+    uv pip install <PACKAGE_SPECIFIER>==<VERSION> --python 3.11 --python-preference only-managed --force-reinstall
     ```
 
     If you determined you needed to use a `PyPI` index URL in the previous step, you'll need to add `--index=<INDEX_URL>` like this:
 
     ```sh
-    uv pip install <PACKAGE_SPECIFIER>=<VERSION> --python 3.11 --python-preference only-managed --index=<INDEX_URL> --force-reinstall
+    uv pip install <PACKAGE_SPECIFIER>==<VERSION> --python 3.11 --python-preference only-managed --index=<INDEX_URL> --force-reinstall
     ```
 
 9. Deactivate and reactivate your venv so that the invokeai-specific commands become available in the environment:

docs/installation/quick_start.md

@@ -99,6 +99,20 @@ We recommend watching our [Getting Started Playlist](https://www.youtube.com/pla
 - Using control layers and reference guides.
 - Refining images with advanced workflows.
 
+## Troubleshooting
+
+If installation fails, retrying the install in Repair Mode may fix it. There's a checkbox to enable this on the Review step of the install flow.
+
+If that doesn't fix it, [clearing the `uv` cache](https://docs.astral.sh/uv/reference/cli/#uv-cache-clean) might do the trick:
+
+- Open and start the dev console (button at the bottom-left of the launcher).
+- Run `uv cache clean`.
+- Retry the installation. Enable Repair Mode for good measure.
+
+If you are still unable to install, try installing to a different location and see if that works.
+
+If you still have problems, ask for help on the Invoke [discord](https://discord.gg/ZmtBAhwWhy).
+
 ## Other Installation Methods
 
 - You can install the Invoke application as a python package. See our [manual install](./manual.md) docs.

docs/installation/requirements.md

@@ -4,7 +4,9 @@ Invoke runs on Windows 10+, macOS 14+ and Linux (Ubuntu 20.04+ is well-tested).
 
 ## Hardware
 
-Hardware requirements vary significantly depending on model and image output size. The requirements below are rough guidelines.
+Hardware requirements vary significantly depending on model and image output size.
+
+The requirements below are rough guidelines for best performance. GPUs with less VRAM typically still work, if a bit slower. Follow the [Low-VRAM mode guide](./features/low-vram.md) to optimize performance.
 
 - All Apple Silicon (M1, M2, etc) Macs work, but 16GB+ memory is recommended.
 - AMD GPUs are supported on Linux only. The VRAM requirements are the same as Nvidia GPUs.

docs/nodes/communityNodes.md

@@ -535,7 +535,7 @@ View:
 **Node Link:** https://github.com/simonfuhrmann/invokeai-stereo
 
 **Example Workflow and Output**
-</br><img src="https://github.com/simonfuhrmann/invokeai-stereo/blob/main/docs/example_promo_03.jpg" width="500" />
+</br><img src="https://raw.githubusercontent.com/simonfuhrmann/invokeai-stereo/refs/heads/main/docs/example_promo_03.jpg" width="600" />
 
 --------------------------------
 ### Simple Skin Detection

invokeai/app/api/routers/boards.py

@@ -7,6 +7,7 @@ from pydantic import BaseModel, Field
 from invokeai.app.api.dependencies import ApiDependencies
 from invokeai.app.services.board_records.board_records_common import BoardChanges, BoardRecordOrderBy
 from invokeai.app.services.boards.boards_common import BoardDTO
+from invokeai.app.services.image_records.image_records_common import ImageCategory
 from invokeai.app.services.shared.pagination import OffsetPaginatedResults
 from invokeai.app.services.shared.sqlite.sqlite_common import SQLiteDirection
 
@@ -87,7 +88,9 @@ async def delete_board(
     try:
         if include_images is True:
             deleted_images = ApiDependencies.invoker.services.board_images.get_all_board_image_names_for_board(
-                board_id=board_id
+                board_id=board_id,
+                categories=None,
+                is_intermediate=None,
             )
             ApiDependencies.invoker.services.images.delete_images_on_board(board_id=board_id)
             ApiDependencies.invoker.services.boards.delete(board_id=board_id)
@@ -98,7 +101,9 @@ async def delete_board(
             )
         else:
             deleted_board_images = ApiDependencies.invoker.services.board_images.get_all_board_image_names_for_board(
-                board_id=board_id
+                board_id=board_id,
+                categories=None,
+                is_intermediate=None,
             )
             ApiDependencies.invoker.services.boards.delete(board_id=board_id)
             return DeleteBoardResult(
@@ -142,10 +147,14 @@ async def list_boards(
 )
 async def list_all_board_image_names(
     board_id: str = Path(description="The id of the board"),
+    categories: list[ImageCategory] | None = Query(default=None, description="The categories of image to include."),
+    is_intermediate: bool | None = Query(default=None, description="Whether to list intermediate images."),
 ) -> list[str]:
     """Gets a list of images for a board"""
 
     image_names = ApiDependencies.invoker.services.board_images.get_all_board_image_names_for_board(
         board_id,
+        categories,
+        is_intermediate,
     )
     return image_names

invokeai/app/api/routers/model_manager.py

@@ -858,6 +858,18 @@ async def get_stats() -> Optional[CacheStats]:
     return ApiDependencies.invoker.services.model_manager.load.ram_cache.stats
 
 
+@model_manager_router.post(
+    "/empty_model_cache",
+    operation_id="empty_model_cache",
+    status_code=200,
+)
+async def empty_model_cache() -> None:
+    """Drop all models from the model cache to free RAM/VRAM. 'Locked' models that are in active use will not be dropped."""
+    # Request 1000GB of room in order to force the cache to drop all models.
+    ApiDependencies.invoker.services.logger.info("Emptying model cache.")
+    ApiDependencies.invoker.services.model_manager.load.ram_cache.make_room(1000 * 2**30)
+
+
 class HFTokenStatus(str, Enum):
     VALID = "valid"
     INVALID = "invalid"

invokeai/app/api/routers/session_queue.py

@@ -10,11 +10,13 @@ from invokeai.app.services.session_queue.session_queue_common import (
     QUEUE_ITEM_STATUS,
     Batch,
     BatchStatus,
+    CancelAllExceptCurrentResult,
     CancelByBatchIDsResult,
     CancelByDestinationResult,
     ClearResult,
     EnqueueBatchResult,
     PruneResult,
+    RetryItemsResult,
     SessionQueueCountsByDestination,
     SessionQueueItem,
     SessionQueueItemDTO,
@@ -46,7 +48,9 @@ async def enqueue_batch(
 ) -> EnqueueBatchResult:
     """Processes a batch and enqueues the output graphs for execution."""
 
-    return ApiDependencies.invoker.services.session_queue.enqueue_batch(queue_id=queue_id, batch=batch, prepend=prepend)
+    return await ApiDependencies.invoker.services.session_queue.enqueue_batch(
+        queue_id=queue_id, batch=batch, prepend=prepend
+    )
 
 
 @session_queue_router.get(
@@ -94,6 +98,18 @@ async def Pause(
     return ApiDependencies.invoker.services.session_processor.pause()
 
 
+@session_queue_router.put(
+    "/{queue_id}/cancel_all_except_current",
+    operation_id="cancel_all_except_current",
+    responses={200: {"model": CancelAllExceptCurrentResult}},
+)
+async def cancel_all_except_current(
+    queue_id: str = Path(description="The queue id to perform this operation on"),
+) -> CancelAllExceptCurrentResult:
+    """Immediately cancels all queue items except in-processing items"""
+    return ApiDependencies.invoker.services.session_queue.cancel_all_except_current(queue_id=queue_id)
+
+
 @session_queue_router.put(
     "/{queue_id}/cancel_by_batch_ids",
     operation_id="cancel_by_batch_ids",
@@ -122,6 +138,19 @@ async def cancel_by_destination(
     )
 
 
+@session_queue_router.put(
+    "/{queue_id}/retry_items_by_id",
+    operation_id="retry_items_by_id",
+    responses={200: {"model": RetryItemsResult}},
+)
+async def retry_items_by_id(
+    queue_id: str = Path(description="The queue id to perform this operation on"),
+    item_ids: list[int] = Body(description="The queue item ids to retry"),
+) -> RetryItemsResult:
+    """Immediately cancels all queue items with the given origin"""
+    return ApiDependencies.invoker.services.session_queue.retry_items_by_id(queue_id=queue_id, item_ids=item_ids)
+
+
 @session_queue_router.put(
     "/{queue_id}/clear",
     operation_id="clear",

invokeai/app/api/routers/utilities.py

@@ -25,6 +25,7 @@ async def parse_dynamicprompts(
     prompt: str = Body(description="The prompt to parse with dynamicprompts"),
     max_prompts: int = Body(ge=1, le=10000, default=1000, description="The max number of prompts to generate"),
     combinatorial: bool = Body(default=True, description="Whether to use the combinatorial generator"),
+    seed: int | None = Body(None, description="The seed to use for random generation. Only used if not combinatorial"),
 ) -> DynamicPromptsResponse:
     """Creates a batch process"""
     max_prompts = min(max_prompts, 10000)
@@ -35,7 +36,7 @@ async def parse_dynamicprompts(
             generator = CombinatorialPromptGenerator()
             prompts = generator.generate(prompt, max_prompts=max_prompts)
         else:
-            generator = RandomPromptGenerator()
+            generator = RandomPromptGenerator(seed=seed)
             prompts = generator.generate(prompt, num_images=max_prompts)
     except ParseException as e:
         prompts = [prompt]

invokeai/app/api_app.py

@@ -1,12 +1,8 @@
 import asyncio
 import logging
-import mimetypes
-import socket
 from contextlib import asynccontextmanager
 from pathlib import Path
 
-import torch
-import uvicorn
 from fastapi import FastAPI, Request
 from fastapi.middleware.cors import CORSMiddleware
 from fastapi.middleware.gzip import GZipMiddleware
@@ -15,11 +11,7 @@ from fastapi.responses import HTMLResponse, RedirectResponse
 from fastapi_events.handlers.local import local_handler
 from fastapi_events.middleware import EventHandlerASGIMiddleware
 from starlette.middleware.base import BaseHTTPMiddleware, RequestResponseEndpoint
-from torch.backends.mps import is_available as is_mps_available
 
-# for PyCharm:
-# noinspection PyUnresolvedReferences
-import invokeai.backend.util.hotfixes  # noqa: F401 (monkeypatching on import)
 import invokeai.frontend.web as web_dir
 from invokeai.app.api.dependencies import ApiDependencies
 from invokeai.app.api.no_cache_staticfiles import NoCacheStaticFiles
@@ -38,31 +30,13 @@ from invokeai.app.api.routers import (
 from invokeai.app.api.sockets import SocketIO
 from invokeai.app.services.config.config_default import get_config
 from invokeai.app.util.custom_openapi import get_openapi_func
-from invokeai.backend.util.devices import TorchDevice
 from invokeai.backend.util.logging import InvokeAILogger
 
 app_config = get_config()
-
-
-if is_mps_available():
-    import invokeai.backend.util.mps_fixes  # noqa: F401 (monkeypatching on import)
-
-
 logger = InvokeAILogger.get_logger(config=app_config)
-# fix for windows mimetypes registry entries being borked
-# see https://github.com/invoke-ai/InvokeAI/discussions/3684#discussioncomment-6391352
-mimetypes.add_type("application/javascript", ".js")
-mimetypes.add_type("text/css", ".css")
-
-torch_device_name = TorchDevice.get_torch_device_name()
-logger.info(f"Using torch device: {torch_device_name}")
 
 loop = asyncio.new_event_loop()
 
-# We may change the port if the default is in use, this global variable is used to store the port so that we can log
-# the correct port when the server starts in the lifespan handler.
-port = app_config.port
-
 
 @asynccontextmanager
 async def lifespan(app: FastAPI):
@@ -71,7 +45,7 @@ async def lifespan(app: FastAPI):
 
     # Log the server address when it starts - in case the network log level is not high enough to see the startup log
     proto = "https" if app_config.ssl_certfile else "http"
-    msg = f"Invoke running on {proto}://{app_config.host}:{port} (Press CTRL+C to quit)"
+    msg = f"Invoke running on {proto}://{app_config.host}:{app_config.port} (Press CTRL+C to quit)"
 
     # Logging this way ignores the logger's log level and _always_ logs the message
     record = logger.makeRecord(
@@ -186,73 +160,3 @@ except RuntimeError:
 app.mount(
     "/static", NoCacheStaticFiles(directory=Path(web_root_path, "static/")), name="static"
 )  # docs favicon is in here
-
-
-def check_cudnn(logger: logging.Logger) -> None:
-    """Check for cuDNN issues that could be causing degraded performance."""
-    if torch.backends.cudnn.is_available():
-        try:
-            # Note: At the time of writing (torch 2.2.1), torch.backends.cudnn.version() only raises an error the first
-            # time it is called. Subsequent calls will return the version number without complaining about a mismatch.
-            cudnn_version = torch.backends.cudnn.version()
-            logger.info(f"cuDNN version: {cudnn_version}")
-        except RuntimeError as e:
-            logger.warning(
-                "Encountered a cuDNN version issue. This may result in degraded performance. This issue is usually "
-                "caused by an incompatible cuDNN version installed in your python environment, or on the host "
-                f"system. Full error message:\n{e}"
-            )
-
-
-def invoke_api() -> None:
-    def find_port(port: int) -> int:
-        """Find a port not in use starting at given port"""
-        # Taken from https://waylonwalker.com/python-find-available-port/, thanks Waylon!
-        # https://github.com/WaylonWalker
-        with socket.socket(socket.AF_INET, socket.SOCK_STREAM) as s:
-            s.settimeout(1)
-            if s.connect_ex(("localhost", port)) == 0:
-                return find_port(port=port + 1)
-            else:
-                return port
-
-    if app_config.dev_reload:
-        try:
-            import jurigged
-        except ImportError as e:
-            logger.error(
-                'Can\'t start `--dev_reload` because jurigged is not found; `pip install -e ".[dev]"` to include development dependencies.',
-                exc_info=e,
-            )
-        else:
-            jurigged.watch(logger=InvokeAILogger.get_logger(name="jurigged").info)
-
-    global port
-    port = find_port(app_config.port)
-    if port != app_config.port:
-        logger.warn(f"Port {app_config.port} in use, using port {port}")
-
-    check_cudnn(logger)
-
-    config = uvicorn.Config(
-        app=app,
-        host=app_config.host,
-        port=port,
-        loop="asyncio",
-        log_level=app_config.log_level_network,
-        ssl_certfile=app_config.ssl_certfile,
-        ssl_keyfile=app_config.ssl_keyfile,
-    )
-    server = uvicorn.Server(config)
-
-    # replace uvicorn's loggers with InvokeAI's for consistent appearance
-    uvicorn_logger = InvokeAILogger.get_logger("uvicorn")
-    uvicorn_logger.handlers.clear()
-    for hdlr in logger.handlers:
-        uvicorn_logger.addHandler(hdlr)
-
-    loop.run_until_complete(server.serve())
-
-
-if __name__ == "__main__":
-    invoke_api()

invokeai/app/invocations/__init__.py

@@ -1,33 +1,5 @@
-import shutil
-import sys
-from importlib.util import module_from_spec, spec_from_file_location
 from pathlib import Path
 
-from invokeai.app.services.config.config_default import get_config
-
-custom_nodes_path = Path(get_config().custom_nodes_path)
-custom_nodes_path.mkdir(parents=True, exist_ok=True)
-
-custom_nodes_init_path = str(custom_nodes_path / "__init__.py")
-custom_nodes_readme_path = str(custom_nodes_path / "README.md")
-
-# copy our custom nodes __init__.py to the custom nodes directory
-shutil.copy(Path(__file__).parent / "custom_nodes/init.py", custom_nodes_init_path)
-shutil.copy(Path(__file__).parent / "custom_nodes/README.md", custom_nodes_readme_path)
-
-# set the same permissions as the destination directory, in case our source is read-only,
-# so that the files are user-writable
-for p in custom_nodes_path.glob("**/*"):
-    p.chmod(custom_nodes_path.stat().st_mode)
-
-# Import custom nodes, see https://docs.python.org/3/library/importlib.html#importing-programmatically
-spec = spec_from_file_location("custom_nodes", custom_nodes_init_path)
-if spec is None or spec.loader is None:
-    raise RuntimeError(f"Could not load custom nodes from {custom_nodes_init_path}")
-module = module_from_spec(spec)
-sys.modules[spec.name] = module
-spec.loader.exec_module(module)
-
 # add core nodes to __all__
 python_files = filter(lambda f: not f.name.startswith("_"), Path(__file__).parent.glob("*.py"))
 __all__ = [f.stem for f in python_files]  # type: ignore

invokeai/app/invocations/baseinvocation.py

@@ -44,8 +44,6 @@ if TYPE_CHECKING:
 
 logger = InvokeAILogger.get_logger()
 
-CUSTOM_NODE_PACK_SUFFIX = "__invokeai-custom-node"
-
 
 class InvalidVersionError(ValueError):
     pass
@@ -240,6 +238,11 @@ class BaseInvocation(ABC, BaseModel):
         """Gets the invocation's output annotation (i.e. the return annotation of its `invoke()` method)."""
         return signature(cls.invoke).return_annotation
 
+    @classmethod
+    def get_invocation_for_type(cls, invocation_type: str) -> BaseInvocation | None:
+        """Gets the invocation class for a given invocation type."""
+        return cls.get_invocations_map().get(invocation_type)
+
     @staticmethod
     def json_schema_extra(schema: dict[str, Any], model_class: Type[BaseInvocation]) -> None:
         """Adds various UI-facing attributes to the invocation's OpenAPI schema."""
@@ -446,8 +449,27 @@ def invocation(
         if re.compile(r"^\S+$").match(invocation_type) is None:
             raise ValueError(f'"invocation_type" must consist of non-whitespace characters, got "{invocation_type}"')
 
+        # The node pack is the module name - will be "invokeai" for built-in nodes
+        node_pack = cls.__module__.split(".")[0]
+
+        # Handle the case where an existing node is being clobbered by the one we are registering
         if invocation_type in BaseInvocation.get_invocation_types():
-            raise ValueError(f'Invocation type "{invocation_type}" already exists')
+            clobbered_invocation = BaseInvocation.get_invocation_for_type(invocation_type)
+            # This should always be true - we just checked if the invocation type was in the set
+            assert clobbered_invocation is not None
+
+            clobbered_node_pack = clobbered_invocation.UIConfig.node_pack
+
+            if clobbered_node_pack == "invokeai":
+                # The node being clobbered is a core node
+                raise ValueError(
+                    f'Cannot load node "{invocation_type}" from node pack "{node_pack}" - a core node with the same type already exists'
+                )
+            else:
+                # The node being clobbered is a custom node
+                raise ValueError(
+                    f'Cannot load node "{invocation_type}" from node pack "{node_pack}" - a node with the same type already exists in node pack "{clobbered_node_pack}"'
+                )
 
         validate_fields(cls.model_fields, invocation_type)
 
@@ -457,8 +479,7 @@ def invocation(
         uiconfig["tags"] = tags
         uiconfig["category"] = category
         uiconfig["classification"] = classification
-        # The node pack is the module name - will be "invokeai" for built-in nodes
-        uiconfig["node_pack"] = cls.__module__.split(".")[0]
+        uiconfig["node_pack"] = node_pack
 
         if version is not None:
             try:

invokeai/app/invocations/batch.py

@@ -0,0 +1,274 @@
+from typing import Literal
+
+from pydantic import BaseModel
+
+from invokeai.app.invocations.baseinvocation import (
+    BaseInvocation,
+    BaseInvocationOutput,
+    Classification,
+    invocation,
+    invocation_output,
+)
+from invokeai.app.invocations.fields import (
+    ImageField,
+    Input,
+    InputField,
+    OutputField,
+)
+from invokeai.app.invocations.primitives import (
+    FloatOutput,
+    ImageOutput,
+    IntegerOutput,
+    StringOutput,
+)
+from invokeai.app.services.shared.invocation_context import InvocationContext
+
+BATCH_GROUP_IDS = Literal[
+    "None",
+    "Group 1",
+    "Group 2",
+    "Group 3",
+    "Group 4",
+    "Group 5",
+]
+
+
+class NotExecutableNodeError(Exception):
+    def __init__(self, message: str = "This class should never be executed or instantiated directly."):
+        super().__init__(message)
+
+    pass
+
+
+class BaseBatchInvocation(BaseInvocation):
+    batch_group_id: BATCH_GROUP_IDS = InputField(
+        default="None",
+        description="The ID of this batch node's group. If provided, all batch nodes in with the same ID will be 'zipped' before execution, and all nodes' collections must be of the same size.",
+        input=Input.Direct,
+        title="Batch Group",
+    )
+
+    def __init__(self):
+        raise NotExecutableNodeError()
+
+
+@invocation(
+    "image_batch",
+    title="Image Batch",
+    tags=["primitives", "image", "batch", "special"],
+    category="primitives",
+    version="1.0.0",
+    classification=Classification.Special,
+)
+class ImageBatchInvocation(BaseBatchInvocation):
+    """Create a batched generation, where the workflow is executed once for each image in the batch."""
+
+    images: list[ImageField] = InputField(
+        default=[],
+        min_length=1,
+        description="The images to batch over",
+    )
+
+    def invoke(self, context: InvocationContext) -> ImageOutput:
+        raise NotExecutableNodeError()
+
+
+@invocation_output("image_generator_output")
+class ImageGeneratorOutput(BaseInvocationOutput):
+    """Base class for nodes that output a collection of boards"""
+
+    images: list[ImageField] = OutputField(description="The generated images")
+
+
+class ImageGeneratorField(BaseModel):
+    pass
+
+
+@invocation(
+    "image_generator",
+    title="Image Generator",
+    tags=["primitives", "board", "image", "batch", "special"],
+    category="primitives",
+    version="1.0.0",
+    classification=Classification.Special,
+)
+class ImageGenerator(BaseInvocation):
+    """Generated a collection of images for use in a batched generation"""
+
+    generator: ImageGeneratorField = InputField(
+        description="The image generator.",
+        input=Input.Direct,
+        title="Generator Type",
+    )
+
+    def __init__(self):
+        raise NotExecutableNodeError()
+
+    def invoke(self, context: InvocationContext) -> ImageGeneratorOutput:
+        raise NotExecutableNodeError()
+
+
+@invocation(
+    "string_batch",
+    title="String Batch",
+    tags=["primitives", "string", "batch", "special"],
+    category="primitives",
+    version="1.0.0",
+    classification=Classification.Special,
+)
+class StringBatchInvocation(BaseBatchInvocation):
+    """Create a batched generation, where the workflow is executed once for each string in the batch."""
+
+    strings: list[str] = InputField(
+        default=[],
+        min_length=1,
+        description="The strings to batch over",
+    )
+
+    def invoke(self, context: InvocationContext) -> StringOutput:
+        raise NotExecutableNodeError()
+
+
+@invocation_output("string_generator_output")
+class StringGeneratorOutput(BaseInvocationOutput):
+    """Base class for nodes that output a collection of strings"""
+
+    strings: list[str] = OutputField(description="The generated strings")
+
+
+class StringGeneratorField(BaseModel):
+    pass
+
+
+@invocation(
+    "string_generator",
+    title="String Generator",
+    tags=["primitives", "string", "number", "batch", "special"],
+    category="primitives",
+    version="1.0.0",
+    classification=Classification.Special,
+)
+class StringGenerator(BaseInvocation):
+    """Generated a range of strings for use in a batched generation"""
+
+    generator: StringGeneratorField = InputField(
+        description="The string generator.",
+        input=Input.Direct,
+        title="Generator Type",
+    )
+
+    def __init__(self):
+        raise NotExecutableNodeError()
+
+    def invoke(self, context: InvocationContext) -> StringGeneratorOutput:
+        raise NotExecutableNodeError()
+
+
+@invocation(
+    "integer_batch",
+    title="Integer Batch",
+    tags=["primitives", "integer", "number", "batch", "special"],
+    category="primitives",
+    version="1.0.0",
+    classification=Classification.Special,
+)
+class IntegerBatchInvocation(BaseBatchInvocation):
+    """Create a batched generation, where the workflow is executed once for each integer in the batch."""
+
+    integers: list[int] = InputField(
+        default=[],
+        min_length=1,
+        description="The integers to batch over",
+    )
+
+    def invoke(self, context: InvocationContext) -> IntegerOutput:
+        raise NotExecutableNodeError()
+
+
+@invocation_output("integer_generator_output")
+class IntegerGeneratorOutput(BaseInvocationOutput):
+    integers: list[int] = OutputField(description="The generated integers")
+
+
+class IntegerGeneratorField(BaseModel):
+    pass
+
+
+@invocation(
+    "integer_generator",
+    title="Integer Generator",
+    tags=["primitives", "int", "number", "batch", "special"],
+    category="primitives",
+    version="1.0.0",
+    classification=Classification.Special,
+)
+class IntegerGenerator(BaseInvocation):
+    """Generated a range of integers for use in a batched generation"""
+
+    generator: IntegerGeneratorField = InputField(
+        description="The integer generator.",
+        input=Input.Direct,
+        title="Generator Type",
+    )
+
+    def __init__(self):
+        raise NotExecutableNodeError()
+
+    def invoke(self, context: InvocationContext) -> IntegerGeneratorOutput:
+        raise NotExecutableNodeError()
+
+
+@invocation(
+    "float_batch",
+    title="Float Batch",
+    tags=["primitives", "float", "number", "batch", "special"],
+    category="primitives",
+    version="1.0.0",
+    classification=Classification.Special,
+)
+class FloatBatchInvocation(BaseBatchInvocation):
+    """Create a batched generation, where the workflow is executed once for each float in the batch."""
+
+    floats: list[float] = InputField(
+        default=[],
+        min_length=1,
+        description="The floats to batch over",
+    )
+
+    def invoke(self, context: InvocationContext) -> FloatOutput:
+        raise NotExecutableNodeError()
+
+
+@invocation_output("float_generator_output")
+class FloatGeneratorOutput(BaseInvocationOutput):
+    """Base class for nodes that output a collection of floats"""
+
+    floats: list[float] = OutputField(description="The generated floats")
+
+
+class FloatGeneratorField(BaseModel):
+    pass
+
+
+@invocation(
+    "float_generator",
+    title="Float Generator",
+    tags=["primitives", "float", "number", "batch", "special"],
+    category="primitives",
+    version="1.0.0",
+    classification=Classification.Special,
+)
+class FloatGenerator(BaseInvocation):
+    """Generated a range of floats for use in a batched generation"""
+
+    generator: FloatGeneratorField = InputField(
+        description="The float generator.",
+        input=Input.Direct,
+        title="Generator Type",
+    )
+
+    def __init__(self):
+        raise NotExecutableNodeError()
+
+    def invoke(self, context: InvocationContext) -> FloatGeneratorOutput:
+        raise NotExecutableNodeError()

invokeai/app/invocations/custom_nodes/init.py

@@ -10,10 +10,12 @@ from pathlib import Path
 from invokeai.backend.util.logging import InvokeAILogger
 
 logger = InvokeAILogger.get_logger()
-loaded_count = 0
+loaded_packs: list[str] = []
+failed_packs: list[str] = []
 
+custom_nodes_dir = Path(__file__).parent
 
-for d in Path(__file__).parent.iterdir():
+for d in custom_nodes_dir.iterdir():
     # skip files
     if not d.is_dir():
         continue
@@ -47,12 +49,16 @@ for d in Path(__file__).parent.iterdir():
         sys.modules[spec.name] = module
         spec.loader.exec_module(module)
 
-        loaded_count += 1
+        loaded_packs.append(module_name)
     except Exception:
+        failed_packs.append(module_name)
         full_error = traceback.format_exc()
-        logger.error(f"Failed to load node pack {module_name}:\n{full_error}")
+        logger.error(f"Failed to load node pack {module_name} (may have partially loaded):\n{full_error}")
 
     del init, module_name
 
+loaded_count = len(loaded_packs)
 if loaded_count > 0:
-    logger.info(f"Loaded {loaded_count} node packs from {Path(__file__).parent}")
+    logger.info(
+        f"Loaded {loaded_count} node pack{'s' if loaded_count != 1 else ''} from {custom_nodes_dir}: {', '.join(loaded_packs)}"
+    )

invokeai/app/invocations/denoise_latents.py

@@ -40,6 +40,7 @@ from invokeai.app.services.shared.invocation_context import InvocationContext
 from invokeai.app.util.controlnet_utils import prepare_control_image
 from invokeai.backend.ip_adapter.ip_adapter import IPAdapter
 from invokeai.backend.model_manager import BaseModelType, ModelVariantType
+from invokeai.backend.model_manager.config import AnyModelConfig
 from invokeai.backend.model_patcher import ModelPatcher
 from invokeai.backend.patches.layer_patcher import LayerPatcher
 from invokeai.backend.patches.model_patch_raw import ModelPatchRaw
@@ -85,6 +86,7 @@ def get_scheduler(
     scheduler_info: ModelIdentifierField,
     scheduler_name: str,
     seed: int,
+    unet_config: AnyModelConfig,
 ) -> Scheduler:
     """Load a scheduler and apply some scheduler-specific overrides."""
     # TODO(ryand): Silently falling back to ddim seems like a bad idea. Look into why this was added and remove if
@@ -103,6 +105,9 @@ def get_scheduler(
         "_backup": scheduler_config,
     }
 
+    if hasattr(unet_config, "prediction_type"):
+        scheduler_config["prediction_type"] = unet_config.prediction_type
+
     # make dpmpp_sde reproducable(seed can be passed only in initializer)
     if scheduler_class is DPMSolverSDEScheduler:
         scheduler_config["noise_sampler_seed"] = seed
@@ -829,6 +834,9 @@ class DenoiseLatentsInvocation(BaseInvocation):
         seed, noise, latents = self.prepare_noise_and_latents(context, self.noise, self.latents)
         _, _, latent_height, latent_width = latents.shape
 
+        # get the unet's config so that we can pass the base to sd_step_callback()
+        unet_config = context.models.get_config(self.unet.unet.key)
+
         conditioning_data = self.get_conditioning_data(
             context=context,
             positive_conditioning_field=self.positive_conditioning,
@@ -848,6 +856,7 @@ class DenoiseLatentsInvocation(BaseInvocation):
             scheduler_info=self.unet.scheduler,
             scheduler_name=self.scheduler,
             seed=seed,
+            unet_config=unet_config,
         )
 
         timesteps, init_timestep, scheduler_step_kwargs = self.init_scheduler(
@@ -859,9 +868,6 @@ class DenoiseLatentsInvocation(BaseInvocation):
             denoising_end=self.denoising_end,
         )
 
-        # get the unet's config so that we can pass the base to sd_step_callback()
-        unet_config = context.models.get_config(self.unet.unet.key)
-
         ### preview
         def step_callback(state: PipelineIntermediateState) -> None:
             context.util.sd_step_callback(state, unet_config.base)
@@ -892,7 +898,7 @@ class DenoiseLatentsInvocation(BaseInvocation):
 
         ### inpaint
         mask, masked_latents, is_gradient_mask = self.prep_inpaint_mask(context, latents)
-        # NOTE: We used to identify inpainting models by inpecting the shape of the loaded UNet model weights. Now we
+        # NOTE: We used to identify inpainting models by inspecting the shape of the loaded UNet model weights. Now we
         # use the ModelVariantType config. During testing, there was a report of a user with models that had an
         # incorrect ModelVariantType value. Re-installing the model fixed the issue. If this issue turns out to be
         # prevalent, we will have to revisit how we initialize the inpainting extensions.
@@ -1030,6 +1036,7 @@ class DenoiseLatentsInvocation(BaseInvocation):
                 scheduler_info=self.unet.scheduler,
                 scheduler_name=self.scheduler,
                 seed=seed,
+                unet_config=unet_config,
             )
 
             pipeline = self.create_pipeline(unet, scheduler)

invokeai/app/invocations/fields.py

@@ -300,6 +300,13 @@ class BoundingBoxField(BaseModel):
             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]]):
     """

invokeai/app/invocations/flux_lora_loader.py

@@ -8,7 +8,7 @@ from invokeai.app.invocations.baseinvocation import (
     invocation_output,
 )
 from invokeai.app.invocations.fields import FieldDescriptions, Input, InputField, OutputField, UIType
-from invokeai.app.invocations.model import CLIPField, LoRAField, ModelIdentifierField, TransformerField
+from invokeai.app.invocations.model import CLIPField, LoRAField, ModelIdentifierField, T5EncoderField, TransformerField
 from invokeai.app.services.shared.invocation_context import InvocationContext
 from invokeai.backend.model_manager.config import BaseModelType
 
@@ -21,6 +21,9 @@ class FluxLoRALoaderOutput(BaseInvocationOutput):
         default=None, description=FieldDescriptions.transformer, title="FLUX Transformer"
     )
     clip: Optional[CLIPField] = OutputField(default=None, description=FieldDescriptions.clip, title="CLIP")
+    t5_encoder: Optional[T5EncoderField] = OutputField(
+        default=None, description=FieldDescriptions.t5_encoder, title="T5 Encoder"
+    )
 
 
 @invocation(
@@ -28,7 +31,7 @@ class FluxLoRALoaderOutput(BaseInvocationOutput):
     title="FLUX LoRA",
     tags=["lora", "model", "flux"],
     category="model",
-    version="1.1.0",
+    version="1.2.0",
     classification=Classification.Prototype,
 )
 class FluxLoRALoaderInvocation(BaseInvocation):
@@ -50,6 +53,12 @@ class FluxLoRALoaderInvocation(BaseInvocation):
         description=FieldDescriptions.clip,
         input=Input.Connection,
     )
+    t5_encoder: T5EncoderField | None = InputField(
+        default=None,
+        title="T5 Encoder",
+        description=FieldDescriptions.t5_encoder,
+        input=Input.Connection,
+    )
 
     def invoke(self, context: InvocationContext) -> FluxLoRALoaderOutput:
         lora_key = self.lora.key
@@ -62,6 +71,8 @@ class FluxLoRALoaderInvocation(BaseInvocation):
             raise ValueError(f'LoRA "{lora_key}" already applied to transformer.')
         if self.clip and any(lora.lora.key == lora_key for lora in self.clip.loras):
             raise ValueError(f'LoRA "{lora_key}" already applied to CLIP encoder.')
+        if self.t5_encoder and any(lora.lora.key == lora_key for lora in self.t5_encoder.loras):
+            raise ValueError(f'LoRA "{lora_key}" already applied to T5 encoder.')
 
         output = FluxLoRALoaderOutput()
 
@@ -82,6 +93,14 @@ class FluxLoRALoaderInvocation(BaseInvocation):
                     weight=self.weight,
                 )
             )
+        if self.t5_encoder is not None:
+            output.t5_encoder = self.t5_encoder.model_copy(deep=True)
+            output.t5_encoder.loras.append(
+                LoRAField(
+                    lora=self.lora,
+                    weight=self.weight,
+                )
+            )
 
         return output
 
@@ -91,14 +110,14 @@ class FluxLoRALoaderInvocation(BaseInvocation):
     title="FLUX LoRA Collection Loader",
     tags=["lora", "model", "flux"],
     category="model",
-    version="1.1.0",
+    version="1.3.0",
     classification=Classification.Prototype,
 )
 class FLUXLoRACollectionLoader(BaseInvocation):
     """Applies a collection of LoRAs to a FLUX transformer."""
 
-    loras: LoRAField | list[LoRAField] = InputField(
-        description="LoRA models and weights. May be a single LoRA or collection.", title="LoRAs"
+    loras: Optional[LoRAField | list[LoRAField]] = InputField(
+        default=None, description="LoRA models and weights. May be a single LoRA or collection.", title="LoRAs"
     )
 
     transformer: Optional[TransformerField] = InputField(
@@ -113,13 +132,30 @@ class FLUXLoRACollectionLoader(BaseInvocation):
         description=FieldDescriptions.clip,
         input=Input.Connection,
     )
+    t5_encoder: T5EncoderField | None = InputField(
+        default=None,
+        title="T5 Encoder",
+        description=FieldDescriptions.t5_encoder,
+        input=Input.Connection,
+    )
 
     def invoke(self, context: InvocationContext) -> FluxLoRALoaderOutput:
         output = FluxLoRALoaderOutput()
         loras = self.loras if isinstance(self.loras, list) else [self.loras]
         added_loras: list[str] = []
 
+        if self.transformer is not None:
+            output.transformer = self.transformer.model_copy(deep=True)
+
+        if self.clip is not None:
+            output.clip = self.clip.model_copy(deep=True)
+
+        if self.t5_encoder is not None:
+            output.t5_encoder = self.t5_encoder.model_copy(deep=True)
+
         for lora in loras:
+            if lora is None:
+                continue
             if lora.lora.key in added_loras:
                 continue
 
@@ -130,14 +166,13 @@ class FLUXLoRACollectionLoader(BaseInvocation):
 
             added_loras.append(lora.lora.key)
 
-            if self.transformer is not None:
-                if output.transformer is None:
-                    output.transformer = self.transformer.model_copy(deep=True)
+            if self.transformer is not None and output.transformer is not None:
                 output.transformer.loras.append(lora)
 
-            if self.clip is not None:
-                if output.clip is None:
-                    output.clip = self.clip.model_copy(deep=True)
+            if self.clip is not None and output.clip is not None:
                 output.clip.loras.append(lora)
 
+            if self.t5_encoder is not None and output.t5_encoder is not None:
+                output.t5_encoder.loras.append(lora)
+
         return output

invokeai/app/invocations/flux_model_loader.py

@@ -10,6 +10,10 @@ from invokeai.app.invocations.baseinvocation import (
 from invokeai.app.invocations.fields import FieldDescriptions, Input, InputField, OutputField, UIType
 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.flux.util import max_seq_lengths
 from invokeai.backend.model_manager.config import (
     CheckpointConfigBase,
@@ -36,7 +40,7 @@ class FluxModelLoaderOutput(BaseInvocationOutput):
     title="Flux Main Model",
     tags=["model", "flux"],
     category="model",
-    version="1.0.4",
+    version="1.0.5",
     classification=Classification.Prototype,
 )
 class FluxModelLoaderInvocation(BaseInvocation):
@@ -74,8 +78,8 @@ class FluxModelLoaderInvocation(BaseInvocation):
         tokenizer = self.clip_embed_model.model_copy(update={"submodel_type": SubModelType.Tokenizer})
         clip_encoder = self.clip_embed_model.model_copy(update={"submodel_type": SubModelType.TextEncoder})
 
-        tokenizer2 = self.t5_encoder_model.model_copy(update={"submodel_type": SubModelType.Tokenizer2})
-        t5_encoder = self.t5_encoder_model.model_copy(update={"submodel_type": SubModelType.TextEncoder2})
+        tokenizer2 = preprocess_t5_tokenizer_model_identifier(self.t5_encoder_model)
+        t5_encoder = preprocess_t5_encoder_model_identifier(self.t5_encoder_model)
 
         transformer_config = context.models.get_config(transformer)
         assert isinstance(transformer_config, CheckpointConfigBase)
@@ -83,7 +87,7 @@ class FluxModelLoaderInvocation(BaseInvocation):
         return FluxModelLoaderOutput(
             transformer=TransformerField(transformer=transformer, loras=[]),
             clip=CLIPField(tokenizer=tokenizer, text_encoder=clip_encoder, loras=[], skipped_layers=0),
-            t5_encoder=T5EncoderField(tokenizer=tokenizer2, text_encoder=t5_encoder),
+            t5_encoder=T5EncoderField(tokenizer=tokenizer2, text_encoder=t5_encoder, loras=[]),
             vae=VAEField(vae=vae),
             max_seq_len=max_seq_lengths[transformer_config.config_path],
         )

invokeai/app/invocations/flux_text_encoder.py

@@ -2,7 +2,7 @@ from contextlib import ExitStack
 from typing import Iterator, Literal, Optional, Tuple
 
 import torch
-from transformers import CLIPTextModel, CLIPTokenizer, T5EncoderModel, T5Tokenizer
+from transformers import CLIPTextModel, CLIPTokenizer, T5EncoderModel, T5Tokenizer, T5TokenizerFast
 
 from invokeai.app.invocations.baseinvocation import BaseInvocation, Classification, invocation
 from invokeai.app.invocations.fields import (
@@ -19,7 +19,7 @@ from invokeai.app.services.shared.invocation_context import InvocationContext
 from invokeai.backend.flux.modules.conditioner import HFEncoder
 from invokeai.backend.model_manager.config import ModelFormat
 from invokeai.backend.patches.layer_patcher import LayerPatcher
-from invokeai.backend.patches.lora_conversions.flux_lora_constants import FLUX_LORA_CLIP_PREFIX
+from invokeai.backend.patches.lora_conversions.flux_lora_constants import FLUX_LORA_CLIP_PREFIX, FLUX_LORA_T5_PREFIX
 from invokeai.backend.patches.model_patch_raw import ModelPatchRaw
 from invokeai.backend.stable_diffusion.diffusion.conditioning_data import ConditioningFieldData, FLUXConditioningInfo
 
@@ -71,12 +71,44 @@ class FluxTextEncoderInvocation(BaseInvocation):
     def _t5_encode(self, context: InvocationContext) -> torch.Tensor:
         prompt = [self.prompt]
 
+        t5_encoder_info = context.models.load(self.t5_encoder.text_encoder)
+        t5_encoder_config = t5_encoder_info.config
+        assert t5_encoder_config is not None
+
         with (
-            context.models.load(self.t5_encoder.text_encoder) as t5_text_encoder,
+            t5_encoder_info.model_on_device() as (cached_weights, t5_text_encoder),
             context.models.load(self.t5_encoder.tokenizer) as t5_tokenizer,
+            ExitStack() as exit_stack,
         ):
             assert isinstance(t5_text_encoder, T5EncoderModel)
-            assert isinstance(t5_tokenizer, T5Tokenizer)
+            assert isinstance(t5_tokenizer, (T5Tokenizer, T5TokenizerFast))
+
+            # Determine if the model is quantized.
+            # If the model is quantized, then we need to apply the LoRA weights as sidecar layers. This results in
+            # slower inference than direct patching, but is agnostic to the quantization format.
+            if t5_encoder_config.format in [ModelFormat.T5Encoder, ModelFormat.Diffusers]:
+                model_is_quantized = False
+            elif t5_encoder_config.format in [
+                ModelFormat.BnbQuantizedLlmInt8b,
+                ModelFormat.BnbQuantizednf4b,
+                ModelFormat.GGUFQuantized,
+            ]:
+                model_is_quantized = True
+            else:
+                raise ValueError(f"Unsupported model format: {t5_encoder_config.format}")
+
+            # Apply LoRA models to the T5 encoder.
+            # Note: We apply the LoRA after the encoder has been moved to its target device for faster patching.
+            exit_stack.enter_context(
+                LayerPatcher.apply_smart_model_patches(
+                    model=t5_text_encoder,
+                    patches=self._t5_lora_iterator(context),
+                    prefix=FLUX_LORA_T5_PREFIX,
+                    dtype=t5_text_encoder.dtype,
+                    cached_weights=cached_weights,
+                    force_sidecar_patching=model_is_quantized,
+                )
+            )
 
             t5_encoder = HFEncoder(t5_text_encoder, t5_tokenizer, False, self.t5_max_seq_len)
 
@@ -132,3 +164,10 @@ class FluxTextEncoderInvocation(BaseInvocation):
             assert isinstance(lora_info.model, ModelPatchRaw)
             yield (lora_info.model, lora.weight)
             del lora_info
+
+    def _t5_lora_iterator(self, context: InvocationContext) -> Iterator[Tuple[ModelPatchRaw, float]]:
+        for lora in self.t5_encoder.loras:
+            lora_info = context.models.load(lora.lora)
+            assert isinstance(lora_info.model, ModelPatchRaw)
+            yield (lora_info.model, lora.weight)
+            del lora_info

invokeai/app/invocations/flux_vae_decode.py

@@ -41,16 +41,11 @@ class FluxVaeDecodeInvocation(BaseInvocation, WithMetadata, WithBoard):
 
     def _estimate_working_memory(self, latents: torch.Tensor, vae: AutoEncoder) -> 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).
         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 = 1090  # Determined experimentally.
+        scaling_constant = 2200  # Determined experimentally.
         working_memory = out_h * out_w * element_size * scaling_constant
-
-        # We add a 20% buffer to the working memory estimate to be safe.
-        working_memory = working_memory * 1.2
         return int(working_memory)
 
     def _vae_decode(self, vae_info: LoadedModel, latents: torch.Tensor) -> Image.Image:

invokeai/app/invocations/ideal_size.py

@@ -21,7 +21,7 @@ class IdealSizeOutput(BaseInvocationOutput):
     "ideal_size",
     title="Ideal Size",
     tags=["latents", "math", "ideal_size"],
-    version="1.0.3",
+    version="1.0.4",
 )
 class IdealSizeInvocation(BaseInvocation):
     """Calculates the ideal size for generation to avoid duplication"""
@@ -41,11 +41,16 @@ class IdealSizeInvocation(BaseInvocation):
     def invoke(self, context: InvocationContext) -> IdealSizeOutput:
         unet_config = context.models.get_config(self.unet.unet.key)
         aspect = self.width / self.height
-        dimension: float = 512
-        if unet_config.base == BaseModelType.StableDiffusion2:
+
+        if unet_config.base == BaseModelType.StableDiffusion1:
+            dimension = 512
+        elif unet_config.base == BaseModelType.StableDiffusion2:
             dimension = 768
-        elif unet_config.base == BaseModelType.StableDiffusionXL:
+        elif unet_config.base in (BaseModelType.StableDiffusionXL, BaseModelType.Flux, BaseModelType.StableDiffusion3):
             dimension = 1024
+        else:
+            raise ValueError(f"Unsupported model type: {unet_config.base}")
+
         dimension = dimension * self.multiplier
         min_dimension = math.floor(dimension * 0.5)
         model_area = dimension * dimension  # hardcoded for now since all models are trained on square images

invokeai/app/invocations/image.py

@@ -13,6 +13,7 @@ from invokeai.app.invocations.baseinvocation import (
 )
 from invokeai.app.invocations.constants import IMAGE_MODES
 from invokeai.app.invocations.fields import (
+    BoundingBoxField,
     ColorField,
     FieldDescriptions,
     ImageField,
@@ -23,6 +24,7 @@ from invokeai.app.invocations.fields import (
 from invokeai.app.invocations.primitives import ImageOutput
 from invokeai.app.services.image_records.image_records_common import ImageCategory
 from invokeai.app.services.shared.invocation_context import InvocationContext
+from invokeai.app.util.misc import SEED_MAX
 from invokeai.backend.image_util.invisible_watermark import InvisibleWatermark
 from invokeai.backend.image_util.safety_checker import SafetyChecker
 
@@ -161,12 +163,12 @@ class ImagePasteInvocation(BaseInvocation, WithMetadata, WithBoard):
     crop: bool = InputField(default=False, description="Crop to base image dimensions")
 
     def invoke(self, context: InvocationContext) -> ImageOutput:
-        base_image = context.images.get_pil(self.base_image.image_name)
-        image = context.images.get_pil(self.image.image_name)
+        base_image = context.images.get_pil(self.base_image.image_name, mode="RGBA")
+        image = context.images.get_pil(self.image.image_name, mode="RGBA")
         mask = None
         if self.mask is not None:
-            mask = context.images.get_pil(self.mask.image_name)
-            mask = ImageOps.invert(mask.convert("L"))
+            mask = context.images.get_pil(self.mask.image_name, mode="L")
+            mask = ImageOps.invert(mask)
         # TODO: probably shouldn't invert mask here... should user be required to do it?
 
         min_x = min(0, self.x)
@@ -176,7 +178,11 @@ class ImagePasteInvocation(BaseInvocation, WithMetadata, WithBoard):
 
         new_image = Image.new(mode="RGBA", size=(max_x - min_x, max_y - min_y), color=(0, 0, 0, 0))
         new_image.paste(base_image, (abs(min_x), abs(min_y)))
-        new_image.paste(image, (max(0, self.x), max(0, self.y)), mask=mask)
+
+        # Create a temporary image to paste the image with transparency
+        temp_image = Image.new("RGBA", new_image.size)
+        temp_image.paste(image, (max(0, self.x), max(0, self.y)), mask=mask)
+        new_image = Image.alpha_composite(new_image, temp_image)
 
         if self.crop:
             base_w, base_h = base_image.size
@@ -301,14 +307,44 @@ class ImageBlurInvocation(BaseInvocation, WithMetadata, WithBoard):
     blur_type: Literal["gaussian", "box"] = InputField(default="gaussian", description="The type of blur")
 
     def invoke(self, context: InvocationContext) -> ImageOutput:
-        image = context.images.get_pil(self.image.image_name)
+        image = context.images.get_pil(self.image.image_name, mode="RGBA")
 
+        # Split the image into RGBA channels
+        r, g, b, a = image.split()
+
+        # Premultiply RGB channels by alpha
+        premultiplied_image = ImageChops.multiply(image, a.convert("RGBA"))
+        premultiplied_image.putalpha(a)
+
+        # Apply the blur
         blur = (
             ImageFilter.GaussianBlur(self.radius) if self.blur_type == "gaussian" else ImageFilter.BoxBlur(self.radius)
         )
-        blur_image = image.filter(blur)
+        blurred_image = premultiplied_image.filter(blur)
 
-        image_dto = context.images.save(image=blur_image)
+        # Split the blurred image into RGBA channels
+        r, g, b, a_orig = blurred_image.split()
+
+        # Convert to float using NumPy. float 32/64 division are much faster than float 16
+        r = numpy.array(r, dtype=numpy.float32)
+        g = numpy.array(g, dtype=numpy.float32)
+        b = numpy.array(b, dtype=numpy.float32)
+        a = numpy.array(a_orig, dtype=numpy.float32) / 255.0  # Normalize alpha to [0, 1]
+
+        # Unpremultiply RGB channels by alpha
+        r /= a + 1e-6  # Add a small epsilon to avoid division by zero
+        g /= a + 1e-6
+        b /= a + 1e-6
+
+        # Convert back to PIL images
+        r = Image.fromarray(numpy.uint8(numpy.clip(r, 0, 255)))
+        g = Image.fromarray(numpy.uint8(numpy.clip(g, 0, 255)))
+        b = Image.fromarray(numpy.uint8(numpy.clip(b, 0, 255)))
+
+        # Merge back into a single image
+        result_image = Image.merge("RGBA", (r, g, b, a_orig))
+
+        image_dto = context.images.save(image=result_image)
 
         return ImageOutput.build(image_dto)
 
@@ -807,7 +843,7 @@ CHANNEL_FORMATS = {
         "value",
     ],
     category="image",
-    version="1.2.2",
+    version="1.2.3",
 )
 class ImageChannelOffsetInvocation(BaseInvocation, WithMetadata, WithBoard):
     """Add or subtract a value from a specific color channel of an image."""
@@ -817,18 +853,22 @@ class ImageChannelOffsetInvocation(BaseInvocation, WithMetadata, WithBoard):
     offset: int = InputField(default=0, ge=-255, le=255, description="The amount to adjust the channel by")
 
     def invoke(self, context: InvocationContext) -> ImageOutput:
-        pil_image = context.images.get_pil(self.image.image_name)
+        image = context.images.get_pil(self.image.image_name, "RGBA")
 
         # extract the channel and mode from the input and reference tuple
         mode = CHANNEL_FORMATS[self.channel][0]
         channel_number = CHANNEL_FORMATS[self.channel][1]
 
         # Convert PIL image to new format
-        converted_image = numpy.array(pil_image.convert(mode)).astype(int)
+        converted_image = numpy.array(image.convert(mode)).astype(int)
         image_channel = converted_image[:, :, channel_number]
 
-        # Adjust the value, clipping to 0..255
-        image_channel = numpy.clip(image_channel + self.offset, 0, 255)
+        if self.channel == "Hue (HSV)":
+            # loop around the values because hue is special
+            image_channel = (image_channel + self.offset) % 256
+        else:
+            # Adjust the value, clipping to 0..255
+            image_channel = numpy.clip(image_channel + self.offset, 0, 255)
 
         # Put the channel back into the image
         converted_image[:, :, channel_number] = image_channel
@@ -836,6 +876,10 @@ class ImageChannelOffsetInvocation(BaseInvocation, WithMetadata, WithBoard):
         # Convert back to RGBA format and output
         pil_image = Image.fromarray(converted_image.astype(numpy.uint8), mode=mode).convert("RGBA")
 
+        # restore the alpha channel
+        if self.channel != "Alpha (RGBA)":
+            pil_image.putalpha(image.getchannel("A"))
+
         image_dto = context.images.save(image=pil_image)
 
         return ImageOutput.build(image_dto)
@@ -863,7 +907,7 @@ class ImageChannelOffsetInvocation(BaseInvocation, WithMetadata, WithBoard):
         "value",
     ],
     category="image",
-    version="1.2.2",
+    version="1.2.3",
 )
 class ImageChannelMultiplyInvocation(BaseInvocation, WithMetadata, WithBoard):
     """Scale a specific color channel of an image."""
@@ -874,14 +918,14 @@ class ImageChannelMultiplyInvocation(BaseInvocation, WithMetadata, WithBoard):
     invert_channel: bool = InputField(default=False, description="Invert the channel after scaling")
 
     def invoke(self, context: InvocationContext) -> ImageOutput:
-        pil_image = context.images.get_pil(self.image.image_name)
+        image = context.images.get_pil(self.image.image_name, "RGBA")
 
         # extract the channel and mode from the input and reference tuple
         mode = CHANNEL_FORMATS[self.channel][0]
         channel_number = CHANNEL_FORMATS[self.channel][1]
 
         # Convert PIL image to new format
-        converted_image = numpy.array(pil_image.convert(mode)).astype(float)
+        converted_image = numpy.array(image.convert(mode)).astype(float)
         image_channel = converted_image[:, :, channel_number]
 
         # Adjust the value, clipping to 0..255
@@ -897,6 +941,10 @@ class ImageChannelMultiplyInvocation(BaseInvocation, WithMetadata, WithBoard):
         # Convert back to RGBA format and output
         pil_image = Image.fromarray(converted_image.astype(numpy.uint8), mode=mode).convert("RGBA")
 
+        # restore the alpha channel
+        if self.channel != "Alpha (RGBA)":
+            pil_image.putalpha(image.getchannel("A"))
+
         image_dto = context.images.save(image=pil_image)
 
         return ImageOutput.build(image_dto)
@@ -962,10 +1010,10 @@ class CanvasPasteBackInvocation(BaseInvocation, WithMetadata, WithBoard):
 
 @invocation(
     "mask_from_id",
-    title="Mask from ID",
+    title="Mask from Segmented Image",
     tags=["image", "mask", "id"],
     category="image",
-    version="1.0.0",
+    version="1.0.1",
 )
 class MaskFromIDInvocation(BaseInvocation, WithMetadata, WithBoard):
     """Generate a mask for a particular color in an ID Map"""
@@ -975,40 +1023,24 @@ class MaskFromIDInvocation(BaseInvocation, WithMetadata, WithBoard):
     threshold: int = InputField(default=100, description="Threshold for color detection")
     invert: bool = InputField(default=False, description="Whether or not to invert the mask")
 
-    def rgba_to_hex(self, rgba_color: tuple[int, int, int, int]):
-        r, g, b, a = rgba_color
-        hex_code = "#{:02X}{:02X}{:02X}{:02X}".format(r, g, b, int(a * 255))
-        return hex_code
+    def invoke(self, context: InvocationContext) -> ImageOutput:
+        image = context.images.get_pil(self.image.image_name, mode="RGBA")
 
-    def id_to_mask(self, id_mask: Image.Image, color: tuple[int, int, int, int], threshold: int = 100):
-        if id_mask.mode != "RGB":
-            id_mask = id_mask.convert("RGB")
-
-        # Can directly just use the tuple but I'll leave this rgba_to_hex here
-        # incase anyone prefers using hex codes directly instead of the color picker
-        hex_color_str = self.rgba_to_hex(color)
-        rgb_color = numpy.array([int(hex_color_str[i : i + 2], 16) for i in (1, 3, 5)])
+        np_color = numpy.array(self.color.tuple())
 
         # Maybe there's a faster way to calculate this distance but I can't think of any right now.
-        color_distance = numpy.linalg.norm(id_mask - rgb_color, axis=-1)
+        color_distance = numpy.linalg.norm(image - np_color, axis=-1)
 
         # Create a mask based on the threshold and the distance calculated above
-        binary_mask = (color_distance < threshold).astype(numpy.uint8) * 255
+        binary_mask = (color_distance < self.threshold).astype(numpy.uint8) * 255
 
         # Convert the mask back to PIL
         binary_mask_pil = Image.fromarray(binary_mask)
 
-        return binary_mask_pil
-
-    def invoke(self, context: InvocationContext) -> ImageOutput:
-        image = context.images.get_pil(self.image.image_name)
-
-        mask = self.id_to_mask(image, self.color.tuple(), self.threshold)
-
         if self.invert:
-            mask = ImageOps.invert(mask)
+            binary_mask_pil = ImageOps.invert(binary_mask_pil)
 
-        image_dto = context.images.save(image=mask, image_category=ImageCategory.MASK)
+        image_dto = context.images.save(image=binary_mask_pil, image_category=ImageCategory.MASK)
 
         return ImageOutput.build(image_dto)
 
@@ -1055,3 +1087,123 @@ class CanvasV2MaskAndCropInvocation(BaseInvocation, WithMetadata, WithBoard):
             image_dto = context.images.save(image=generated_image)
 
         return ImageOutput.build(image_dto)
+
+
+@invocation(
+    "img_noise",
+    title="Add Image Noise",
+    tags=["image", "noise"],
+    category="image",
+    version="1.0.1",
+)
+class ImageNoiseInvocation(BaseInvocation, WithMetadata, WithBoard):
+    """Add noise to an image"""
+
+    image: ImageField = InputField(description="The image to add noise to")
+    seed: int = InputField(
+        default=0,
+        ge=0,
+        le=SEED_MAX,
+        description=FieldDescriptions.seed,
+    )
+    noise_type: Literal["gaussian", "salt_and_pepper"] = InputField(
+        default="gaussian",
+        description="The type of noise to add",
+    )
+    amount: float = InputField(default=0.1, ge=0, le=1, description="The amount of noise to add")
+    noise_color: bool = InputField(default=True, description="Whether to add colored noise")
+    size: int = InputField(default=1, ge=1, description="The size of the noise points")
+
+    def invoke(self, context: InvocationContext) -> ImageOutput:
+        image = context.images.get_pil(self.image.image_name, mode="RGBA")
+
+        # Save out the alpha channel
+        alpha = image.getchannel("A")
+
+        # Set the seed for numpy random
+        rs = numpy.random.RandomState(numpy.random.MT19937(numpy.random.SeedSequence(self.seed)))
+
+        if self.noise_type == "gaussian":
+            if self.noise_color:
+                noise = rs.normal(0, 1, (image.height // self.size, image.width // self.size, 3)) * 255
+            else:
+                noise = rs.normal(0, 1, (image.height // self.size, image.width // self.size)) * 255
+                noise = numpy.stack([noise] * 3, axis=-1)
+        elif self.noise_type == "salt_and_pepper":
+            if self.noise_color:
+                noise = rs.choice(
+                    [0, 255], (image.height // self.size, image.width // self.size, 3), p=[1 - self.amount, self.amount]
+                )
+            else:
+                noise = rs.choice(
+                    [0, 255], (image.height // self.size, image.width // self.size), p=[1 - self.amount, self.amount]
+                )
+                noise = numpy.stack([noise] * 3, axis=-1)
+
+        noise = Image.fromarray(noise.astype(numpy.uint8), mode="RGB").resize(
+            (image.width, image.height), Image.Resampling.NEAREST
+        )
+        noisy_image = Image.blend(image.convert("RGB"), noise, self.amount).convert("RGBA")
+
+        # Paste back the alpha channel
+        noisy_image.putalpha(alpha)
+
+        image_dto = context.images.save(image=noisy_image)
+
+        return ImageOutput.build(image_dto)
+
+
+@invocation(
+    "crop_image_to_bounding_box",
+    title="Crop Image to Bounding Box",
+    category="image",
+    version="1.0.0",
+    tags=["image", "crop"],
+    classification=Classification.Beta,
+)
+class CropImageToBoundingBoxInvocation(BaseInvocation, WithMetadata, WithBoard):
+    """Crop an image to the given bounding box. If the bounding box is omitted, the image is cropped to the non-transparent pixels."""
+
+    image: ImageField = InputField(description="The image to crop")
+    bounding_box: BoundingBoxField | None = InputField(
+        default=None, description="The bounding box to crop the image to"
+    )
+
+    def invoke(self, context: InvocationContext) -> ImageOutput:
+        image = context.images.get_pil(self.image.image_name)
+
+        bounding_box = self.bounding_box.tuple() if self.bounding_box is not None else image.getbbox()
+
+        cropped_image = image.crop(bounding_box)
+
+        image_dto = context.images.save(image=cropped_image)
+        return ImageOutput.build(image_dto)
+
+
+@invocation(
+    "paste_image_into_bounding_box",
+    title="Paste Image into Bounding Box",
+    category="image",
+    version="1.0.0",
+    tags=["image", "crop"],
+    classification=Classification.Beta,
+)
+class PasteImageIntoBoundingBoxInvocation(BaseInvocation, WithMetadata, WithBoard):
+    """Paste the source image into the target image at the given bounding box.
+
+    The source image must be the same size as the bounding box, and the bounding box must fit within the target image."""
+
+    source_image: ImageField = InputField(description="The image to paste")
+    target_image: ImageField = InputField(description="The image to paste into")
+    bounding_box: BoundingBoxField = InputField(description="The bounding box to paste the image into")
+
+    def invoke(self, context: InvocationContext) -> ImageOutput:
+        source_image = context.images.get_pil(self.source_image.image_name, mode="RGBA")
+        target_image = context.images.get_pil(self.target_image.image_name, mode="RGBA")
+
+        bounding_box = self.bounding_box.tuple()
+
+        target_image.paste(source_image, bounding_box, source_image)
+
+        image_dto = context.images.save(image=target_image)
+        return ImageOutput.build(image_dto)

invokeai/app/invocations/latents_to_image.py

@@ -60,7 +60,7 @@ class LatentsToImageInvocation(BaseInvocation, WithMetadata, WithBoard):
         # 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 = 960  # Determined experimentally.
+        scaling_constant = 2200  # Determined experimentally.
 
         if use_tiling:
             tile_size = self.tile_size
@@ -84,9 +84,7 @@ class LatentsToImageInvocation(BaseInvocation, WithMetadata, WithBoard):
             # If we are running in FP32, then we should account for the likely increase in model size (~250MB).
             working_memory += 250 * 2**20
 
-        # We add 20% to the working memory estimate to be safe.
-        working_memory = int(working_memory * 1.2)
-        return working_memory
+        return int(working_memory)
 
     @torch.no_grad()
     def invoke(self, context: InvocationContext) -> ImageOutput:

invokeai/app/invocations/load_custom_nodes.py

@@ -0,0 +1,40 @@
+import shutil
+import sys
+from importlib.util import module_from_spec, spec_from_file_location
+from pathlib import Path
+
+
+def load_custom_nodes(custom_nodes_path: Path):
+    """
+    Loads all custom nodes from the custom_nodes_path directory.
+
+    This function copies a custom __init__.py file to the custom_nodes_path directory, effectively turning it into a
+    python module.
+
+    The custom __init__.py file itself imports all the custom node packs as python modules from the custom_nodes_path
+    directory.
+
+    Then,the custom __init__.py file is programmatically imported using importlib. As it executes, it imports all the
+    custom node packs as python modules.
+    """
+    custom_nodes_path.mkdir(parents=True, exist_ok=True)
+
+    custom_nodes_init_path = str(custom_nodes_path / "__init__.py")
+    custom_nodes_readme_path = str(custom_nodes_path / "README.md")
+
+    # copy our custom nodes __init__.py to the custom nodes directory
+    shutil.copy(Path(__file__).parent / "custom_nodes/init.py", custom_nodes_init_path)
+    shutil.copy(Path(__file__).parent / "custom_nodes/README.md", custom_nodes_readme_path)
+
+    # set the same permissions as the destination directory, in case our source is read-only,
+    # so that the files are user-writable
+    for p in custom_nodes_path.glob("**/*"):
+        p.chmod(custom_nodes_path.stat().st_mode)
+
+    # Import custom nodes, see https://docs.python.org/3/library/importlib.html#importing-programmatically
+    spec = spec_from_file_location("custom_nodes", custom_nodes_init_path)
+    if spec is None or spec.loader is None:
+        raise RuntimeError(f"Could not load custom nodes from {custom_nodes_init_path}")
+    module = module_from_spec(spec)
+    sys.modules[spec.name] = module
+    spec.loader.exec_module(module)

invokeai/app/invocations/mask.py

@@ -2,9 +2,22 @@ import numpy as np
 import torch
 from PIL import Image
 
-from invokeai.app.invocations.baseinvocation import BaseInvocation, Classification, InvocationContext, invocation
-from invokeai.app.invocations.fields import ImageField, InputField, TensorField, WithBoard, WithMetadata
-from invokeai.app.invocations.primitives import ImageOutput, MaskOutput
+from invokeai.app.invocations.baseinvocation import (
+    BaseInvocation,
+    Classification,
+    InvocationContext,
+    invocation,
+)
+from invokeai.app.invocations.fields import (
+    BoundingBoxField,
+    ColorField,
+    ImageField,
+    InputField,
+    TensorField,
+    WithBoard,
+    WithMetadata,
+)
+from invokeai.app.invocations.primitives import BoundingBoxOutput, ImageOutput, MaskOutput
 from invokeai.backend.image_util.util import pil_to_np
 
 
@@ -73,7 +86,7 @@ class AlphaMaskToTensorInvocation(BaseInvocation):
     title="Invert Tensor Mask",
     tags=["conditioning"],
     category="conditioning",
-    version="1.0.0",
+    version="1.1.0",
     classification=Classification.Beta,
 )
 class InvertTensorMaskInvocation(BaseInvocation):
@@ -83,6 +96,15 @@ class InvertTensorMaskInvocation(BaseInvocation):
 
     def invoke(self, context: InvocationContext) -> MaskOutput:
         mask = context.tensors.load(self.mask.tensor_name)
+
+        # Verify dtype and shape.
+        assert mask.dtype == torch.bool
+        assert mask.dim() in [2, 3]
+
+        # Unsqueeze the channel dimension if it is missing. The MaskOutput type expects a single channel.
+        if mask.dim() == 2:
+            mask = mask.unsqueeze(0)
+
         inverted = ~mask
 
         return MaskOutput(
@@ -201,3 +223,48 @@ class ApplyMaskTensorToImageInvocation(BaseInvocation, WithMetadata, WithBoard):
         image_dto = context.images.save(image=masked_image)
 
         return ImageOutput.build(image_dto)
+
+
+WHITE = ColorField(r=255, g=255, b=255, a=255)
+
+
+@invocation(
+    "get_image_mask_bounding_box",
+    title="Get Image Mask Bounding Box",
+    tags=["mask"],
+    category="mask",
+    version="1.0.0",
+    classification=Classification.Beta,
+)
+class GetMaskBoundingBoxInvocation(BaseInvocation):
+    """Gets the bounding box of the given mask image."""
+
+    mask: ImageField = InputField(description="The mask to crop.")
+    margin: int = InputField(default=0, description="Margin to add to the bounding box.")
+    mask_color: ColorField = InputField(default=WHITE, description="Color of the mask in the image.")
+
+    def invoke(self, context: InvocationContext) -> BoundingBoxOutput:
+        mask = context.images.get_pil(self.mask.image_name, mode="RGBA")
+        mask_np = np.array(mask)
+
+        # Convert mask_color to RGBA tuple
+        mask_color_rgb = self.mask_color.tuple()
+
+        # Find the bounding box of the mask color
+        y, x = np.where(np.all(mask_np == mask_color_rgb, axis=-1))
+
+        if len(x) == 0 or len(y) == 0:
+            # No pixels found with the given color
+            return BoundingBoxOutput(bounding_box=BoundingBoxField(x_min=0, y_min=0, x_max=0, y_max=0))
+
+        left, upper, right, lower = x.min(), y.min(), x.max(), y.max()
+
+        # Add the margin
+        left = max(0, left - self.margin)
+        upper = max(0, upper - self.margin)
+        right = min(mask_np.shape[1], right + self.margin)
+        lower = min(mask_np.shape[0], lower + self.margin)
+
+        bounding_box = BoundingBoxField(x_min=left, y_min=upper, x_max=right, y_max=lower)
+
+        return BoundingBoxOutput(bounding_box=bounding_box)

invokeai/app/invocations/metadata.py

@@ -18,6 +18,7 @@ from invokeai.app.invocations.fields import (
     UIType,
 )
 from invokeai.app.invocations.model import ModelIdentifierField
+from invokeai.app.invocations.primitives import StringOutput
 from invokeai.app.services.shared.invocation_context import InvocationContext
 from invokeai.app.util.controlnet_utils import CONTROLNET_MODE_VALUES, CONTROLNET_RESIZE_VALUES
 from invokeai.version.invokeai_version import __version__
@@ -275,3 +276,34 @@ class CoreMetadataInvocation(BaseInvocation):
         return MetadataOutput(metadata=MetadataField.model_validate(as_dict))
 
     model_config = ConfigDict(extra="allow")
+
+
+@invocation(
+    "metadata_field_extractor",
+    title="Metadata Field Extractor",
+    tags=["metadata"],
+    category="metadata",
+    version="1.0.0",
+    classification=Classification.Deprecated,
+)
+class MetadataFieldExtractorInvocation(BaseInvocation):
+    """Extracts the text value from an image's metadata given a key.
+    Raises an error if the image has no metadata or if the value is not a string (nesting not permitted)."""
+
+    image: ImageField = InputField(description="The image to extract metadata from")
+    key: str = InputField(description="The key in the image's metadata to extract the value from")
+
+    def invoke(self, context: InvocationContext) -> StringOutput:
+        image_name = self.image.image_name
+
+        metadata = context.images.get_metadata(image_name=image_name)
+        if not metadata:
+            raise ValueError(f"No metadata found on image {image_name}")
+
+        try:
+            val = metadata.root[self.key]
+            if not isinstance(val, str):
+                raise ValueError(f"Metadata at key '{self.key}' must be a string")
+            return StringOutput(value=val)
+        except KeyError as e:
+            raise ValueError(f"No key '{self.key}' found in the metadata for {image_name}") from e

invokeai/app/invocations/model.py

@@ -68,6 +68,7 @@ class CLIPField(BaseModel):
 class T5EncoderField(BaseModel):
     tokenizer: ModelIdentifierField = Field(description="Info to load tokenizer submodel")
     text_encoder: ModelIdentifierField = Field(description="Info to load text_encoder submodel")
+    loras: List[LoRAField] = Field(description="LoRAs to apply on model loading")
 
 
 class VAEField(BaseModel):
@@ -205,7 +206,7 @@ class LoRALoaderInvocation(BaseInvocation):
         lora_key = self.lora.key
 
         if not context.models.exists(lora_key):
-            raise Exception(f"Unkown lora: {lora_key}!")
+            raise Exception(f"Unknown lora: {lora_key}!")
 
         if self.unet is not None and any(lora.lora.key == lora_key for lora in self.unet.loras):
             raise Exception(f'LoRA "{lora_key}" already applied to unet')
@@ -256,12 +257,12 @@ class LoRASelectorInvocation(BaseInvocation):
         return LoRASelectorOutput(lora=LoRAField(lora=self.lora, weight=self.weight))
 
 
-@invocation("lora_collection_loader", title="LoRA Collection Loader", tags=["model"], category="model", version="1.0.0")
+@invocation("lora_collection_loader", title="LoRA Collection Loader", tags=["model"], category="model", version="1.1.0")
 class LoRACollectionLoader(BaseInvocation):
     """Applies a collection of LoRAs to the provided UNet and CLIP models."""
 
-    loras: LoRAField | list[LoRAField] = InputField(
-        description="LoRA models and weights. May be a single LoRA or collection.", title="LoRAs"
+    loras: Optional[LoRAField | list[LoRAField]] = InputField(
+        default=None, description="LoRA models and weights. May be a single LoRA or collection.", title="LoRAs"
     )
     unet: Optional[UNetField] = InputField(
         default=None,
@@ -281,7 +282,14 @@ class LoRACollectionLoader(BaseInvocation):
         loras = self.loras if isinstance(self.loras, list) else [self.loras]
         added_loras: list[str] = []
 
+        if self.unet is not None:
+            output.unet = self.unet.model_copy(deep=True)
+        if self.clip is not None:
+            output.clip = self.clip.model_copy(deep=True)
+
         for lora in loras:
+            if lora is None:
+                continue
             if lora.lora.key in added_loras:
                 continue
 
@@ -292,14 +300,10 @@ class LoRACollectionLoader(BaseInvocation):
 
             added_loras.append(lora.lora.key)
 
-            if self.unet is not None:
-                if output.unet is None:
-                    output.unet = self.unet.model_copy(deep=True)
+            if self.unet is not None and output.unet is not None:
                 output.unet.loras.append(lora)
 
-            if self.clip is not None:
-                if output.clip is None:
-                    output.clip = self.clip.model_copy(deep=True)
+            if self.clip is not None and output.clip is not None:
                 output.clip.loras.append(lora)
 
         return output
@@ -399,13 +403,13 @@ class SDXLLoRALoaderInvocation(BaseInvocation):
     title="SDXL LoRA Collection Loader",
     tags=["model"],
     category="model",
-    version="1.0.0",
+    version="1.1.0",
 )
 class SDXLLoRACollectionLoader(BaseInvocation):
     """Applies a collection of SDXL LoRAs to the provided UNet and CLIP models."""
 
-    loras: LoRAField | list[LoRAField] = InputField(
-        description="LoRA models and weights. May be a single LoRA or collection.", title="LoRAs"
+    loras: Optional[LoRAField | list[LoRAField]] = InputField(
+        default=None, description="LoRA models and weights. May be a single LoRA or collection.", title="LoRAs"
     )
     unet: Optional[UNetField] = InputField(
         default=None,
@@ -431,7 +435,18 @@ class SDXLLoRACollectionLoader(BaseInvocation):
         loras = self.loras if isinstance(self.loras, list) else [self.loras]
         added_loras: list[str] = []
 
+        if self.unet is not None:
+            output.unet = self.unet.model_copy(deep=True)
+
+        if self.clip is not None:
+            output.clip = self.clip.model_copy(deep=True)
+
+        if self.clip2 is not None:
+            output.clip2 = self.clip2.model_copy(deep=True)
+
         for lora in loras:
+            if lora is None:
+                continue
             if lora.lora.key in added_loras:
                 continue
 
@@ -442,19 +457,13 @@ class SDXLLoRACollectionLoader(BaseInvocation):
 
             added_loras.append(lora.lora.key)
 
-            if self.unet is not None:
-                if output.unet is None:
-                    output.unet = self.unet.model_copy(deep=True)
+            if self.unet is not None and output.unet is not None:
                 output.unet.loras.append(lora)
 
-            if self.clip is not None:
-                if output.clip is None:
-                    output.clip = self.clip.model_copy(deep=True)
+            if self.clip is not None and output.clip is not None:
                 output.clip.loras.append(lora)
 
-            if self.clip2 is not None:
-                if output.clip2 is None:
-                    output.clip2 = self.clip2.model_copy(deep=True)
+            if self.clip2 is not None and output.clip2 is not None:
                 output.clip2.loras.append(lora)
 
         return output
@@ -472,7 +481,7 @@ class VAELoaderInvocation(BaseInvocation):
         key = self.vae_model.key
 
         if not context.models.exists(key):
-            raise Exception(f"Unkown vae: {key}!")
+            raise Exception(f"Unknown vae: {key}!")
 
         return VAEOutput(vae=VAEField(vae=self.vae_model))
 

invokeai/app/invocations/primitives.py

@@ -7,7 +7,6 @@ import torch
 from invokeai.app.invocations.baseinvocation import (
     BaseInvocation,
     BaseInvocationOutput,
-    Classification,
     invocation,
     invocation_output,
 )
@@ -266,13 +265,9 @@ class ImageInvocation(BaseInvocation):
     image: ImageField = InputField(description="The image to load")
 
     def invoke(self, context: InvocationContext) -> ImageOutput:
-        image = context.images.get_pil(self.image.image_name)
+        image_dto = context.images.get_dto(self.image.image_name)
 
-        return ImageOutput(
-            image=ImageField(image_name=self.image.image_name),
-            width=image.width,
-            height=image.height,
-        )
+        return ImageOutput.build(image_dto=image_dto)
 
 
 @invocation(
@@ -417,6 +412,7 @@ class ColorInvocation(BaseInvocation):
 class MaskOutput(BaseInvocationOutput):
     """A torch mask tensor."""
 
+    # shape: [1, H, W], dtype: bool
     mask: TensorField = OutputField(description="The mask.")
     width: int = OutputField(description="The width of the mask in pixels.")
     height: int = OutputField(description="The height of the mask in pixels.")
@@ -539,23 +535,3 @@ class BoundingBoxInvocation(BaseInvocation):
 
 
 # endregion
-
-
-@invocation(
-    "image_batch",
-    title="Image Batch",
-    tags=["primitives", "image", "batch", "internal"],
-    category="primitives",
-    version="1.0.0",
-    classification=Classification.Special,
-)
-class ImageBatchInvocation(BaseInvocation):
-    """Create a batched generation, where the workflow is executed once for each image in the batch."""
-
-    images: list[ImageField] = InputField(min_length=1, description="The images to batch over", input=Input.Direct)
-
-    def __init__(self):
-        raise NotImplementedError("This class should never be executed or instantiated directly.")
-
-    def invoke(self, context: InvocationContext) -> ImageOutput:
-        raise NotImplementedError("This class should never be executed or instantiated directly.")

invokeai/app/invocations/sd3_latents_to_image.py

@@ -43,16 +43,11 @@ class SD3LatentsToImageInvocation(BaseInvocation, WithMetadata, WithBoard):
 
     def _estimate_working_memory(self, latents: torch.Tensor, vae: AutoencoderKL) -> 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).
         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 = 1230  # Determined experimentally.
+        scaling_constant = 2200  # Determined experimentally.
         working_memory = out_h * out_w * element_size * scaling_constant
-
-        # We add a 20% buffer to the working memory estimate to be safe.
-        working_memory = working_memory * 1.2
         return int(working_memory)
 
     @torch.no_grad()

invokeai/app/invocations/sd3_model_loader.py

@@ -10,6 +10,10 @@ from invokeai.app.invocations.baseinvocation import (
 from invokeai.app.invocations.fields import FieldDescriptions, Input, InputField, OutputField, UIType
 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.config import SubModelType
 
 
@@ -88,21 +92,13 @@ class Sd3ModelLoaderInvocation(BaseInvocation):
             if self.clip_g_model
             else self.model.model_copy(update={"submodel_type": SubModelType.TextEncoder2})
         )
-        tokenizer_t5 = (
-            self.t5_encoder_model.model_copy(update={"submodel_type": SubModelType.Tokenizer3})
-            if self.t5_encoder_model
-            else self.model.model_copy(update={"submodel_type": SubModelType.Tokenizer3})
-        )
-        t5_encoder = (
-            self.t5_encoder_model.model_copy(update={"submodel_type": SubModelType.TextEncoder3})
-            if self.t5_encoder_model
-            else self.model.model_copy(update={"submodel_type": SubModelType.TextEncoder3})
-        )
+        tokenizer_t5 = preprocess_t5_tokenizer_model_identifier(self.t5_encoder_model or self.model)
+        t5_encoder = preprocess_t5_encoder_model_identifier(self.t5_encoder_model or self.model)
 
         return Sd3ModelLoaderOutput(
             transformer=TransformerField(transformer=transformer, loras=[]),
             clip_l=CLIPField(tokenizer=tokenizer_l, text_encoder=clip_encoder_l, loras=[], skipped_layers=0),
             clip_g=CLIPField(tokenizer=tokenizer_g, text_encoder=clip_encoder_g, loras=[], skipped_layers=0),
-            t5_encoder=T5EncoderField(tokenizer=tokenizer_t5, text_encoder=t5_encoder),
+            t5_encoder=T5EncoderField(tokenizer=tokenizer_t5, text_encoder=t5_encoder, loras=[]),
             vae=VAEField(vae=vae),
         )

invokeai/app/invocations/segment_anything.py

@@ -49,7 +49,7 @@ class SAMPointsField(BaseModel):
     title="Segment Anything",
     tags=["prompt", "segmentation"],
     category="segmentation",
-    version="1.1.0",
+    version="1.2.0",
 )
 class SegmentAnythingInvocation(BaseInvocation):
     """Runs a Segment Anything Model."""
@@ -96,8 +96,10 @@ class SegmentAnythingInvocation(BaseInvocation):
             # masks contains bool values, so we merge them via max-reduce.
             combined_mask, _ = torch.stack(masks).max(dim=0)
 
+        # Unsqueeze the channel dimension.
+        combined_mask = combined_mask.unsqueeze(0)
         mask_tensor_name = context.tensors.save(combined_mask)
-        height, width = combined_mask.shape
+        _, height, width = combined_mask.shape
         return MaskOutput(mask=TensorField(tensor_name=mask_tensor_name), width=width, height=height)
 
     @staticmethod

invokeai/app/invocations/tiled_multi_diffusion_denoise_latents.py

@@ -218,6 +218,7 @@ class TiledMultiDiffusionDenoiseLatents(BaseInvocation):
                 scheduler_info=self.unet.scheduler,
                 scheduler_name=self.scheduler,
                 seed=seed,
+                unet_config=unet_config,
             )
             pipeline = self.create_pipeline(unet=unet, scheduler=scheduler)
 

invokeai/app/invocations/util.py

@@ -9,6 +9,6 @@ def validate_weights(weights: Union[float, list[float]]) -> None:
 
 
 def validate_begin_end_step(begin_step_percent: float, end_step_percent: float) -> None:
-    """Validate that begin_step_percent is less than end_step_percent"""
-    if begin_step_percent >= end_step_percent:
+    """Validate that begin_step_percent is less than or equal to end_step_percent"""
+    if begin_step_percent > end_step_percent:
         raise ValueError("Begin step percent must be less than or equal to end step percent")

invokeai/app/run_app.py

@@ -1,12 +1,82 @@
-"""This is a wrapper around the main app entrypoint, to allow for CLI args to be parsed before running the app."""
+import uvicorn
+
+from invokeai.app.invocations.load_custom_nodes import load_custom_nodes
+from invokeai.app.services.config.config_default import get_config
+from invokeai.app.util.torch_cuda_allocator import configure_torch_cuda_allocator
+from invokeai.backend.util.logging import InvokeAILogger
+from invokeai.frontend.cli.arg_parser import InvokeAIArgs
+
+
+def get_app():
+    """Import the app and event loop. We wrap this in a function to more explicitly control when it happens, because
+    importing from api_app does a bunch of stuff - it's more like calling a function than importing a module.
+    """
+    from invokeai.app.api_app import app, loop
+
+    return app, loop
 
 
 def run_app() -> None:
-    # Before doing _anything_, parse CLI args!
-    from invokeai.frontend.cli.arg_parser import InvokeAIArgs
-
+    """The main entrypoint for the app."""
+    # Parse the CLI arguments.
     InvokeAIArgs.parse_args()
 
-    from invokeai.app.api_app import invoke_api
+    # Load config.
+    app_config = get_config()
 
-    invoke_api()
+    logger = InvokeAILogger.get_logger(config=app_config)
+
+    # Configure the torch CUDA memory allocator.
+    # NOTE: It is important that this happens before torch is imported.
+    if app_config.pytorch_cuda_alloc_conf:
+        configure_torch_cuda_allocator(app_config.pytorch_cuda_alloc_conf, logger)
+
+    # Import from startup_utils here to avoid importing torch before configure_torch_cuda_allocator() is called.
+    from invokeai.app.util.startup_utils import (
+        apply_monkeypatches,
+        check_cudnn,
+        enable_dev_reload,
+        find_open_port,
+        register_mime_types,
+    )
+
+    # Find an open port, and modify the config accordingly.
+    orig_config_port = app_config.port
+    app_config.port = find_open_port(app_config.port)
+    if orig_config_port != app_config.port:
+        logger.warning(f"Port {orig_config_port} is already in use. Using port {app_config.port}.")
+
+    # Miscellaneous startup tasks.
+    apply_monkeypatches()
+    register_mime_types()
+    if app_config.dev_reload:
+        enable_dev_reload()
+    check_cudnn(logger)
+
+    # Initialize the app and event loop.
+    app, loop = get_app()
+
+    # Load custom nodes. This must be done after importing the Graph class, which itself imports all modules from the
+    # invocations module. The ordering here is implicit, but important - we want to load custom nodes after all the
+    # core nodes have been imported so that we can catch when a custom node clobbers a core node.
+    load_custom_nodes(custom_nodes_path=app_config.custom_nodes_path)
+
+    # Start the server.
+    config = uvicorn.Config(
+        app=app,
+        host=app_config.host,
+        port=app_config.port,
+        loop="asyncio",
+        log_level=app_config.log_level_network,
+        ssl_certfile=app_config.ssl_certfile,
+        ssl_keyfile=app_config.ssl_keyfile,
+    )
+    server = uvicorn.Server(config)
+
+    # replace uvicorn's loggers with InvokeAI's for consistent appearance
+    uvicorn_logger = InvokeAILogger.get_logger("uvicorn")
+    uvicorn_logger.handlers.clear()
+    for hdlr in logger.handlers:
+        uvicorn_logger.addHandler(hdlr)
+
+    loop.run_until_complete(server.serve())

invokeai/app/services/board_image_records/board_image_records_base.py

@@ -1,6 +1,8 @@
 from abc import ABC, abstractmethod
 from typing import Optional
 
+from invokeai.app.services.image_records.image_records_common import ImageCategory
+
 
 class BoardImageRecordStorageBase(ABC):
     """Abstract base class for the one-to-many board-image relationship record storage."""
@@ -26,6 +28,8 @@ class BoardImageRecordStorageBase(ABC):
     def get_all_board_image_names_for_board(
         self,
         board_id: str,
+        categories: list[ImageCategory] | None,
+        is_intermediate: bool | None,
     ) -> list[str]:
         """Gets all board images for a board, as a list of the image names."""
         pass

invokeai/app/services/board_image_records/board_image_records_sqlite.py

@@ -1,23 +1,20 @@
 import sqlite3
-import threading
 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 ImageRecord, deserialize_image_record
+from invokeai.app.services.image_records.image_records_common import (
+    ImageCategory,
+    ImageRecord,
+    deserialize_image_record,
+)
 from invokeai.app.services.shared.pagination import OffsetPaginatedResults
 from invokeai.app.services.shared.sqlite.sqlite_database import SqliteDatabase
 
 
 class SqliteBoardImageRecordStorage(BoardImageRecordStorageBase):
-    _conn: sqlite3.Connection
-    _cursor: sqlite3.Cursor
-    _lock: threading.RLock
-
     def __init__(self, db: SqliteDatabase) -> None:
         super().__init__()
-        self._lock = db.lock
         self._conn = db.conn
-        self._cursor = self._conn.cursor()
 
     def add_image_to_board(
         self,
@@ -25,8 +22,8 @@ class SqliteBoardImageRecordStorage(BoardImageRecordStorageBase):
         image_name: str,
     ) -> None:
         try:
-            self._lock.acquire()
-            self._cursor.execute(
+            cursor = self._conn.cursor()
+            cursor.execute(
                 """--sql
                 INSERT INTO board_images (board_id, image_name)
                 VALUES (?, ?)
@@ -38,16 +35,14 @@ class SqliteBoardImageRecordStorage(BoardImageRecordStorageBase):
         except sqlite3.Error as e:
             self._conn.rollback()
             raise e
-        finally:
-            self._lock.release()
 
     def remove_image_from_board(
         self,
         image_name: str,
     ) -> None:
         try:
-            self._lock.acquire()
-            self._cursor.execute(
+            cursor = self._conn.cursor()
+            cursor.execute(
                 """--sql
                 DELETE FROM board_images
                 WHERE image_name = ?;
@@ -58,8 +53,6 @@ class SqliteBoardImageRecordStorage(BoardImageRecordStorageBase):
         except sqlite3.Error as e:
             self._conn.rollback()
             raise e
-        finally:
-            self._lock.release()
 
     def get_images_for_board(
         self,
@@ -68,96 +61,108 @@ class SqliteBoardImageRecordStorage(BoardImageRecordStorageBase):
         limit: int = 10,
     ) -> OffsetPaginatedResults[ImageRecord]:
         # TODO: this isn't paginated yet?
-        try:
-            self._lock.acquire()
-            self._cursor.execute(
-                """--sql
-                SELECT images.*
-                FROM board_images
-                INNER JOIN images ON board_images.image_name = images.image_name
-                WHERE board_images.board_id = ?
-                ORDER BY board_images.updated_at DESC;
-                """,
-                (board_id,),
-            )
-            result = cast(list[sqlite3.Row], self._cursor.fetchall())
-            images = [deserialize_image_record(dict(r)) for r in result]
+        cursor = self._conn.cursor()
+        cursor.execute(
+            """--sql
+            SELECT images.*
+            FROM board_images
+            INNER JOIN images ON board_images.image_name = images.image_name
+            WHERE board_images.board_id = ?
+            ORDER BY board_images.updated_at DESC;
+            """,
+            (board_id,),
+        )
+        result = cast(list[sqlite3.Row], cursor.fetchall())
+        images = [deserialize_image_record(dict(r)) for r in result]
 
-            self._cursor.execute(
-                """--sql
-                SELECT COUNT(*) FROM images WHERE 1=1;
-                """
-            )
-            count = cast(int, self._cursor.fetchone()[0])
+        cursor.execute(
+            """--sql
+            SELECT COUNT(*) FROM images WHERE 1=1;
+            """
+        )
+        count = cast(int, cursor.fetchone()[0])
 
-        except sqlite3.Error as e:
-            self._conn.rollback()
-            raise e
-        finally:
-            self._lock.release()
         return OffsetPaginatedResults(items=images, offset=offset, limit=limit, total=count)
 
-    def get_all_board_image_names_for_board(self, board_id: str) -> list[str]:
-        try:
-            self._lock.acquire()
-            self._cursor.execute(
-                """--sql
-                SELECT image_name
-                FROM board_images
-                WHERE board_id = ?;
-                """,
-                (board_id,),
-            )
-            result = cast(list[sqlite3.Row], self._cursor.fetchall())
-            image_names = [r[0] for r in result]
-            return image_names
-        except sqlite3.Error as e:
-            self._conn.rollback()
-            raise e
-        finally:
-            self._lock.release()
+    def get_all_board_image_names_for_board(
+        self,
+        board_id: str,
+        categories: list[ImageCategory] | None,
+        is_intermediate: bool | None,
+    ) -> list[str]:
+        params: list[str | bool] = []
+
+        # Base query is a join between images and board_images
+        stmt = """
+                SELECT images.image_name
+                FROM images
+                LEFT JOIN board_images ON board_images.image_name = images.image_name
+                WHERE 1=1
+                AND board_images.board_id = ?
+                """
+        params.append(board_id)
+
+        # Add the category filter
+        if categories is not None:
+            # Convert the enum values to unique list of strings
+            category_strings = [c.value for c in set(categories)]
+            # Create the correct length of placeholders
+            placeholders = ",".join("?" * len(category_strings))
+            stmt += f"""--sql
+                AND images.image_category IN ( {placeholders} )
+                """
+
+            # Unpack the included categories into the query params
+            for c in category_strings:
+                params.append(c)
+
+        # Add the is_intermediate filter
+        if is_intermediate is not None:
+            stmt += """--sql
+                AND images.is_intermediate = ?
+                """
+            params.append(is_intermediate)
+
+        # Put a ring on it
+        stmt += ";"
+
+        # Execute the query
+        cursor = self._conn.cursor()
+        cursor.execute(stmt, params)
+
+        result = cast(list[sqlite3.Row], cursor.fetchall())
+        image_names = [r[0] for r in result]
+        return image_names
 
     def get_board_for_image(
         self,
         image_name: str,
     ) -> Optional[str]:
-        try:
-            self._lock.acquire()
-            self._cursor.execute(
-                """--sql
+        cursor = self._conn.cursor()
+        cursor.execute(
+            """--sql
                 SELECT board_id
                 FROM board_images
                 WHERE image_name = ?;
                 """,
-                (image_name,),
-            )
-            result = self._cursor.fetchone()
-            if result is None:
-                return None
-            return cast(str, result[0])
-        except sqlite3.Error as e:
-            self._conn.rollback()
-            raise e
-        finally:
-            self._lock.release()
+            (image_name,),
+        )
+        result = cursor.fetchone()
+        if result is None:
+            return None
+        return cast(str, result[0])
 
     def get_image_count_for_board(self, board_id: str) -> int:
-        try:
-            self._lock.acquire()
-            self._cursor.execute(
-                """--sql
+        cursor = self._conn.cursor()
+        cursor.execute(
+            """--sql
                 SELECT COUNT(*)
                 FROM board_images
                 INNER JOIN images ON board_images.image_name = images.image_name
                 WHERE images.is_intermediate = FALSE
                 AND board_images.board_id = ?;
                 """,
-                (board_id,),
-            )
-            count = cast(int, self._cursor.fetchone()[0])
-            return count
-        except sqlite3.Error as e:
-            self._conn.rollback()
-            raise e
-        finally:
-            self._lock.release()
+            (board_id,),
+        )
+        count = cast(int, cursor.fetchone()[0])
+        return count

invokeai/app/services/board_images/board_images_base.py

@@ -1,6 +1,8 @@
 from abc import ABC, abstractmethod
 from typing import Optional
 
+from invokeai.app.services.image_records.image_records_common import ImageCategory
+
 
 class BoardImagesServiceABC(ABC):
     """High-level service for board-image relationship management."""
@@ -26,6 +28,8 @@ class BoardImagesServiceABC(ABC):
     def get_all_board_image_names_for_board(
         self,
         board_id: str,
+        categories: list[ImageCategory] | None,
+        is_intermediate: bool | None,
     ) -> list[str]:
         """Gets all board images for a board, as a list of the image names."""
         pass

invokeai/app/services/board_images/board_images_default.py

@@ -1,6 +1,7 @@
 from typing import Optional
 
 from invokeai.app.services.board_images.board_images_base import BoardImagesServiceABC
+from invokeai.app.services.image_records.image_records_common import ImageCategory
 from invokeai.app.services.invoker import Invoker
 
 
@@ -26,8 +27,14 @@ class BoardImagesService(BoardImagesServiceABC):
     def get_all_board_image_names_for_board(
         self,
         board_id: str,
+        categories: list[ImageCategory] | None,
+        is_intermediate: bool | None,
     ) -> list[str]:
-        return self.__invoker.services.board_image_records.get_all_board_image_names_for_board(board_id)
+        return self.__invoker.services.board_image_records.get_all_board_image_names_for_board(
+            board_id,
+            categories,
+            is_intermediate,
+        )
 
     def get_board_for_image(
         self,

invokeai/app/services/board_records/board_records_sqlite.py

@@ -1,5 +1,4 @@
 import sqlite3
-import threading
 from typing import Union, cast
 
 from invokeai.app.services.board_records.board_records_base import BoardRecordStorageBase
@@ -19,20 +18,14 @@ from invokeai.app.util.misc import uuid_string
 
 
 class SqliteBoardRecordStorage(BoardRecordStorageBase):
-    _conn: sqlite3.Connection
-    _cursor: sqlite3.Cursor
-    _lock: threading.RLock
-
     def __init__(self, db: SqliteDatabase) -> None:
         super().__init__()
-        self._lock = db.lock
         self._conn = db.conn
-        self._cursor = self._conn.cursor()
 
     def delete(self, board_id: str) -> None:
         try:
-            self._lock.acquire()
-            self._cursor.execute(
+            cursor = self._conn.cursor()
+            cursor.execute(
                 """--sql
                 DELETE FROM boards
                 WHERE board_id = ?;
@@ -40,14 +33,9 @@ class SqliteBoardRecordStorage(BoardRecordStorageBase):
                 (board_id,),
             )
             self._conn.commit()
-        except sqlite3.Error as e:
-            self._conn.rollback()
-            raise BoardRecordDeleteException from e
         except Exception as e:
             self._conn.rollback()
             raise BoardRecordDeleteException from e
-        finally:
-            self._lock.release()
 
     def save(
         self,
@@ -55,8 +43,8 @@ class SqliteBoardRecordStorage(BoardRecordStorageBase):
     ) -> BoardRecord:
         try:
             board_id = uuid_string()
-            self._lock.acquire()
-            self._cursor.execute(
+            cursor = self._conn.cursor()
+            cursor.execute(
                 """--sql
                 INSERT OR IGNORE INTO boards (board_id, board_name)
                 VALUES (?, ?);
@@ -67,8 +55,6 @@ class SqliteBoardRecordStorage(BoardRecordStorageBase):
         except sqlite3.Error as e:
             self._conn.rollback()
             raise BoardRecordSaveException from e
-        finally:
-            self._lock.release()
         return self.get(board_id)
 
     def get(
@@ -76,8 +62,8 @@ class SqliteBoardRecordStorage(BoardRecordStorageBase):
         board_id: str,
     ) -> BoardRecord:
         try:
-            self._lock.acquire()
-            self._cursor.execute(
+            cursor = self._conn.cursor()
+            cursor.execute(
                 """--sql
                 SELECT *
                 FROM boards
@@ -86,12 +72,9 @@ class SqliteBoardRecordStorage(BoardRecordStorageBase):
                 (board_id,),
             )
 
-            result = cast(Union[sqlite3.Row, None], self._cursor.fetchone())
+            result = cast(Union[sqlite3.Row, None], cursor.fetchone())
         except sqlite3.Error as e:
-            self._conn.rollback()
             raise BoardRecordNotFoundException from e
-        finally:
-            self._lock.release()
         if result is None:
             raise BoardRecordNotFoundException
         return BoardRecord(**dict(result))
@@ -102,11 +85,10 @@ class SqliteBoardRecordStorage(BoardRecordStorageBase):
         changes: BoardChanges,
     ) -> BoardRecord:
         try:
-            self._lock.acquire()
-
+            cursor = self._conn.cursor()
             # Change the name of a board
             if changes.board_name is not None:
-                self._cursor.execute(
+                cursor.execute(
                     """--sql
                     UPDATE boards
                     SET board_name = ?
@@ -117,7 +99,7 @@ class SqliteBoardRecordStorage(BoardRecordStorageBase):
 
             # Change the cover image of a board
             if changes.cover_image_name is not None:
-                self._cursor.execute(
+                cursor.execute(
                     """--sql
                     UPDATE boards
                     SET cover_image_name = ?
@@ -128,7 +110,7 @@ class SqliteBoardRecordStorage(BoardRecordStorageBase):
 
             # Change the archived status of a board
             if changes.archived is not None:
-                self._cursor.execute(
+                cursor.execute(
                     """--sql
                     UPDATE boards
                     SET archived = ?
@@ -141,8 +123,6 @@ class SqliteBoardRecordStorage(BoardRecordStorageBase):
         except sqlite3.Error as e:
             self._conn.rollback()
             raise BoardRecordSaveException from e
-        finally:
-            self._lock.release()
         return self.get(board_id)
 
     def get_many(
@@ -153,11 +133,10 @@ class SqliteBoardRecordStorage(BoardRecordStorageBase):
         limit: int = 10,
         include_archived: bool = False,
     ) -> OffsetPaginatedResults[BoardRecord]:
-        try:
-            self._lock.acquire()
+        cursor = self._conn.cursor()
 
-            # Build base query
-            base_query = """
+        # Build base query
+        base_query = """
                 SELECT *
                 FROM boards
                 {archived_filter}
@@ -165,81 +144,67 @@ class SqliteBoardRecordStorage(BoardRecordStorageBase):
                 LIMIT ? OFFSET ?;
             """
 
-            # Determine archived filter condition
-            archived_filter = "" if include_archived else "WHERE archived = 0"
+        # Determine archived filter condition
+        archived_filter = "" if include_archived else "WHERE archived = 0"
 
-            final_query = base_query.format(
-                archived_filter=archived_filter, order_by=order_by.value, direction=direction.value
-            )
+        final_query = base_query.format(
+            archived_filter=archived_filter, order_by=order_by.value, direction=direction.value
+        )
 
-            # Execute query to fetch boards
-            self._cursor.execute(final_query, (limit, offset))
+        # Execute query to fetch boards
+        cursor.execute(final_query, (limit, offset))
 
-            result = cast(list[sqlite3.Row], self._cursor.fetchall())
-            boards = [deserialize_board_record(dict(r)) for r in result]
+        result = cast(list[sqlite3.Row], cursor.fetchall())
+        boards = [deserialize_board_record(dict(r)) for r in result]
 
-            # Determine count query
-            if include_archived:
-                count_query = """
+        # Determine count query
+        if include_archived:
+            count_query = """
                     SELECT COUNT(*)
                     FROM boards;
                 """
-            else:
-                count_query = """
+        else:
+            count_query = """
                     SELECT COUNT(*)
                     FROM boards
                     WHERE archived = 0;
                 """
 
-            # Execute count query
-            self._cursor.execute(count_query)
+        # Execute count query
+        cursor.execute(count_query)
 
-            count = cast(int, self._cursor.fetchone()[0])
+        count = cast(int, cursor.fetchone()[0])
 
-            return OffsetPaginatedResults[BoardRecord](items=boards, offset=offset, limit=limit, total=count)
-
-        except sqlite3.Error as e:
-            self._conn.rollback()
-            raise e
-        finally:
-            self._lock.release()
+        return OffsetPaginatedResults[BoardRecord](items=boards, offset=offset, limit=limit, total=count)
 
     def get_all(
         self, order_by: BoardRecordOrderBy, direction: SQLiteDirection, include_archived: bool = False
     ) -> list[BoardRecord]:
-        try:
-            self._lock.acquire()
-
-            if order_by == BoardRecordOrderBy.Name:
-                base_query = """
+        cursor = self._conn.cursor()
+        if order_by == BoardRecordOrderBy.Name:
+            base_query = """
                     SELECT *
                     FROM boards
                     {archived_filter}
                     ORDER BY LOWER(board_name) {direction}
                 """
-            else:
-                base_query = """
+        else:
+            base_query = """
                     SELECT *
                     FROM boards
                     {archived_filter}
                     ORDER BY {order_by} {direction}
                 """
 
-            archived_filter = "" if include_archived else "WHERE archived = 0"
+        archived_filter = "" if include_archived else "WHERE archived = 0"
 
-            final_query = base_query.format(
-                archived_filter=archived_filter, order_by=order_by.value, direction=direction.value
-            )
+        final_query = base_query.format(
+            archived_filter=archived_filter, order_by=order_by.value, direction=direction.value
+        )
 
-            self._cursor.execute(final_query)
+        cursor.execute(final_query)
 
-            result = cast(list[sqlite3.Row], self._cursor.fetchall())
-            boards = [deserialize_board_record(dict(r)) for r in result]
+        result = cast(list[sqlite3.Row], cursor.fetchall())
+        boards = [deserialize_board_record(dict(r)) for r in result]
 
-            return boards
-
-        except sqlite3.Error as e:
-            self._conn.rollback()
-            raise e
-        finally:
-            self._lock.release()
+        return boards

invokeai/app/services/bulk_download/bulk_download_default.py

@@ -63,7 +63,11 @@ class BulkDownloadService(BulkDownloadBase):
         return [self._invoker.services.images.get_dto(image_name) for image_name in image_names]
 
     def _board_handler(self, board_id: str) -> list[ImageDTO]:
-        image_names = self._invoker.services.board_image_records.get_all_board_image_names_for_board(board_id)
+        image_names = self._invoker.services.board_image_records.get_all_board_image_names_for_board(
+            board_id,
+            categories=None,
+            is_intermediate=None,
+        )
         return self._image_handler(image_names)
 
     def generate_item_id(self, board_id: Optional[str]) -> str:

invokeai/app/services/config/config_default.py

@@ -87,9 +87,11 @@ class InvokeAIAppConfig(BaseSettings):
         log_memory_usage: If True, a memory snapshot will be captured before and after every model cache operation, and the result will be logged (at debug level). There is a time cost to capturing the memory snapshots, so it is recommended to only enable this feature if you are actively inspecting the model cache's behaviour.
         device_working_mem_gb: The amount of working memory to keep available on the compute device (in GB). Has no effect if running on CPU. If you are experiencing OOM errors, try increasing this value.
         enable_partial_loading: Enable partial loading of models. This enables models to run with reduced VRAM requirements (at the cost of slower speed) by streaming the model from RAM to VRAM as its used. In some edge cases, partial loading can cause models to run more slowly if they were previously being fully loaded into VRAM.
+        keep_ram_copy_of_weights: Whether to keep a full RAM copy of a model's weights when the model is loaded in VRAM. Keeping a RAM copy increases average RAM usage, but speeds up model switching and LoRA patching (assuming there is sufficient RAM). Set this to False if RAM pressure is consistently high.
         ram: DEPRECATED: This setting is no longer used. It has been replaced by `max_cache_ram_gb`, but most users will not need to use this config since automatic cache size limits should work well in most cases. This config setting will be removed once the new model cache behavior is stable.
         vram: DEPRECATED: This setting is no longer used. It has been replaced by `max_cache_vram_gb`, but most users will not need to use this config since automatic cache size limits should work well in most cases. This config setting will be removed once the new model cache behavior is stable.
         lazy_offload: DEPRECATED: This setting is no longer used. Lazy-offloading is enabled by default. This config setting will be removed once the new model cache behavior is stable.
+        pytorch_cuda_alloc_conf: Configure the Torch CUDA memory allocator. This will impact peak reserved VRAM usage and performance. Setting to "backend:cudaMallocAsync" works well on many systems. The optimal configuration is highly dependent on the system configuration (device type, VRAM, CUDA driver version, etc.), so must be tuned experimentally.
         device: Preferred execution device. `auto` will choose the device depending on the hardware platform and the installed torch capabilities.<br>Valid values: `auto`, `cpu`, `cuda`, `cuda:1`, `mps`
         precision: Floating point precision. `float16` will consume half the memory of `float32` but produce slightly lower-quality images. The `auto` setting will guess the proper precision based on your video card and operating system.<br>Valid values: `auto`, `float16`, `bfloat16`, `float32`
         sequential_guidance: Whether to calculate guidance in serial instead of in parallel, lowering memory requirements.
@@ -162,11 +164,15 @@ class InvokeAIAppConfig(BaseSettings):
     log_memory_usage:              bool = Field(default=False,              description="If True, a memory snapshot will be captured before and after every model cache operation, and the result will be logged (at debug level). There is a time cost to capturing the memory snapshots, so it is recommended to only enable this feature if you are actively inspecting the model cache's behaviour.")
     device_working_mem_gb:        float = Field(default=3,                  description="The amount of working memory to keep available on the compute device (in GB). Has no effect if running on CPU. If you are experiencing OOM errors, try increasing this value.")
     enable_partial_loading:        bool = Field(default=False,              description="Enable partial loading of models. This enables models to run with reduced VRAM requirements (at the cost of slower speed) by streaming the model from RAM to VRAM as its used. In some edge cases, partial loading can cause models to run more slowly if they were previously being fully loaded into VRAM.")
+    keep_ram_copy_of_weights:      bool = Field(default=True,              description="Whether to keep a full RAM copy of a model's weights when the model is loaded in VRAM. Keeping a RAM copy increases average RAM usage, but speeds up model switching and LoRA patching (assuming there is sufficient RAM). Set this to False if RAM pressure is consistently high.")
     # Deprecated CACHE configs
     ram:                Optional[float] = Field(default=None, gt=0,         description="DEPRECATED: This setting is no longer used. It has been replaced by `max_cache_ram_gb`, but most users will not need to use this config since automatic cache size limits should work well in most cases. This config setting will be removed once the new model cache behavior is stable.")
     vram:               Optional[float] = Field(default=None, ge=0,         description="DEPRECATED: This setting is no longer used. It has been replaced by `max_cache_vram_gb`, but most users will not need to use this config since automatic cache size limits should work well in most cases. This config setting will be removed once the new model cache behavior is stable.")
     lazy_offload:                  bool = Field(default=True,               description="DEPRECATED: This setting is no longer used. Lazy-offloading is enabled by default. This config setting will be removed once the new model cache behavior is stable.")
 
+    # PyTorch Memory Allocator
+    pytorch_cuda_alloc_conf: Optional[str] = Field(default=None,            description="Configure the Torch CUDA memory allocator. This will impact peak reserved VRAM usage and performance. Setting to \"backend:cudaMallocAsync\" works well on many systems. The optimal configuration is highly dependent on the system configuration (device type, VRAM, CUDA driver version, etc.), so must be tuned experimentally.")
+
     # DEVICE
     device:                      DEVICE = Field(default="auto",             description="Preferred execution device. `auto` will choose the device depending on the hardware platform and the installed torch capabilities.")
     precision:                PRECISION = Field(default="auto",             description="Floating point precision. `float16` will consume half the memory of `float32` but produce slightly lower-quality images. The `auto` setting will guess the proper precision based on your video card and operating system.")

invokeai/app/services/events/events_base.py

@@ -28,6 +28,7 @@ from invokeai.app.services.events.events_common import (
     ModelLoadCompleteEvent,
     ModelLoadStartedEvent,
     QueueClearedEvent,
+    QueueItemsRetriedEvent,
     QueueItemStatusChangedEvent,
 )
 
@@ -39,6 +40,7 @@ if TYPE_CHECKING:
     from invokeai.app.services.session_queue.session_queue_common import (
         BatchStatus,
         EnqueueBatchResult,
+        RetryItemsResult,
         SessionQueueItem,
         SessionQueueStatus,
     )
@@ -99,6 +101,10 @@ class EventServiceBase:
         """Emitted when a batch is enqueued"""
         self.dispatch(BatchEnqueuedEvent.build(enqueue_result))
 
+    def emit_queue_items_retried(self, retry_result: "RetryItemsResult") -> None:
+        """Emitted when a list of queue items are retried"""
+        self.dispatch(QueueItemsRetriedEvent.build(retry_result))
+
     def emit_queue_cleared(self, queue_id: str) -> None:
         """Emitted when a queue is cleared"""
         self.dispatch(QueueClearedEvent.build(queue_id))

invokeai/app/services/events/events_common.py

@@ -10,6 +10,7 @@ from invokeai.app.services.session_queue.session_queue_common import (
     QUEUE_ITEM_STATUS,
     BatchStatus,
     EnqueueBatchResult,
+    RetryItemsResult,
     SessionQueueItem,
     SessionQueueStatus,
 )
@@ -290,6 +291,22 @@ class BatchEnqueuedEvent(QueueEventBase):
         )
 
 
+@payload_schema.register
+class QueueItemsRetriedEvent(QueueEventBase):
+    """Event model for queue_items_retried"""
+
+    __event_name__ = "queue_items_retried"
+
+    retried_item_ids: list[int] = Field(description="The IDs of the queue items that were retried")
+
+    @classmethod
+    def build(cls, retry_result: RetryItemsResult) -> "QueueItemsRetriedEvent":
+        return cls(
+            queue_id=retry_result.queue_id,
+            retried_item_ids=retry_result.retried_item_ids,
+        )
+
+
 @payload_schema.register
 class QueueClearedEvent(QueueEventBase):
     """Event model for queue_cleared"""

invokeai/app/services/image_records/image_records_sqlite.py

@@ -1,5 +1,4 @@
 import sqlite3
-import threading
 from datetime import datetime
 from typing import Optional, Union, cast
 
@@ -22,21 +21,14 @@ from invokeai.app.services.shared.sqlite.sqlite_database import SqliteDatabase
 
 
 class SqliteImageRecordStorage(ImageRecordStorageBase):
-    _conn: sqlite3.Connection
-    _cursor: sqlite3.Cursor
-    _lock: threading.RLock
-
     def __init__(self, db: SqliteDatabase) -> None:
         super().__init__()
-        self._lock = db.lock
         self._conn = db.conn
-        self._cursor = self._conn.cursor()
 
     def get(self, image_name: str) -> ImageRecord:
         try:
-            self._lock.acquire()
-
-            self._cursor.execute(
+            cursor = self._conn.cursor()
+            cursor.execute(
                 f"""--sql
                 SELECT {IMAGE_DTO_COLS} FROM images
                 WHERE image_name = ?;
@@ -44,12 +36,9 @@ class SqliteImageRecordStorage(ImageRecordStorageBase):
                 (image_name,),
             )
 
-            result = cast(Optional[sqlite3.Row], self._cursor.fetchone())
+            result = cast(Optional[sqlite3.Row], cursor.fetchone())
         except sqlite3.Error as e:
-            self._conn.rollback()
             raise ImageRecordNotFoundException from e
-        finally:
-            self._lock.release()
 
         if not result:
             raise ImageRecordNotFoundException
@@ -58,9 +47,8 @@ class SqliteImageRecordStorage(ImageRecordStorageBase):
 
     def get_metadata(self, image_name: str) -> Optional[MetadataField]:
         try:
-            self._lock.acquire()
-
-            self._cursor.execute(
+            cursor = self._conn.cursor()
+            cursor.execute(
                 """--sql
                 SELECT metadata FROM images
                 WHERE image_name = ?;
@@ -68,7 +56,7 @@ class SqliteImageRecordStorage(ImageRecordStorageBase):
                 (image_name,),
             )
 
-            result = cast(Optional[sqlite3.Row], self._cursor.fetchone())
+            result = cast(Optional[sqlite3.Row], cursor.fetchone())
 
             if not result:
                 raise ImageRecordNotFoundException
@@ -77,10 +65,7 @@ class SqliteImageRecordStorage(ImageRecordStorageBase):
             metadata_raw = cast(Optional[str], as_dict.get("metadata", None))
             return MetadataFieldValidator.validate_json(metadata_raw) if metadata_raw is not None else None
         except sqlite3.Error as e:
-            self._conn.rollback()
             raise ImageRecordNotFoundException from e
-        finally:
-            self._lock.release()
 
     def update(
         self,
@@ -88,10 +73,10 @@ class SqliteImageRecordStorage(ImageRecordStorageBase):
         changes: ImageRecordChanges,
     ) -> None:
         try:
-            self._lock.acquire()
+            cursor = self._conn.cursor()
             # Change the category of the image
             if changes.image_category is not None:
-                self._cursor.execute(
+                cursor.execute(
                     """--sql
                     UPDATE images
                     SET image_category = ?
@@ -102,7 +87,7 @@ class SqliteImageRecordStorage(ImageRecordStorageBase):
 
             # Change the session associated with the image
             if changes.session_id is not None:
-                self._cursor.execute(
+                cursor.execute(
                     """--sql
                     UPDATE images
                     SET session_id = ?
@@ -113,7 +98,7 @@ class SqliteImageRecordStorage(ImageRecordStorageBase):
 
             # Change the image's `is_intermediate`` flag
             if changes.is_intermediate is not None:
-                self._cursor.execute(
+                cursor.execute(
                     """--sql
                     UPDATE images
                     SET is_intermediate = ?
@@ -124,7 +109,7 @@ class SqliteImageRecordStorage(ImageRecordStorageBase):
 
             # Change the image's `starred`` state
             if changes.starred is not None:
-                self._cursor.execute(
+                cursor.execute(
                     """--sql
                     UPDATE images
                     SET starred = ?
@@ -137,8 +122,6 @@ class SqliteImageRecordStorage(ImageRecordStorageBase):
         except sqlite3.Error as e:
             self._conn.rollback()
             raise ImageRecordSaveException from e
-        finally:
-            self._lock.release()
 
     def get_many(
         self,
@@ -152,110 +135,104 @@ class SqliteImageRecordStorage(ImageRecordStorageBase):
         board_id: Optional[str] = None,
         search_term: Optional[str] = None,
     ) -> OffsetPaginatedResults[ImageRecord]:
-        try:
-            self._lock.acquire()
+        cursor = self._conn.cursor()
 
-            # Manually build two queries - one for the count, one for the records
-            count_query = """--sql
-            SELECT COUNT(*)
-            FROM images
-            LEFT JOIN board_images ON board_images.image_name = images.image_name
-            WHERE 1=1
+        # Manually build two queries - one for the count, one for the records
+        count_query = """--sql
+        SELECT COUNT(*)
+        FROM images
+        LEFT JOIN board_images ON board_images.image_name = images.image_name
+        WHERE 1=1
+        """
+
+        images_query = f"""--sql
+        SELECT {IMAGE_DTO_COLS}
+        FROM images
+        LEFT JOIN board_images ON board_images.image_name = images.image_name
+        WHERE 1=1
+        """
+
+        query_conditions = ""
+        query_params: list[Union[int, str, bool]] = []
+
+        if image_origin is not None:
+            query_conditions += """--sql
+            AND images.image_origin = ?
+            """
+            query_params.append(image_origin.value)
+
+        if categories is not None:
+            # Convert the enum values to unique list of strings
+            category_strings = [c.value for c in set(categories)]
+            # Create the correct length of placeholders
+            placeholders = ",".join("?" * len(category_strings))
+
+            query_conditions += f"""--sql
+            AND images.image_category IN ( {placeholders} )
             """
 
-            images_query = f"""--sql
-            SELECT {IMAGE_DTO_COLS}
-            FROM images
-            LEFT JOIN board_images ON board_images.image_name = images.image_name
-            WHERE 1=1
+            # Unpack the included categories into the query params
+            for c in category_strings:
+                query_params.append(c)
+
+        if is_intermediate is not None:
+            query_conditions += """--sql
+            AND images.is_intermediate = ?
             """
 
-            query_conditions = ""
-            query_params: list[Union[int, str, bool]] = []
+            query_params.append(is_intermediate)
 
-            if image_origin is not None:
-                query_conditions += """--sql
-                AND images.image_origin = ?
-                """
-                query_params.append(image_origin.value)
+        # board_id of "none" is reserved for images without a board
+        if board_id == "none":
+            query_conditions += """--sql
+            AND board_images.board_id IS NULL
+            """
+        elif board_id is not None:
+            query_conditions += """--sql
+            AND board_images.board_id = ?
+            """
+            query_params.append(board_id)
 
-            if categories is not None:
-                # Convert the enum values to unique list of strings
-                category_strings = [c.value for c in set(categories)]
-                # Create the correct length of placeholders
-                placeholders = ",".join("?" * len(category_strings))
+        # Search term condition
+        if search_term:
+            query_conditions += """--sql
+            AND images.metadata LIKE ?
+            """
+            query_params.append(f"%{search_term.lower()}%")
 
-                query_conditions += f"""--sql
-                AND images.image_category IN ( {placeholders} )
-                """
+        if starred_first:
+            query_pagination = f"""--sql
+            ORDER BY images.starred DESC, images.created_at {order_dir.value} LIMIT ? OFFSET ?
+            """
+        else:
+            query_pagination = f"""--sql
+            ORDER BY images.created_at {order_dir.value} LIMIT ? OFFSET ?
+            """
 
-                # Unpack the included categories into the query params
-                for c in category_strings:
-                    query_params.append(c)
+        # Final images query with pagination
+        images_query += query_conditions + query_pagination + ";"
+        # Add all the parameters
+        images_params = query_params.copy()
+        # Add the pagination parameters
+        images_params.extend([limit, offset])
 
-            if is_intermediate is not None:
-                query_conditions += """--sql
-                AND images.is_intermediate = ?
-                """
+        # Build the list of images, deserializing each row
+        cursor.execute(images_query, images_params)
+        result = cast(list[sqlite3.Row], cursor.fetchall())
+        images = [deserialize_image_record(dict(r)) for r in result]
 
-                query_params.append(is_intermediate)
-
-            # board_id of "none" is reserved for images without a board
-            if board_id == "none":
-                query_conditions += """--sql
-                AND board_images.board_id IS NULL
-                """
-            elif board_id is not None:
-                query_conditions += """--sql
-                AND board_images.board_id = ?
-                """
-                query_params.append(board_id)
-
-            # Search term condition
-            if search_term:
-                query_conditions += """--sql
-                AND images.metadata LIKE ?
-                """
-                query_params.append(f"%{search_term.lower()}%")
-
-            if starred_first:
-                query_pagination = f"""--sql
-                ORDER BY images.starred DESC, images.created_at {order_dir.value} LIMIT ? OFFSET ?
-                """
-            else:
-                query_pagination = f"""--sql
-                ORDER BY images.created_at {order_dir.value} LIMIT ? OFFSET ?
-                """
-
-            # Final images query with pagination
-            images_query += query_conditions + query_pagination + ";"
-            # Add all the parameters
-            images_params = query_params.copy()
-            # Add the pagination parameters
-            images_params.extend([limit, offset])
-
-            # Build the list of images, deserializing each row
-            self._cursor.execute(images_query, images_params)
-            result = cast(list[sqlite3.Row], self._cursor.fetchall())
-            images = [deserialize_image_record(dict(r)) for r in result]
-
-            # Set up and execute the count query, without pagination
-            count_query += query_conditions + ";"
-            count_params = query_params.copy()
-            self._cursor.execute(count_query, count_params)
-            count = cast(int, self._cursor.fetchone()[0])
-        except sqlite3.Error as e:
-            self._conn.rollback()
-            raise e
-        finally:
-            self._lock.release()
+        # Set up and execute the count query, without pagination
+        count_query += query_conditions + ";"
+        count_params = query_params.copy()
+        cursor.execute(count_query, count_params)
+        count = cast(int, cursor.fetchone()[0])
 
         return OffsetPaginatedResults(items=images, offset=offset, limit=limit, total=count)
 
     def delete(self, image_name: str) -> None:
         try:
-            self._lock.acquire()
-            self._cursor.execute(
+            cursor = self._conn.cursor()
+            cursor.execute(
                 """--sql
                 DELETE FROM images
                 WHERE image_name = ?;
@@ -266,58 +243,48 @@ class SqliteImageRecordStorage(ImageRecordStorageBase):
         except sqlite3.Error as e:
             self._conn.rollback()
             raise ImageRecordDeleteException from e
-        finally:
-            self._lock.release()
 
     def delete_many(self, image_names: list[str]) -> None:
         try:
-            placeholders = ",".join("?" for _ in image_names)
+            cursor = self._conn.cursor()
 
-            self._lock.acquire()
+            placeholders = ",".join("?" for _ in image_names)
 
             # Construct the SQLite query with the placeholders
             query = f"DELETE FROM images WHERE image_name IN ({placeholders})"
 
             # Execute the query with the list of IDs as parameters
-            self._cursor.execute(query, image_names)
+            cursor.execute(query, image_names)
 
             self._conn.commit()
         except sqlite3.Error as e:
             self._conn.rollback()
             raise ImageRecordDeleteException from e
-        finally:
-            self._lock.release()
 
     def get_intermediates_count(self) -> int:
-        try:
-            self._lock.acquire()
-            self._cursor.execute(
-                """--sql
-                SELECT COUNT(*) FROM images
-                WHERE is_intermediate = TRUE;
-                """
-            )
-            count = cast(int, self._cursor.fetchone()[0])
-            self._conn.commit()
-            return count
-        except sqlite3.Error as e:
-            self._conn.rollback()
-            raise ImageRecordDeleteException from e
-        finally:
-            self._lock.release()
+        cursor = self._conn.cursor()
+        cursor.execute(
+            """--sql
+            SELECT COUNT(*) FROM images
+            WHERE is_intermediate = TRUE;
+            """
+        )
+        count = cast(int, cursor.fetchone()[0])
+        self._conn.commit()
+        return count
 
     def delete_intermediates(self) -> list[str]:
         try:
-            self._lock.acquire()
-            self._cursor.execute(
+            cursor = self._conn.cursor()
+            cursor.execute(
                 """--sql
                 SELECT image_name FROM images
                 WHERE is_intermediate = TRUE;
                 """
             )
-            result = cast(list[sqlite3.Row], self._cursor.fetchall())
+            result = cast(list[sqlite3.Row], cursor.fetchall())
             image_names = [r[0] for r in result]
-            self._cursor.execute(
+            cursor.execute(
                 """--sql
                 DELETE FROM images
                 WHERE is_intermediate = TRUE;
@@ -328,8 +295,6 @@ class SqliteImageRecordStorage(ImageRecordStorageBase):
         except sqlite3.Error as e:
             self._conn.rollback()
             raise ImageRecordDeleteException from e
-        finally:
-            self._lock.release()
 
     def save(
         self,
@@ -346,8 +311,8 @@ class SqliteImageRecordStorage(ImageRecordStorageBase):
         metadata: Optional[str] = None,
     ) -> datetime:
         try:
-            self._lock.acquire()
-            self._cursor.execute(
+            cursor = self._conn.cursor()
+            cursor.execute(
                 """--sql
                 INSERT OR IGNORE INTO images (
                     image_name,
@@ -380,7 +345,7 @@ class SqliteImageRecordStorage(ImageRecordStorageBase):
             )
             self._conn.commit()
 
-            self._cursor.execute(
+            cursor.execute(
                 """--sql
                 SELECT created_at
                 FROM images
@@ -389,34 +354,30 @@ class SqliteImageRecordStorage(ImageRecordStorageBase):
                 (image_name,),
             )
 
-            created_at = datetime.fromisoformat(self._cursor.fetchone()[0])
+            created_at = datetime.fromisoformat(cursor.fetchone()[0])
 
             return created_at
         except sqlite3.Error as e:
             self._conn.rollback()
             raise ImageRecordSaveException from e
-        finally:
-            self._lock.release()
 
     def get_most_recent_image_for_board(self, board_id: str) -> Optional[ImageRecord]:
-        try:
-            self._lock.acquire()
-            self._cursor.execute(
-                """--sql
-                SELECT images.*
-                FROM images
-                JOIN board_images ON images.image_name = board_images.image_name
-                WHERE board_images.board_id = ?
-                AND images.is_intermediate = FALSE
-                ORDER BY images.starred DESC, images.created_at DESC
-                LIMIT 1;
-                """,
-                (board_id,),
-            )
+        cursor = self._conn.cursor()
+        cursor.execute(
+            """--sql
+            SELECT images.*
+            FROM images
+            JOIN board_images ON images.image_name = board_images.image_name
+            WHERE board_images.board_id = ?
+            AND images.is_intermediate = FALSE
+            ORDER BY images.starred DESC, images.created_at DESC
+            LIMIT 1;
+            """,
+            (board_id,),
+        )
+
+        result = cast(Optional[sqlite3.Row], cursor.fetchone())
 
-            result = cast(Optional[sqlite3.Row], self._cursor.fetchone())
-        finally:
-            self._lock.release()
         if result is None:
             return None
 

invokeai/app/services/images/images_default.py

@@ -265,7 +265,11 @@ class ImageService(ImageServiceABC):
 
     def delete_images_on_board(self, board_id: str):
         try:
-            image_names = self.__invoker.services.board_image_records.get_all_board_image_names_for_board(board_id)
+            image_names = self.__invoker.services.board_image_records.get_all_board_image_names_for_board(
+                board_id,
+                categories=None,
+                is_intermediate=None,
+            )
             for image_name in image_names:
                 self.__invoker.services.image_files.delete(image_name)
             self.__invoker.services.image_records.delete_many(image_names)
@@ -278,7 +282,7 @@ class ImageService(ImageServiceABC):
             self.__invoker.services.logger.error("Failed to delete image files")
             raise
         except Exception as e:
-            self.__invoker.services.logger.error("Problem deleting image records and files")
+            self.__invoker.services.logger.error(f"Problem deleting image records and files: {str(e)}")
             raise e
 
     def delete_intermediates(self) -> int:

invokeai/app/services/model_images/model_images_common.py

@@ -1,4 +1,4 @@
-# TODO: Should these excpetions subclass existing python exceptions?
+# TODO: Should these exceptions subclass existing python exceptions?
 class ModelImageFileNotFoundException(Exception):
     """Raised when an image file is not found in storage."""
 

invokeai/app/services/model_manager/model_manager_default.py

@@ -84,6 +84,7 @@ class ModelManagerService(ModelManagerServiceBase):
         ram_cache = ModelCache(
             execution_device_working_mem_gb=app_config.device_working_mem_gb,
             enable_partial_loading=app_config.enable_partial_loading,
+            keep_ram_copy_of_weights=app_config.keep_ram_copy_of_weights,
             max_ram_cache_size_gb=app_config.max_cache_ram_gb,
             max_vram_cache_size_gb=app_config.max_cache_vram_gb,
             execution_device=execution_device or TorchDevice.choose_torch_device(),

invokeai/app/services/model_records/model_records_sql.py

@@ -78,7 +78,6 @@ class ModelRecordServiceSQL(ModelRecordServiceBase):
         """
         super().__init__()
         self._db = db
-        self._cursor = db.conn.cursor()
         self._logger = logger
 
     @property
@@ -96,38 +95,38 @@ class ModelRecordServiceSQL(ModelRecordServiceBase):
 
         Can raise DuplicateModelException and InvalidModelConfigException exceptions.
         """
-        with self._db.lock:
-            try:
-                self._cursor.execute(
-                    """--sql
-                    INSERT INTO models (
-                       id,
-                       config
-                      )
-                    VALUES (?,?);
-                    """,
-                    (
-                        config.key,
-                        config.model_dump_json(),
-                    ),
-                )
-                self._db.conn.commit()
+        try:
+            cursor = self._db.conn.cursor()
+            cursor.execute(
+                """--sql
+                INSERT INTO models (
+                    id,
+                    config
+                    )
+                VALUES (?,?);
+                """,
+                (
+                    config.key,
+                    config.model_dump_json(),
+                ),
+            )
+            self._db.conn.commit()
 
-            except sqlite3.IntegrityError as e:
-                self._db.conn.rollback()
-                if "UNIQUE constraint failed" in str(e):
-                    if "models.path" in str(e):
-                        msg = f"A model with path '{config.path}' is already installed"
-                    elif "models.name" in str(e):
-                        msg = f"A model with name='{config.name}', type='{config.type}', base='{config.base}' is already installed"
-                    else:
-                        msg = f"A model with key '{config.key}' is already installed"
-                    raise DuplicateModelException(msg) from e
+        except sqlite3.IntegrityError as e:
+            self._db.conn.rollback()
+            if "UNIQUE constraint failed" in str(e):
+                if "models.path" in str(e):
+                    msg = f"A model with path '{config.path}' is already installed"
+                elif "models.name" in str(e):
+                    msg = f"A model with name='{config.name}', type='{config.type}', base='{config.base}' is already installed"
                 else:
-                    raise e
-            except sqlite3.Error as e:
-                self._db.conn.rollback()
+                    msg = f"A model with key '{config.key}' is already installed"
+                raise DuplicateModelException(msg) from e
+            else:
                 raise e
+        except sqlite3.Error as e:
+            self._db.conn.rollback()
+            raise e
 
         return self.get_model(config.key)
 
@@ -139,21 +138,21 @@ class ModelRecordServiceSQL(ModelRecordServiceBase):
 
         Can raise an UnknownModelException
         """
-        with self._db.lock:
-            try:
-                self._cursor.execute(
-                    """--sql
-                    DELETE FROM models
-                    WHERE id=?;
-                    """,
-                    (key,),
-                )
-                if self._cursor.rowcount == 0:
-                    raise UnknownModelException("model not found")
-                self._db.conn.commit()
-            except sqlite3.Error as e:
-                self._db.conn.rollback()
-                raise e
+        try:
+            cursor = self._db.conn.cursor()
+            cursor.execute(
+                """--sql
+                DELETE FROM models
+                WHERE id=?;
+                """,
+                (key,),
+            )
+            if cursor.rowcount == 0:
+                raise UnknownModelException("model not found")
+            self._db.conn.commit()
+        except sqlite3.Error as e:
+            self._db.conn.rollback()
+            raise e
 
     def update_model(self, key: str, changes: ModelRecordChanges) -> AnyModelConfig:
         record = self.get_model(key)
@@ -164,23 +163,23 @@ class ModelRecordServiceSQL(ModelRecordServiceBase):
 
         json_serialized = record.model_dump_json()
 
-        with self._db.lock:
-            try:
-                self._cursor.execute(
-                    """--sql
-                    UPDATE models
-                    SET
-                        config=?
-                    WHERE id=?;
-                    """,
-                    (json_serialized, key),
-                )
-                if self._cursor.rowcount == 0:
-                    raise UnknownModelException("model not found")
-                self._db.conn.commit()
-            except sqlite3.Error as e:
-                self._db.conn.rollback()
-                raise e
+        try:
+            cursor = self._db.conn.cursor()
+            cursor.execute(
+                """--sql
+                UPDATE models
+                SET
+                    config=?
+                WHERE id=?;
+                """,
+                (json_serialized, key),
+            )
+            if cursor.rowcount == 0:
+                raise UnknownModelException("model not found")
+            self._db.conn.commit()
+        except sqlite3.Error as e:
+            self._db.conn.rollback()
+            raise e
 
         return self.get_model(key)
 
@@ -192,33 +191,33 @@ class ModelRecordServiceSQL(ModelRecordServiceBase):
 
         Exceptions: UnknownModelException
         """
-        with self._db.lock:
-            self._cursor.execute(
-                """--sql
-                SELECT config, strftime('%s',updated_at) FROM models
-                WHERE id=?;
-                """,
-                (key,),
-            )
-            rows = self._cursor.fetchone()
-            if not rows:
-                raise UnknownModelException("model not found")
-            model = ModelConfigFactory.make_config(json.loads(rows[0]), timestamp=rows[1])
+        cursor = self._db.conn.cursor()
+        cursor.execute(
+            """--sql
+            SELECT config, strftime('%s',updated_at) FROM models
+            WHERE id=?;
+            """,
+            (key,),
+        )
+        rows = cursor.fetchone()
+        if not rows:
+            raise UnknownModelException("model not found")
+        model = ModelConfigFactory.make_config(json.loads(rows[0]), timestamp=rows[1])
         return model
 
     def get_model_by_hash(self, hash: str) -> AnyModelConfig:
-        with self._db.lock:
-            self._cursor.execute(
-                """--sql
-                SELECT config, strftime('%s',updated_at) FROM models
-                WHERE hash=?;
-                """,
-                (hash,),
-            )
-            rows = self._cursor.fetchone()
-            if not rows:
-                raise UnknownModelException("model not found")
-            model = ModelConfigFactory.make_config(json.loads(rows[0]), timestamp=rows[1])
+        cursor = self._db.conn.cursor()
+        cursor.execute(
+            """--sql
+            SELECT config, strftime('%s',updated_at) FROM models
+            WHERE hash=?;
+            """,
+            (hash,),
+        )
+        rows = cursor.fetchone()
+        if not rows:
+            raise UnknownModelException("model not found")
+        model = ModelConfigFactory.make_config(json.loads(rows[0]), timestamp=rows[1])
         return model
 
     def exists(self, key: str) -> bool:
@@ -227,16 +226,15 @@ class ModelRecordServiceSQL(ModelRecordServiceBase):
 
         :param key: Unique key for the model to be deleted
         """
-        count = 0
-        with self._db.lock:
-            self._cursor.execute(
-                """--sql
-                select count(*) FROM models
-                WHERE id=?;
-                """,
-                (key,),
-            )
-            count = self._cursor.fetchone()[0]
+        cursor = self._db.conn.cursor()
+        cursor.execute(
+            """--sql
+            select count(*) FROM models
+            WHERE id=?;
+            """,
+            (key,),
+        )
+        count = cursor.fetchone()[0]
         return count > 0
 
     def search_by_attr(
@@ -284,17 +282,18 @@ class ModelRecordServiceSQL(ModelRecordServiceBase):
             where_clause.append("format=?")
             bindings.append(model_format)
         where = f"WHERE {' AND '.join(where_clause)}" if where_clause else ""
-        with self._db.lock:
-            self._cursor.execute(
-                f"""--sql
-                SELECT config, strftime('%s',updated_at)
-                FROM models
-                {where}
-                ORDER BY {ordering[order_by]} -- using ? to bind doesn't work here for some reason;
-                """,
-                tuple(bindings),
-            )
-            result = self._cursor.fetchall()
+
+        cursor = self._db.conn.cursor()
+        cursor.execute(
+            f"""--sql
+            SELECT config, strftime('%s',updated_at)
+            FROM models
+            {where}
+            ORDER BY {ordering[order_by]} -- using ? to bind doesn't work here for some reason;
+            """,
+            tuple(bindings),
+        )
+        result = cursor.fetchall()
 
         # Parse the model configs.
         results: list[AnyModelConfig] = []
@@ -313,34 +312,28 @@ class ModelRecordServiceSQL(ModelRecordServiceBase):
 
     def search_by_path(self, path: Union[str, Path]) -> List[AnyModelConfig]:
         """Return models with the indicated path."""
-        results = []
-        with self._db.lock:
-            self._cursor.execute(
-                """--sql
-                SELECT config, strftime('%s',updated_at) FROM models
-                WHERE path=?;
-                """,
-                (str(path),),
-            )
-            results = [
-                ModelConfigFactory.make_config(json.loads(x[0]), timestamp=x[1]) for x in self._cursor.fetchall()
-            ]
+        cursor = self._db.conn.cursor()
+        cursor.execute(
+            """--sql
+            SELECT config, strftime('%s',updated_at) FROM models
+            WHERE path=?;
+            """,
+            (str(path),),
+        )
+        results = [ModelConfigFactory.make_config(json.loads(x[0]), timestamp=x[1]) for x in cursor.fetchall()]
         return results
 
     def search_by_hash(self, hash: str) -> List[AnyModelConfig]:
         """Return models with the indicated hash."""
-        results = []
-        with self._db.lock:
-            self._cursor.execute(
-                """--sql
-                SELECT config, strftime('%s',updated_at) FROM models
-                WHERE hash=?;
-                """,
-                (hash,),
-            )
-            results = [
-                ModelConfigFactory.make_config(json.loads(x[0]), timestamp=x[1]) for x in self._cursor.fetchall()
-            ]
+        cursor = self._db.conn.cursor()
+        cursor.execute(
+            """--sql
+            SELECT config, strftime('%s',updated_at) FROM models
+            WHERE hash=?;
+            """,
+            (hash,),
+        )
+        results = [ModelConfigFactory.make_config(json.loads(x[0]), timestamp=x[1]) for x in cursor.fetchall()]
         return results
 
     def list_models(
@@ -356,33 +349,32 @@ class ModelRecordServiceSQL(ModelRecordServiceBase):
             ModelRecordOrderBy.Format: "format",
         }
 
-        # Lock so that the database isn't updated while we're doing the two queries.
-        with self._db.lock:
-            # query1: get the total number of model configs
-            self._cursor.execute(
-                """--sql
-                select count(*) from models;
-                """,
-                (),
-            )
-            total = int(self._cursor.fetchone()[0])
+        cursor = self._db.conn.cursor()
 
-            # query2: fetch key fields
-            self._cursor.execute(
-                f"""--sql
-                SELECT config
-                FROM models
-                ORDER BY {ordering[order_by]} -- using ? to bind doesn't work here for some reason
-                LIMIT ?
-                OFFSET ?;
-                """,
-                (
-                    per_page,
-                    page * per_page,
-                ),
-            )
-            rows = self._cursor.fetchall()
-            items = [ModelSummary.model_validate(dict(x)) for x in rows]
-            return PaginatedResults(
-                page=page, pages=ceil(total / per_page), per_page=per_page, total=total, items=items
-            )
+        # Lock so that the database isn't updated while we're doing the two queries.
+        # query1: get the total number of model configs
+        cursor.execute(
+            """--sql
+            select count(*) from models;
+            """,
+            (),
+        )
+        total = int(cursor.fetchone()[0])
+
+        # query2: fetch key fields
+        cursor.execute(
+            f"""--sql
+            SELECT config
+            FROM models
+            ORDER BY {ordering[order_by]} -- using ? to bind doesn't work here for some reason
+            LIMIT ?
+            OFFSET ?;
+            """,
+            (
+                per_page,
+                page * per_page,
+            ),
+        )
+        rows = cursor.fetchall()
+        items = [ModelSummary.model_validate(dict(x)) for x in rows]
+        return PaginatedResults(page=page, pages=ceil(total / per_page), per_page=per_page, total=total, items=items)

invokeai/app/services/session_queue/session_queue_base.py

@@ -1,10 +1,11 @@
 from abc import ABC, abstractmethod
-from typing import Optional
+from typing import Any, Coroutine, Optional
 
 from invokeai.app.services.session_queue.session_queue_common import (
     QUEUE_ITEM_STATUS,
     Batch,
     BatchStatus,
+    CancelAllExceptCurrentResult,
     CancelByBatchIDsResult,
     CancelByDestinationResult,
     CancelByQueueIDResult,
@@ -13,6 +14,7 @@ from invokeai.app.services.session_queue.session_queue_common import (
     IsEmptyResult,
     IsFullResult,
     PruneResult,
+    RetryItemsResult,
     SessionQueueCountsByDestination,
     SessionQueueItem,
     SessionQueueItemDTO,
@@ -31,7 +33,7 @@ class SessionQueueBase(ABC):
         pass
 
     @abstractmethod
-    def enqueue_batch(self, queue_id: str, batch: Batch, prepend: bool) -> EnqueueBatchResult:
+    def enqueue_batch(self, queue_id: str, batch: Batch, prepend: bool) -> Coroutine[Any, Any, EnqueueBatchResult]:
         """Enqueues all permutations of a batch for execution."""
         pass
 
@@ -112,6 +114,11 @@ class SessionQueueBase(ABC):
         """Cancels all queue items with matching queue ID"""
         pass
 
+    @abstractmethod
+    def cancel_all_except_current(self, queue_id: str) -> CancelAllExceptCurrentResult:
+        """Cancels all queue items except in-progress items"""
+        pass
+
     @abstractmethod
     def list_queue_items(
         self,
@@ -133,3 +140,8 @@ class SessionQueueBase(ABC):
     def set_queue_item_session(self, item_id: int, session: GraphExecutionState) -> SessionQueueItem:
         """Sets the session for a session queue item. Use this to update the session state."""
         pass
+
+    @abstractmethod
+    def retry_items_by_id(self, queue_id: str, item_ids: list[int]) -> RetryItemsResult:
+        """Retries the given queue items"""
+        pass

invokeai/app/services/session_queue/session_queue_common.py

@@ -1,7 +1,7 @@
 import datetime
 import json
 from itertools import chain, product
-from typing import Generator, Iterable, Literal, NamedTuple, Optional, TypeAlias, Union, cast
+from typing import Generator, Literal, Optional, TypeAlias, Union, cast
 
 from pydantic import (
     AliasChoices,
@@ -108,8 +108,16 @@ class Batch(BaseModel):
             return v
         for batch_data_list in v:
             for datum in batch_data_list:
+                if not datum.items:
+                    continue
+
+                # Special handling for numbers - they can be mixed
+                # TODO(psyche): Update BatchDatum to have a `type` field to specify the type of the items, then we can have strict float and int fields
+                if all(isinstance(item, (int, float)) for item in datum.items):
+                    continue
+
                 # Get the type of the first item in the list
-                first_item_type = type(datum.items[0]) if datum.items else None
+                first_item_type = type(datum.items[0])
                 for item in datum.items:
                     if type(item) is not first_item_type:
                         raise BatchItemsTypeError("All items in a batch must have the same type")
@@ -226,6 +234,9 @@ class SessionQueueItemWithoutGraph(BaseModel):
     field_values: Optional[list[NodeFieldValue]] = Field(
         default=None, description="The field values that were used for this queue item"
     )
+    retried_from_item_id: Optional[int] = Field(
+        default=None, description="The item_id of the queue item that this item was retried from"
+    )
 
     @classmethod
     def queue_item_dto_from_dict(cls, queue_item_dict: dict) -> "SessionQueueItemDTO":
@@ -336,6 +347,11 @@ class EnqueueBatchResult(BaseModel):
     priority: int = Field(description="The priority of the enqueued batch")
 
 
+class RetryItemsResult(BaseModel):
+    queue_id: str = Field(description="The ID of the queue")
+    retried_item_ids: list[int] = Field(description="The IDs of the queue items that were retried")
+
+
 class ClearResult(BaseModel):
     """Result of clearing the session queue"""
 
@@ -366,6 +382,12 @@ class CancelByQueueIDResult(CancelByBatchIDsResult):
     pass
 
 
+class CancelAllExceptCurrentResult(CancelByBatchIDsResult):
+    """Result of canceling all except current"""
+
+    pass
+
+
 class IsEmptyResult(BaseModel):
     """Result of checking if the session queue is empty"""
 
@@ -384,61 +406,143 @@ class IsFullResult(BaseModel):
 # region Util
 
 
-def populate_graph(graph: Graph, node_field_values: Iterable[NodeFieldValue]) -> Graph:
+def create_session_nfv_tuples(batch: Batch, maximum: int) -> Generator[tuple[str, str, str], None, None]:
     """
-    Populates the given graph with the given batch data items.
-    """
-    graph_clone = graph.model_copy(deep=True)
-    for item in node_field_values:
-        node = graph_clone.get_node(item.node_path)
-        if node is None:
-            continue
-        setattr(node, item.field_name, item.value)
-        graph_clone.update_node(item.node_path, node)
-    return graph_clone
+    Given a batch and a maximum number of sessions to create, generate a tuple of session_id, session_json, and
+    field_values_json for each session.
 
+    The batch has a "source" graph and a data property. The data property is a list of lists of BatchDatum objects.
+    Each BatchDatum has a field identifier (e.g. a node id and field name), and a list of values to substitute into
+    the field.
 
-def create_session_nfv_tuples(
-    batch: Batch, maximum: int
-) -> Generator[tuple[GraphExecutionState, list[NodeFieldValue], Optional[WorkflowWithoutID]], None, None]:
-    """
-    Create all graph permutations from the given batch data and graph. Yields tuples
-    of the form (graph, batch_data_items) where batch_data_items is the list of BatchDataItems
-    that was applied to the graph.
+    This structure allows us to create a new graph for every possible permutation of BatchDatum objects:
+    - Each BatchDatum can be "expanded" into a dict of node-field-value tuples - one for each item in the BatchDatum.
+    - Zip each inner list of expanded BatchDatum objects together. Call this a "batch_data_list".
+    - Take the cartesian product of all zipped batch_data_lists, resulting in a list of permutations of BatchDatum
+    - Take the cartesian product of all zipped batch_data_lists, resulting in a list of lists of BatchDatum objects.
+        Each inner list now represents the substitution values for a single permutation (session).
+    - For each permutation, substitute the values into the graph
+
+    This function is optimized for performance, as it is used to generate a large number of sessions at once.
+
+    Args:
+        batch: The batch to generate sessions from
+        maximum: The maximum number of sessions to generate
+
+    Returns:
+        A generator that yields tuples of session_id, session_json, and field_values_json for each session. The
+        generator will stop early if the maximum number of sessions is reached.
     """
 
     # TODO: Should this be a class method on Batch?
 
-    data: list[list[tuple[NodeFieldValue]]] = []
+    data: list[list[tuple[dict]]] = []
     batch_data_collection = batch.data if batch.data is not None else []
-    for batch_datum_list in batch_data_collection:
-        # each batch_datum_list needs to be convered to NodeFieldValues and then zipped
 
-        node_field_values_to_zip: list[list[NodeFieldValue]] = []
+    for batch_datum_list in batch_data_collection:
+        node_field_values_to_zip: list[list[dict]] = []
+        # Expand each BatchDatum into a list of dicts - one for each item in the BatchDatum
         for batch_datum in batch_datum_list:
             node_field_values = [
-                NodeFieldValue(node_path=batch_datum.node_path, field_name=batch_datum.field_name, value=item)
+                # Note: A tuple here is slightly faster than a dict, but we need the object in dict form to be inserted
+                # in the session_queue table anyways. So, overall creating NFVs as dicts is faster.
+                {"node_path": batch_datum.node_path, "field_name": batch_datum.field_name, "value": item}
                 for item in batch_datum.items
             ]
             node_field_values_to_zip.append(node_field_values)
+        # Zip the dicts together to create a list of dicts for each permutation
         data.append(list(zip(*node_field_values_to_zip, strict=True)))  # type: ignore [arg-type]
 
-    # create generator to yield session,nfv tuples
+    # We serialize the graph and session once, then mutate the graph dict in place for each session.
+    #
+    # This sounds scary, but it's actually fine.
+    #
+    # The batch prep logic injects field values into the same fields for each generated session.
+    #
+    # For example, after the product operation, we'll end up with a list of node-field-value tuples like this:
+    # [
+    #   (
+    #     {"node_path": "1", "field_name": "a", "value": 1},
+    #     {"node_path": "2", "field_name": "b", "value": 2},
+    #     {"node_path": "3", "field_name": "c", "value": 3},
+    #   ),
+    #   (
+    #     {"node_path": "1", "field_name": "a", "value": 4},
+    #     {"node_path": "2", "field_name": "b", "value": 5},
+    #     {"node_path": "3", "field_name": "c", "value": 6},
+    #   )
+    # ]
+    #
+    # Note that each tuple has the same length, and each tuple substitutes values in for exactly the same node fields.
+    # No matter the complexity of the batch, this property holds true.
+    #
+    # This means each permutation's substitution can be done in-place on the same graph dict, because it overwrites the
+    # previous mutation. We only need to serialize the graph once, and then we can mutate it in place for each session.
+    #
+    # Previously, we had created new Graph objects for each session, but this was very slow for large (1k+ session
+    # batches). We then tried dumping the graph to dict and using deep-copy to create a new dict for each session,
+    # but this was also slow.
+    #
+    # Overall, we achieved a 100x speedup by mutating the graph dict in place for each session over creating new Graph
+    # objects for each session.
+    #
+    # We will also mutate the session dict in place, setting a new ID for each session and setting the mutated graph
+    # dict as the session's graph.
+
+    # Dump the batch's graph to a dict once
+    graph_as_dict = batch.graph.model_dump(warnings=False, exclude_none=True)
+
+    # We must provide a Graph object when creating the "dummy" session dict, but we don't actually use it. It will be
+    # overwritten for each session by the mutated graph_as_dict.
+    session_dict = GraphExecutionState(graph=Graph()).model_dump(warnings=False, exclude_none=True)
+
+    # Now we can create a generator that yields the session_id, session_json, and field_values_json for each session.
     count = 0
+
+    # Each batch may have multiple runs, so we need to generate the same number of sessions for each run. The total is
+    # still limited by the maximum number of sessions.
     for _ in range(batch.runs):
         for d in product(*data):
             if count >= maximum:
+                # We've reached the maximum number of sessions we may generate
                 return
+
+            # Flatten the list of lists of dicts into a single list of dicts
+            # TODO(psyche): Is the a more efficient way to do this?
             flat_node_field_values = list(chain.from_iterable(d))
-            graph = populate_graph(batch.graph, flat_node_field_values)
-            yield (GraphExecutionState(graph=graph), flat_node_field_values, batch.workflow)
+
+            # Need a fresh ID for each session
+            session_id = uuid_string()
+
+            # Mutate the session dict in place
+            session_dict["id"] = session_id
+
+            # Substitute the values into the graph
+            for nfv in flat_node_field_values:
+                graph_as_dict["nodes"][nfv["node_path"]][nfv["field_name"]] = nfv["value"]
+
+            # Mutate the session dict in place
+            session_dict["graph"] = graph_as_dict
+
+            # Serialize the session and field values
+            # Note the use of pydantic's to_jsonable_python to handle serialization of any python object, including sets.
+            session_json = json.dumps(session_dict, default=to_jsonable_python)
+            field_values_json = json.dumps(flat_node_field_values, default=to_jsonable_python)
+
+            # Yield the session_id, session_json, and field_values_json
+            yield (session_id, session_json, field_values_json)
+
+            # Increment the count so we know when to stop
             count += 1
 
 
 def calc_session_count(batch: Batch) -> int:
     """
-    Calculates the number of sessions that would be created by the batch, without incurring
-    the overhead of actually generating them. Adapted from `create_sessions().
+    Calculates the number of sessions that would be created by the batch, without incurring the overhead of actually
+    creating them, as is done in `create_session_nfv_tuples()`.
+
+    The count is used to communicate to the user how many sessions were _requested_ to be created, as opposed to how
+    many were _actually_ created (which may be less due to the maximum number of sessions).
     """
     # TODO: Should this be a class method on Batch?
     if not batch.data:
@@ -454,41 +558,75 @@ def calc_session_count(batch: Batch) -> int:
     return len(data_product) * batch.runs
 
 
-class SessionQueueValueToInsert(NamedTuple):
-    """A tuple of values to insert into the session_queue table"""
-
-    # Careful with the ordering of this - it must match the insert statement
-    queue_id: str  # queue_id
-    session: str  # session json
-    session_id: str  # session_id
-    batch_id: str  # batch_id
-    field_values: Optional[str]  # field_values json
-    priority: int  # priority
-    workflow: Optional[str]  # workflow json
-    origin: str | None
-    destination: str | None
+ValueToInsertTuple: TypeAlias = tuple[
+    str,  # queue_id
+    str,  # session (as stringified JSON)
+    str,  # session_id
+    str,  # batch_id
+    str | None,  # field_values (optional, as stringified JSON)
+    int,  # priority
+    str | None,  # workflow (optional, as stringified JSON)
+    str | None,  # origin (optional)
+    str | None,  # destination (optional)
+    int | None,  # retried_from_item_id (optional, this is always None for new items)
+]
+"""A type alias for the tuple of values to insert into the session queue table."""
 
 
-ValuesToInsert: TypeAlias = list[SessionQueueValueToInsert]
+def prepare_values_to_insert(
+    queue_id: str, batch: Batch, priority: int, max_new_queue_items: int
+) -> list[ValueToInsertTuple]:
+    """
+    Given a batch, prepare the values to insert into the session queue table. The list of tuples can be used with an
+    `executemany` statement to insert multiple rows at once.
 
+    Args:
+        queue_id: The ID of the queue to insert the items into
+        batch: The batch to prepare the values for
+        priority: The priority of the queue items
+        max_new_queue_items: The maximum number of queue items to insert
 
-def prepare_values_to_insert(queue_id: str, batch: Batch, priority: int, max_new_queue_items: int) -> ValuesToInsert:
-    values_to_insert: ValuesToInsert = []
-    for session, field_values, workflow in create_session_nfv_tuples(batch, max_new_queue_items):
-        # sessions must have unique id
-        session.id = uuid_string()
+    Returns:
+        A list of tuples to insert into the session queue table. Each tuple contains the following values:
+        - queue_id
+        - session (as stringified JSON)
+        - session_id
+        - batch_id
+        - field_values (optional, as stringified JSON)
+        - priority
+        - workflow (optional, as stringified JSON)
+        - origin (optional)
+        - destination (optional)
+        - retried_from_item_id (optional, this is always None for new items)
+    """
+
+    # A tuple is a fast and memory-efficient way to store the values to insert. Previously, we used a NamedTuple, but
+    # measured a ~5% performance improvement by using a normal tuple instead. For very large batches (10k+ items), the
+    # this difference becomes noticeable.
+    #
+    # So, despite the inferior DX with normal tuples, we use one here for performance reasons.
+
+    values_to_insert: list[ValueToInsertTuple] = []
+
+    # pydantic's to_jsonable_python handles serialization of any python object, including sets, which json.dumps does
+    # not support by default. Apparently there are sets somewhere in the graph.
+
+    # The same workflow is used for all sessions in the batch - serialize it once
+    workflow_json = json.dumps(batch.workflow, default=to_jsonable_python) if batch.workflow else None
+
+    for session_id, session_json, field_values_json in create_session_nfv_tuples(batch, max_new_queue_items):
         values_to_insert.append(
-            SessionQueueValueToInsert(
-                queue_id,  # queue_id
-                session.model_dump_json(warnings=False, exclude_none=True),  # session (json)
-                session.id,  # session_id
-                batch.batch_id,  # batch_id
-                # must use pydantic_encoder bc field_values is a list of models
-                json.dumps(field_values, default=to_jsonable_python) if field_values else None,  # field_values (json)
-                priority,  # priority
-                json.dumps(workflow, default=to_jsonable_python) if workflow else None,  # workflow (json)
-                batch.origin,  # origin
-                batch.destination,  # destination
+            (
+                queue_id,
+                session_json,
+                session_id,
+                batch.batch_id,
+                field_values_json,
+                priority,
+                workflow_json,
+                batch.origin,
+                batch.destination,
+                None,
             )
         )
     return values_to_insert

invokeai/app/services/session_queue/session_queue_sqlite.py

@@ -1,7 +1,10 @@
+import asyncio
+import json
 import sqlite3
-import threading
 from typing import Optional, Union, cast
 
+from pydantic_core import to_jsonable_python
+
 from invokeai.app.services.invoker import Invoker
 from invokeai.app.services.session_queue.session_queue_base import SessionQueueBase
 from invokeai.app.services.session_queue.session_queue_common import (
@@ -9,6 +12,7 @@ from invokeai.app.services.session_queue.session_queue_common import (
     QUEUE_ITEM_STATUS,
     Batch,
     BatchStatus,
+    CancelAllExceptCurrentResult,
     CancelByBatchIDsResult,
     CancelByDestinationResult,
     CancelByQueueIDResult,
@@ -17,6 +21,7 @@ from invokeai.app.services.session_queue.session_queue_common import (
     IsEmptyResult,
     IsFullResult,
     PruneResult,
+    RetryItemsResult,
     SessionQueueCountsByDestination,
     SessionQueueItem,
     SessionQueueItemDTO,
@@ -32,9 +37,6 @@ from invokeai.app.services.shared.sqlite.sqlite_database import SqliteDatabase
 
 class SqliteSessionQueue(SessionQueueBase):
     __invoker: Invoker
-    __conn: sqlite3.Connection
-    __cursor: sqlite3.Cursor
-    __lock: threading.RLock
 
     def start(self, invoker: Invoker) -> None:
         self.__invoker = invoker
@@ -50,9 +52,7 @@ class SqliteSessionQueue(SessionQueueBase):
 
     def __init__(self, db: SqliteDatabase) -> None:
         super().__init__()
-        self.__lock = db.lock
-        self.__conn = db.conn
-        self.__cursor = self.__conn.cursor()
+        self._conn = db.conn
 
     def _set_in_progress_to_canceled(self) -> None:
         """
@@ -60,8 +60,8 @@ class SqliteSessionQueue(SessionQueueBase):
         This is necessary because the invoker may have been killed while processing a queue item.
         """
         try:
-            self.__lock.acquire()
-            self.__cursor.execute(
+            cursor = self._conn.cursor()
+            cursor.execute(
                 """--sql
                 UPDATE session_queue
                 SET status = 'canceled'
@@ -69,14 +69,13 @@ class SqliteSessionQueue(SessionQueueBase):
                 """
             )
         except Exception:
-            self.__conn.rollback()
+            self._conn.rollback()
             raise
-        finally:
-            self.__lock.release()
 
     def _get_current_queue_size(self, queue_id: str) -> int:
         """Gets the current number of pending queue items"""
-        self.__cursor.execute(
+        cursor = self._conn.cursor()
+        cursor.execute(
             """--sql
             SELECT count(*)
             FROM session_queue
@@ -86,11 +85,12 @@ class SqliteSessionQueue(SessionQueueBase):
             """,
             (queue_id,),
         )
-        return cast(int, self.__cursor.fetchone()[0])
+        return cast(int, cursor.fetchone()[0])
 
     def _get_highest_priority(self, queue_id: str) -> int:
         """Gets the highest priority value in the queue"""
-        self.__cursor.execute(
+        cursor = self._conn.cursor()
+        cursor.execute(
             """--sql
             SELECT MAX(priority)
             FROM session_queue
@@ -100,12 +100,14 @@ class SqliteSessionQueue(SessionQueueBase):
             """,
             (queue_id,),
         )
-        return cast(Union[int, None], self.__cursor.fetchone()[0]) or 0
+        return cast(Union[int, None], cursor.fetchone()[0]) or 0
 
-    def enqueue_batch(self, queue_id: str, batch: Batch, prepend: bool) -> EnqueueBatchResult:
+    async def enqueue_batch(self, queue_id: str, batch: Batch, prepend: bool) -> EnqueueBatchResult:
+        return await asyncio.to_thread(self._enqueue_batch, queue_id, batch, prepend)
+
+    def _enqueue_batch(self, queue_id: str, batch: Batch, prepend: bool) -> EnqueueBatchResult:
         try:
-            self.__lock.acquire()
-
+            cursor = self._conn.cursor()
             # TODO: how does this work in a multi-user scenario?
             current_queue_size = self._get_current_queue_size(queue_id)
             max_queue_size = self.__invoker.services.configuration.max_queue_size
@@ -127,19 +129,17 @@ class SqliteSessionQueue(SessionQueueBase):
             if requested_count > enqueued_count:
                 values_to_insert = values_to_insert[:max_new_queue_items]
 
-            self.__cursor.executemany(
+            cursor.executemany(
                 """--sql
-                INSERT INTO session_queue (queue_id, session, session_id, batch_id, field_values, priority, workflow, origin, destination)
-                VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?)
+                INSERT INTO session_queue (queue_id, session, session_id, batch_id, field_values, priority, workflow, origin, destination, retried_from_item_id)
+                VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?, ?)
                 """,
                 values_to_insert,
             )
-            self.__conn.commit()
+            self._conn.commit()
         except Exception:
-            self.__conn.rollback()
+            self._conn.rollback()
             raise
-        finally:
-            self.__lock.release()
         enqueue_result = EnqueueBatchResult(
             queue_id=queue_id,
             requested=requested_count,
@@ -151,25 +151,19 @@ class SqliteSessionQueue(SessionQueueBase):
         return enqueue_result
 
     def dequeue(self) -> Optional[SessionQueueItem]:
-        try:
-            self.__lock.acquire()
-            self.__cursor.execute(
-                """--sql
-                SELECT *
-                FROM session_queue
-                WHERE status = 'pending'
-                ORDER BY
-                  priority DESC,
-                  item_id ASC
-                LIMIT 1
-                """
-            )
-            result = cast(Union[sqlite3.Row, None], self.__cursor.fetchone())
-        except Exception:
-            self.__conn.rollback()
-            raise
-        finally:
-            self.__lock.release()
+        cursor = self._conn.cursor()
+        cursor.execute(
+            """--sql
+            SELECT *
+            FROM session_queue
+            WHERE status = 'pending'
+            ORDER BY
+                priority DESC,
+                item_id ASC
+            LIMIT 1
+            """
+        )
+        result = cast(Union[sqlite3.Row, None], cursor.fetchone())
         if result is None:
             return None
         queue_item = SessionQueueItem.queue_item_from_dict(dict(result))
@@ -177,52 +171,40 @@ class SqliteSessionQueue(SessionQueueBase):
         return queue_item
 
     def get_next(self, queue_id: str) -> Optional[SessionQueueItem]:
-        try:
-            self.__lock.acquire()
-            self.__cursor.execute(
-                """--sql
-                SELECT *
-                FROM session_queue
-                WHERE
-                  queue_id = ?
-                  AND status = 'pending'
-                ORDER BY
-                  priority DESC,
-                  created_at ASC
-                LIMIT 1
-                """,
-                (queue_id,),
-            )
-            result = cast(Union[sqlite3.Row, None], self.__cursor.fetchone())
-        except Exception:
-            self.__conn.rollback()
-            raise
-        finally:
-            self.__lock.release()
+        cursor = self._conn.cursor()
+        cursor.execute(
+            """--sql
+            SELECT *
+            FROM session_queue
+            WHERE
+                queue_id = ?
+                AND status = 'pending'
+            ORDER BY
+                priority DESC,
+                created_at ASC
+            LIMIT 1
+            """,
+            (queue_id,),
+        )
+        result = cast(Union[sqlite3.Row, None], cursor.fetchone())
         if result is None:
             return None
         return SessionQueueItem.queue_item_from_dict(dict(result))
 
     def get_current(self, queue_id: str) -> Optional[SessionQueueItem]:
-        try:
-            self.__lock.acquire()
-            self.__cursor.execute(
-                """--sql
-                SELECT *
-                FROM session_queue
-                WHERE
-                  queue_id = ?
-                  AND status = 'in_progress'
-                LIMIT 1
-                """,
-                (queue_id,),
-            )
-            result = cast(Union[sqlite3.Row, None], self.__cursor.fetchone())
-        except Exception:
-            self.__conn.rollback()
-            raise
-        finally:
-            self.__lock.release()
+        cursor = self._conn.cursor()
+        cursor.execute(
+            """--sql
+            SELECT *
+            FROM session_queue
+            WHERE
+                queue_id = ?
+                AND status = 'in_progress'
+            LIMIT 1
+            """,
+            (queue_id,),
+        )
+        result = cast(Union[sqlite3.Row, None], cursor.fetchone())
         if result is None:
             return None
         return SessionQueueItem.queue_item_from_dict(dict(result))
@@ -236,8 +218,8 @@ class SqliteSessionQueue(SessionQueueBase):
         error_traceback: Optional[str] = None,
     ) -> SessionQueueItem:
         try:
-            self.__lock.acquire()
-            self.__cursor.execute(
+            cursor = self._conn.cursor()
+            cursor.execute(
                 """--sql
                 UPDATE session_queue
                 SET status = ?, error_type = ?, error_message = ?, error_traceback = ?
@@ -245,12 +227,10 @@ class SqliteSessionQueue(SessionQueueBase):
                 """,
                 (status, error_type, error_message, error_traceback, item_id),
             )
-            self.__conn.commit()
+            self._conn.commit()
         except Exception:
-            self.__conn.rollback()
+            self._conn.rollback()
             raise
-        finally:
-            self.__lock.release()
         queue_item = self.get_queue_item(item_id)
         batch_status = self.get_batch_status(queue_id=queue_item.queue_id, batch_id=queue_item.batch_id)
         queue_status = self.get_queue_status(queue_id=queue_item.queue_id)
@@ -258,48 +238,36 @@ class SqliteSessionQueue(SessionQueueBase):
         return queue_item
 
     def is_empty(self, queue_id: str) -> IsEmptyResult:
-        try:
-            self.__lock.acquire()
-            self.__cursor.execute(
-                """--sql
-                SELECT count(*)
-                FROM session_queue
-                WHERE queue_id = ?
-                """,
-                (queue_id,),
-            )
-            is_empty = cast(int, self.__cursor.fetchone()[0]) == 0
-        except Exception:
-            self.__conn.rollback()
-            raise
-        finally:
-            self.__lock.release()
+        cursor = self._conn.cursor()
+        cursor.execute(
+            """--sql
+            SELECT count(*)
+            FROM session_queue
+            WHERE queue_id = ?
+            """,
+            (queue_id,),
+        )
+        is_empty = cast(int, cursor.fetchone()[0]) == 0
         return IsEmptyResult(is_empty=is_empty)
 
     def is_full(self, queue_id: str) -> IsFullResult:
-        try:
-            self.__lock.acquire()
-            self.__cursor.execute(
-                """--sql
-                SELECT count(*)
-                FROM session_queue
-                WHERE queue_id = ?
-                """,
-                (queue_id,),
-            )
-            max_queue_size = self.__invoker.services.configuration.max_queue_size
-            is_full = cast(int, self.__cursor.fetchone()[0]) >= max_queue_size
-        except Exception:
-            self.__conn.rollback()
-            raise
-        finally:
-            self.__lock.release()
+        cursor = self._conn.cursor()
+        cursor.execute(
+            """--sql
+            SELECT count(*)
+            FROM session_queue
+            WHERE queue_id = ?
+            """,
+            (queue_id,),
+        )
+        max_queue_size = self.__invoker.services.configuration.max_queue_size
+        is_full = cast(int, cursor.fetchone()[0]) >= max_queue_size
         return IsFullResult(is_full=is_full)
 
     def clear(self, queue_id: str) -> ClearResult:
         try:
-            self.__lock.acquire()
-            self.__cursor.execute(
+            cursor = self._conn.cursor()
+            cursor.execute(
                 """--sql
                 SELECT COUNT(*)
                 FROM session_queue
@@ -307,8 +275,8 @@ class SqliteSessionQueue(SessionQueueBase):
                 """,
                 (queue_id,),
             )
-            count = self.__cursor.fetchone()[0]
-            self.__cursor.execute(
+            count = cursor.fetchone()[0]
+            cursor.execute(
                 """--sql
                 DELETE
                 FROM session_queue
@@ -316,17 +284,16 @@ class SqliteSessionQueue(SessionQueueBase):
                 """,
                 (queue_id,),
             )
-            self.__conn.commit()
+            self._conn.commit()
         except Exception:
-            self.__conn.rollback()
+            self._conn.rollback()
             raise
-        finally:
-            self.__lock.release()
         self.__invoker.services.events.emit_queue_cleared(queue_id)
         return ClearResult(deleted=count)
 
     def prune(self, queue_id: str) -> PruneResult:
         try:
+            cursor = self._conn.cursor()
             where = """--sql
                 WHERE
                   queue_id = ?
@@ -336,8 +303,7 @@ class SqliteSessionQueue(SessionQueueBase):
                     OR status = 'canceled'
                   )
                 """
-            self.__lock.acquire()
-            self.__cursor.execute(
+            cursor.execute(
                 f"""--sql
                 SELECT COUNT(*)
                 FROM session_queue
@@ -345,8 +311,8 @@ class SqliteSessionQueue(SessionQueueBase):
                 """,
                 (queue_id,),
             )
-            count = self.__cursor.fetchone()[0]
-            self.__cursor.execute(
+            count = cursor.fetchone()[0]
+            cursor.execute(
                 f"""--sql
                 DELETE
                 FROM session_queue
@@ -354,12 +320,10 @@ class SqliteSessionQueue(SessionQueueBase):
                 """,
                 (queue_id,),
             )
-            self.__conn.commit()
+            self._conn.commit()
         except Exception:
-            self.__conn.rollback()
+            self._conn.rollback()
             raise
-        finally:
-            self.__lock.release()
         return PruneResult(deleted=count)
 
     def cancel_queue_item(self, item_id: int) -> SessionQueueItem:
@@ -388,8 +352,8 @@ class SqliteSessionQueue(SessionQueueBase):
 
     def cancel_by_batch_ids(self, queue_id: str, batch_ids: list[str]) -> CancelByBatchIDsResult:
         try:
+            cursor = self._conn.cursor()
             current_queue_item = self.get_current(queue_id)
-            self.__lock.acquire()
             placeholders = ", ".join(["?" for _ in batch_ids])
             where = f"""--sql
                 WHERE
@@ -400,7 +364,7 @@ class SqliteSessionQueue(SessionQueueBase):
                   AND status != 'failed'
                 """
             params = [queue_id] + batch_ids
-            self.__cursor.execute(
+            cursor.execute(
                 f"""--sql
                 SELECT COUNT(*)
                 FROM session_queue
@@ -408,8 +372,8 @@ class SqliteSessionQueue(SessionQueueBase):
                 """,
                 tuple(params),
             )
-            count = self.__cursor.fetchone()[0]
-            self.__cursor.execute(
+            count = cursor.fetchone()[0]
+            cursor.execute(
                 f"""--sql
                 UPDATE session_queue
                 SET status = 'canceled'
@@ -417,20 +381,18 @@ class SqliteSessionQueue(SessionQueueBase):
                 """,
                 tuple(params),
             )
-            self.__conn.commit()
+            self._conn.commit()
             if current_queue_item is not None and current_queue_item.batch_id in batch_ids:
                 self._set_queue_item_status(current_queue_item.item_id, "canceled")
         except Exception:
-            self.__conn.rollback()
+            self._conn.rollback()
             raise
-        finally:
-            self.__lock.release()
         return CancelByBatchIDsResult(canceled=count)
 
     def cancel_by_destination(self, queue_id: str, destination: str) -> CancelByDestinationResult:
         try:
+            cursor = self._conn.cursor()
             current_queue_item = self.get_current(queue_id)
-            self.__lock.acquire()
             where = """--sql
                 WHERE
                   queue_id == ?
@@ -440,7 +402,7 @@ class SqliteSessionQueue(SessionQueueBase):
                   AND status != 'failed'
                 """
             params = (queue_id, destination)
-            self.__cursor.execute(
+            cursor.execute(
                 f"""--sql
                 SELECT COUNT(*)
                 FROM session_queue
@@ -448,8 +410,8 @@ class SqliteSessionQueue(SessionQueueBase):
                 """,
                 params,
             )
-            count = self.__cursor.fetchone()[0]
-            self.__cursor.execute(
+            count = cursor.fetchone()[0]
+            cursor.execute(
                 f"""--sql
                 UPDATE session_queue
                 SET status = 'canceled'
@@ -457,20 +419,18 @@ class SqliteSessionQueue(SessionQueueBase):
                 """,
                 params,
             )
-            self.__conn.commit()
+            self._conn.commit()
             if current_queue_item is not None and current_queue_item.destination == destination:
                 self._set_queue_item_status(current_queue_item.item_id, "canceled")
         except Exception:
-            self.__conn.rollback()
+            self._conn.rollback()
             raise
-        finally:
-            self.__lock.release()
         return CancelByDestinationResult(canceled=count)
 
     def cancel_by_queue_id(self, queue_id: str) -> CancelByQueueIDResult:
         try:
+            cursor = self._conn.cursor()
             current_queue_item = self.get_current(queue_id)
-            self.__lock.acquire()
             where = """--sql
                 WHERE
                   queue_id is ?
@@ -479,7 +439,7 @@ class SqliteSessionQueue(SessionQueueBase):
                   AND status != 'failed'
                 """
             params = [queue_id]
-            self.__cursor.execute(
+            cursor.execute(
                 f"""--sql
                 SELECT COUNT(*)
                 FROM session_queue
@@ -487,8 +447,8 @@ class SqliteSessionQueue(SessionQueueBase):
                 """,
                 tuple(params),
             )
-            count = self.__cursor.fetchone()[0]
-            self.__cursor.execute(
+            count = cursor.fetchone()[0]
+            cursor.execute(
                 f"""--sql
                 UPDATE session_queue
                 SET status = 'canceled'
@@ -496,7 +456,7 @@ class SqliteSessionQueue(SessionQueueBase):
                 """,
                 tuple(params),
             )
-            self.__conn.commit()
+            self._conn.commit()
             if current_queue_item is not None and current_queue_item.queue_id == queue_id:
                 batch_status = self.get_batch_status(queue_id=queue_id, batch_id=current_queue_item.batch_id)
                 queue_status = self.get_queue_status(queue_id=queue_id)
@@ -504,41 +464,64 @@ class SqliteSessionQueue(SessionQueueBase):
                     current_queue_item, batch_status, queue_status
                 )
         except Exception:
-            self.__conn.rollback()
+            self._conn.rollback()
             raise
-        finally:
-            self.__lock.release()
         return CancelByQueueIDResult(canceled=count)
 
-    def get_queue_item(self, item_id: int) -> SessionQueueItem:
+    def cancel_all_except_current(self, queue_id: str) -> CancelAllExceptCurrentResult:
         try:
-            self.__lock.acquire()
-            self.__cursor.execute(
-                """--sql
-                SELECT * FROM session_queue
+            cursor = self._conn.cursor()
+            where = """--sql
                 WHERE
-                  item_id = ?
+                  queue_id == ?
+                  AND status == 'pending'
+                """
+            cursor.execute(
+                f"""--sql
+                SELECT COUNT(*)
+                FROM session_queue
+                {where};
                 """,
-                (item_id,),
+                (queue_id,),
             )
-            result = cast(Union[sqlite3.Row, None], self.__cursor.fetchone())
+            count = cursor.fetchone()[0]
+            cursor.execute(
+                f"""--sql
+                UPDATE session_queue
+                SET status = 'canceled'
+                {where};
+                """,
+                (queue_id,),
+            )
+            self._conn.commit()
         except Exception:
-            self.__conn.rollback()
+            self._conn.rollback()
             raise
-        finally:
-            self.__lock.release()
+        return CancelAllExceptCurrentResult(canceled=count)
+
+    def get_queue_item(self, item_id: int) -> SessionQueueItem:
+        cursor = self._conn.cursor()
+        cursor.execute(
+            """--sql
+            SELECT * FROM session_queue
+            WHERE
+                item_id = ?
+            """,
+            (item_id,),
+        )
+        result = cast(Union[sqlite3.Row, None], cursor.fetchone())
         if result is None:
             raise SessionQueueItemNotFoundError(f"No queue item with id {item_id}")
         return SessionQueueItem.queue_item_from_dict(dict(result))
 
     def set_queue_item_session(self, item_id: int, session: GraphExecutionState) -> SessionQueueItem:
         try:
+            cursor = self._conn.cursor()
             # Use exclude_none so we don't end up with a bunch of nulls in the graph - this can cause validation errors
             # when the graph is loaded. Graph execution occurs purely in memory - the session saved here is not referenced
             # during execution.
             session_json = session.model_dump_json(warnings=False, exclude_none=True)
-            self.__lock.acquire()
-            self.__cursor.execute(
+            cursor.execute(
                 """--sql
                 UPDATE session_queue
                 SET session = ?
@@ -546,12 +529,10 @@ class SqliteSessionQueue(SessionQueueBase):
                 """,
                 (session_json, item_id),
             )
-            self.__conn.commit()
+            self._conn.commit()
         except Exception:
-            self.__conn.rollback()
+            self._conn.rollback()
             raise
-        finally:
-            self.__lock.release()
         return self.get_queue_item(item_id)
 
     def list_queue_items(
@@ -562,83 +543,71 @@ class SqliteSessionQueue(SessionQueueBase):
         cursor: Optional[int] = None,
         status: Optional[QUEUE_ITEM_STATUS] = None,
     ) -> CursorPaginatedResults[SessionQueueItemDTO]:
-        try:
-            item_id = cursor
-            self.__lock.acquire()
-            query = """--sql
-                SELECT item_id,
-                    status,
-                    priority,
-                    field_values,
-                    error_type,
-                    error_message,
-                    error_traceback,
-                    created_at,
-                    updated_at,
-                    completed_at,
-                    started_at,
-                    session_id,
-                    batch_id,
-                    queue_id,
-                    origin,
-                    destination
-                FROM session_queue
-                WHERE queue_id = ?
-            """
-            params: list[Union[str, int]] = [queue_id]
-
-            if status is not None:
-                query += """--sql
-                    AND status = ?
-                    """
-                params.append(status)
-
-            if item_id is not None:
-                query += """--sql
-                    AND (priority < ?) OR (priority = ? AND item_id > ?)
-                    """
-                params.extend([priority, priority, item_id])
+        cursor_ = self._conn.cursor()
+        item_id = cursor
+        query = """--sql
+            SELECT item_id,
+                status,
+                priority,
+                field_values,
+                error_type,
+                error_message,
+                error_traceback,
+                created_at,
+                updated_at,
+                completed_at,
+                started_at,
+                session_id,
+                batch_id,
+                queue_id,
+                origin,
+                destination
+            FROM session_queue
+            WHERE queue_id = ?
+        """
+        params: list[Union[str, int]] = [queue_id]
 
+        if status is not None:
             query += """--sql
-                ORDER BY
-                  priority DESC,
-                  item_id ASC
-                LIMIT ?
+                AND status = ?
                 """
-            params.append(limit + 1)
-            self.__cursor.execute(query, params)
-            results = cast(list[sqlite3.Row], self.__cursor.fetchall())
-            items = [SessionQueueItemDTO.queue_item_dto_from_dict(dict(result)) for result in results]
-            has_more = False
-            if len(items) > limit:
-                # remove the extra item
-                items.pop()
-                has_more = True
-        except Exception:
-            self.__conn.rollback()
-            raise
-        finally:
-            self.__lock.release()
+            params.append(status)
+
+        if item_id is not None:
+            query += """--sql
+                AND (priority < ?) OR (priority = ? AND item_id > ?)
+                """
+            params.extend([priority, priority, item_id])
+
+        query += """--sql
+            ORDER BY
+                priority DESC,
+                item_id ASC
+            LIMIT ?
+            """
+        params.append(limit + 1)
+        cursor_.execute(query, params)
+        results = cast(list[sqlite3.Row], cursor_.fetchall())
+        items = [SessionQueueItemDTO.queue_item_dto_from_dict(dict(result)) for result in results]
+        has_more = False
+        if len(items) > limit:
+            # remove the extra item
+            items.pop()
+            has_more = True
         return CursorPaginatedResults(items=items, limit=limit, has_more=has_more)
 
     def get_queue_status(self, queue_id: str) -> SessionQueueStatus:
-        try:
-            self.__lock.acquire()
-            self.__cursor.execute(
-                """--sql
-                SELECT status, count(*)
-                FROM session_queue
-                WHERE queue_id = ?
-                GROUP BY status
-                """,
-                (queue_id,),
-            )
-            counts_result = cast(list[sqlite3.Row], self.__cursor.fetchall())
-        except Exception:
-            self.__conn.rollback()
-            raise
-        finally:
-            self.__lock.release()
+        cursor = self._conn.cursor()
+        cursor.execute(
+            """--sql
+            SELECT status, count(*)
+            FROM session_queue
+            WHERE queue_id = ?
+            GROUP BY status
+            """,
+            (queue_id,),
+        )
+        counts_result = cast(list[sqlite3.Row], cursor.fetchall())
 
         current_item = self.get_current(queue_id=queue_id)
         total = sum(row[1] for row in counts_result)
@@ -657,29 +626,23 @@ class SqliteSessionQueue(SessionQueueBase):
         )
 
     def get_batch_status(self, queue_id: str, batch_id: str) -> BatchStatus:
-        try:
-            self.__lock.acquire()
-            self.__cursor.execute(
-                """--sql
-                SELECT status, count(*), origin, destination
-                FROM session_queue
-                WHERE
-                  queue_id = ?
-                  AND batch_id = ?
-                GROUP BY status
-                """,
-                (queue_id, batch_id),
-            )
-            result = cast(list[sqlite3.Row], self.__cursor.fetchall())
-            total = sum(row[1] for row in result)
-            counts: dict[str, int] = {row[0]: row[1] for row in result}
-            origin = result[0]["origin"] if result else None
-            destination = result[0]["destination"] if result else None
-        except Exception:
-            self.__conn.rollback()
-            raise
-        finally:
-            self.__lock.release()
+        cursor = self._conn.cursor()
+        cursor.execute(
+            """--sql
+            SELECT status, count(*), origin, destination
+            FROM session_queue
+            WHERE
+                queue_id = ?
+                AND batch_id = ?
+            GROUP BY status
+            """,
+            (queue_id, batch_id),
+        )
+        result = cast(list[sqlite3.Row], cursor.fetchall())
+        total = sum(row[1] for row in result)
+        counts: dict[str, int] = {row[0]: row[1] for row in result}
+        origin = result[0]["origin"] if result else None
+        destination = result[0]["destination"] if result else None
 
         return BatchStatus(
             batch_id=batch_id,
@@ -695,24 +658,18 @@ class SqliteSessionQueue(SessionQueueBase):
         )
 
     def get_counts_by_destination(self, queue_id: str, destination: str) -> SessionQueueCountsByDestination:
-        try:
-            self.__lock.acquire()
-            self.__cursor.execute(
-                """--sql
-                SELECT status, count(*)
-                FROM session_queue
-                WHERE queue_id = ?
-                AND destination = ?
-                GROUP BY status
-                """,
-                (queue_id, destination),
-            )
-            counts_result = cast(list[sqlite3.Row], self.__cursor.fetchall())
-        except Exception:
-            self.__conn.rollback()
-            raise
-        finally:
-            self.__lock.release()
+        cursor = self._conn.cursor()
+        cursor.execute(
+            """--sql
+            SELECT status, count(*)
+            FROM session_queue
+            WHERE queue_id = ?
+            AND destination = ?
+            GROUP BY status
+            """,
+            (queue_id, destination),
+        )
+        counts_result = cast(list[sqlite3.Row], cursor.fetchall())
 
         total = sum(row[1] for row in counts_result)
         counts: dict[str, int] = {row[0]: row[1] for row in counts_result}
@@ -727,3 +684,68 @@ class SqliteSessionQueue(SessionQueueBase):
             canceled=counts.get("canceled", 0),
             total=total,
         )
+
+    def retry_items_by_id(self, queue_id: str, item_ids: list[int]) -> RetryItemsResult:
+        """Retries the given queue items"""
+        try:
+            cursor = self._conn.cursor()
+            values_to_insert: list[tuple] = []
+            retried_item_ids: list[int] = []
+
+            for item_id in item_ids:
+                queue_item = self.get_queue_item(item_id)
+
+                if queue_item.status not in ("failed", "canceled"):
+                    continue
+
+                retried_item_ids.append(item_id)
+
+                field_values_json = (
+                    json.dumps(queue_item.field_values, default=to_jsonable_python) if queue_item.field_values else None
+                )
+                workflow_json = (
+                    json.dumps(queue_item.workflow, default=to_jsonable_python) if queue_item.workflow else None
+                )
+                cloned_session = GraphExecutionState(graph=queue_item.session.graph)
+                cloned_session_json = cloned_session.model_dump_json(warnings=False, exclude_none=True)
+
+                retried_from_item_id = (
+                    queue_item.retried_from_item_id
+                    if queue_item.retried_from_item_id is not None
+                    else queue_item.item_id
+                )
+
+                value_to_insert = (
+                    queue_item.queue_id,
+                    queue_item.batch_id,
+                    queue_item.destination,
+                    field_values_json,
+                    queue_item.origin,
+                    queue_item.priority,
+                    workflow_json,
+                    cloned_session_json,
+                    cloned_session.id,
+                    retried_from_item_id,
+                )
+                values_to_insert.append(value_to_insert)
+
+            # TODO(psyche): Handle max queue size?
+
+            cursor.executemany(
+                """--sql
+                INSERT INTO session_queue (queue_id, session, session_id, batch_id, field_values, priority, workflow, origin, destination, retried_from_item_id)
+                VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?, ?)
+                """,
+                values_to_insert,
+            )
+
+            self._conn.commit()
+        except Exception:
+            self._conn.rollback()
+            raise
+        retry_result = RetryItemsResult(
+            queue_id=queue_id,
+            retried_item_ids=retried_item_ids,
+        )
+        self.__invoker.services.events.emit_queue_items_retried(retry_result)
+        return retry_result

invokeai/app/services/shared/graph.py

@@ -51,15 +51,18 @@ class Edge(BaseModel):
     source: EdgeConnection = Field(description="The connection for the edge's from node and field")
     destination: EdgeConnection = Field(description="The connection for the edge's to node and field")
 
+    def __str__(self):
+        return f"{self.source.node_id}.{self.source.field} -> {self.destination.node_id}.{self.destination.field}"
 
-def get_output_field(node: BaseInvocation, field: str) -> Any:
+
+def get_output_field_type(node: BaseInvocation, field: str) -> Any:
     node_type = type(node)
     node_outputs = get_type_hints(node_type.get_output_annotation())
     node_output_field = node_outputs.get(field) or None
     return node_output_field
 
 
-def get_input_field(node: BaseInvocation, field: str) -> Any:
+def get_input_field_type(node: BaseInvocation, field: str) -> Any:
     node_type = type(node)
     node_inputs = get_type_hints(node_type)
     node_input_field = node_inputs.get(field) or None
@@ -93,6 +96,10 @@ def is_list_or_contains_list(t):
     return False
 
 
+def is_any(t: Any) -> bool:
+    return t == Any or Any in get_args(t)
+
+
 def are_connection_types_compatible(from_type: Any, to_type: Any) -> bool:
     if not from_type:
         return False
@@ -102,13 +109,7 @@ def are_connection_types_compatible(from_type: Any, to_type: Any) -> bool:
     # TODO: this is pretty forgiving on generic types. Clean that up (need to handle optionals and such)
     if from_type and to_type:
         # Ports are compatible
-        if (
-            from_type == to_type
-            or from_type == Any
-            or to_type == Any
-            or Any in get_args(from_type)
-            or Any in get_args(to_type)
-        ):
+        if from_type == to_type or is_any(from_type) or is_any(to_type):
             return True
 
         if from_type in get_args(to_type):
@@ -140,10 +141,10 @@ def are_connections_compatible(
     """Determines if a connection between fields of two nodes is compatible."""
 
     # TODO: handle iterators and collectors
-    from_node_field = get_output_field(from_node, from_field)
-    to_node_field = get_input_field(to_node, to_field)
+    from_type = get_output_field_type(from_node, from_field)
+    to_type = get_input_field_type(to_node, to_field)
 
-    return are_connection_types_compatible(from_node_field, to_node_field)
+    return are_connection_types_compatible(from_type, to_type)
 
 
 T = TypeVar("T")
@@ -440,17 +441,19 @@ class Graph(BaseModel):
                 self.get_node(edge.destination.node_id),
                 edge.destination.field,
             ):
-                raise InvalidEdgeError(
-                    f"Invalid edge from {edge.source.node_id}.{edge.source.field} to {edge.destination.node_id}.{edge.destination.field}"
-                )
+                raise InvalidEdgeError(f"Edge source and target types do not match ({edge})")
 
         # Validate all iterators & collectors
         # TODO: may need to validate all iterators & collectors in subgraphs so edge connections in parent graphs will be available
         for node in self.nodes.values():
-            if isinstance(node, IterateInvocation) and not self._is_iterator_connection_valid(node.id):
-                raise InvalidEdgeError(f"Invalid iterator node {node.id}")
-            if isinstance(node, CollectInvocation) and not self._is_collector_connection_valid(node.id):
-                raise InvalidEdgeError(f"Invalid collector node {node.id}")
+            if isinstance(node, IterateInvocation):
+                err = self._is_iterator_connection_valid(node.id)
+                if err is not None:
+                    raise InvalidEdgeError(f"Invalid iterator node ({node.id}): {err}")
+            if isinstance(node, CollectInvocation):
+                err = self._is_collector_connection_valid(node.id)
+                if err is not None:
+                    raise InvalidEdgeError(f"Invalid collector node ({node.id}): {err}")
 
         return None
 
@@ -477,11 +480,11 @@ class Graph(BaseModel):
 
     def _is_destination_field_Any(self, edge: Edge) -> bool:
         """Checks if the destination field for an edge is of type typing.Any"""
-        return get_input_field(self.get_node(edge.destination.node_id), edge.destination.field) == Any
+        return get_input_field_type(self.get_node(edge.destination.node_id), edge.destination.field) == Any
 
     def _is_destination_field_list_of_Any(self, edge: Edge) -> bool:
         """Checks if the destination field for an edge is of type typing.Any"""
-        return get_input_field(self.get_node(edge.destination.node_id), edge.destination.field) == list[Any]
+        return get_input_field_type(self.get_node(edge.destination.node_id), edge.destination.field) == list[Any]
 
     def _validate_edge(self, edge: Edge):
         """Validates that a new edge doesn't create a cycle in the graph"""
@@ -491,55 +494,40 @@ class Graph(BaseModel):
             from_node = self.get_node(edge.source.node_id)
             to_node = self.get_node(edge.destination.node_id)
         except NodeNotFoundError:
-            raise InvalidEdgeError("One or both nodes don't exist: {edge.source.node_id} -> {edge.destination.node_id}")
+            raise InvalidEdgeError(f"One or both nodes don't exist ({edge})")
 
         # Validate that an edge to this node+field doesn't already exist
         input_edges = self._get_input_edges(edge.destination.node_id, edge.destination.field)
         if len(input_edges) > 0 and not isinstance(to_node, CollectInvocation):
-            raise InvalidEdgeError(
-                f"Edge to node {edge.destination.node_id} field {edge.destination.field} already exists"
-            )
+            raise InvalidEdgeError(f"Edge already exists ({edge})")
 
         # Validate that no cycles would be created
         g = self.nx_graph_flat()
         g.add_edge(edge.source.node_id, edge.destination.node_id)
         if not nx.is_directed_acyclic_graph(g):
-            raise InvalidEdgeError(
-                f"Edge creates a cycle in the graph: {edge.source.node_id} -> {edge.destination.node_id}"
-            )
+            raise InvalidEdgeError(f"Edge creates a cycle in the graph ({edge})")
 
         # Validate that the field types are compatible
         if not are_connections_compatible(from_node, edge.source.field, to_node, edge.destination.field):
-            raise InvalidEdgeError(
-                f"Fields are incompatible: cannot connect {edge.source.node_id}.{edge.source.field} to {edge.destination.node_id}.{edge.destination.field}"
-            )
+            raise InvalidEdgeError(f"Field types are incompatible ({edge})")
 
         # Validate if iterator output type matches iterator input type (if this edge results in both being set)
         if isinstance(to_node, IterateInvocation) and edge.destination.field == "collection":
-            if not self._is_iterator_connection_valid(edge.destination.node_id, new_input=edge.source):
-                raise InvalidEdgeError(
-                    f"Iterator input type does not match iterator output type: {edge.source.node_id}.{edge.source.field} to {edge.destination.node_id}.{edge.destination.field}"
-                )
+            err = self._is_iterator_connection_valid(edge.destination.node_id, new_input=edge.source)
+            if err is not None:
+                raise InvalidEdgeError(f"Iterator input type does not match iterator output type ({edge}): {err}")
 
         # Validate if iterator input type matches output type (if this edge results in both being set)
         if isinstance(from_node, IterateInvocation) and edge.source.field == "item":
-            if not self._is_iterator_connection_valid(edge.source.node_id, new_output=edge.destination):
-                raise InvalidEdgeError(
-                    f"Iterator output type does not match iterator input type:, {edge.source.node_id}.{edge.source.field} to {edge.destination.node_id}.{edge.destination.field}"
-                )
+            err = self._is_iterator_connection_valid(edge.source.node_id, new_output=edge.destination)
+            if err is not None:
+                raise InvalidEdgeError(f"Iterator output type does not match iterator input type ({edge}): {err}")
 
         # Validate if collector input type matches output type (if this edge results in both being set)
         if isinstance(to_node, CollectInvocation) and edge.destination.field == "item":
-            if not self._is_collector_connection_valid(edge.destination.node_id, new_input=edge.source):
-                raise InvalidEdgeError(
-                    f"Collector output type does not match collector input type: {edge.source.node_id}.{edge.source.field} to {edge.destination.node_id}.{edge.destination.field}"
-                )
-
-        # Validate that we are not connecting collector to iterator (currently unsupported)
-        if isinstance(from_node, CollectInvocation) and isinstance(to_node, IterateInvocation):
-            raise InvalidEdgeError(
-                f"Cannot connect collector to iterator: {edge.source.node_id}.{edge.source.field} to {edge.destination.node_id}.{edge.destination.field}"
-            )
+            err = self._is_collector_connection_valid(edge.destination.node_id, new_input=edge.source)
+            if err is not None:
+                raise InvalidEdgeError(f"Collector output type does not match collector input type ({edge}): {err}")
 
         # Validate if collector output type matches input type (if this edge results in both being set) - skip if the destination field is not Any or list[Any]
         if (
@@ -548,10 +536,9 @@ class Graph(BaseModel):
             and not self._is_destination_field_list_of_Any(edge)
             and not self._is_destination_field_Any(edge)
         ):
-            if not self._is_collector_connection_valid(edge.source.node_id, new_output=edge.destination):
-                raise InvalidEdgeError(
-                    f"Collector input type does not match collector output type: {edge.source.node_id}.{edge.source.field} to {edge.destination.node_id}.{edge.destination.field}"
-                )
+            err = self._is_collector_connection_valid(edge.source.node_id, new_output=edge.destination)
+            if err is not None:
+                raise InvalidEdgeError(f"Collector input type does not match collector output type ({edge}): {err}")
 
     def has_node(self, node_id: str) -> bool:
         """Determines whether or not a node exists in the graph."""
@@ -634,7 +621,7 @@ class Graph(BaseModel):
         node_id: str,
         new_input: Optional[EdgeConnection] = None,
         new_output: Optional[EdgeConnection] = None,
-    ) -> bool:
+    ) -> str | None:
         inputs = [e.source for e in self._get_input_edges(node_id, "collection")]
         outputs = [e.destination for e in self._get_output_edges(node_id, "item")]
 
@@ -645,29 +632,47 @@ class Graph(BaseModel):
 
         # Only one input is allowed for iterators
         if len(inputs) > 1:
-            return False
+            return "Iterator may only have one input edge"
+
+        input_node = self.get_node(inputs[0].node_id)
 
         # Get input and output fields (the fields linked to the iterator's input/output)
-        input_field = get_output_field(self.get_node(inputs[0].node_id), inputs[0].field)
-        output_fields = [get_input_field(self.get_node(e.node_id), e.field) for e in outputs]
+        input_field_type = get_output_field_type(input_node, inputs[0].field)
+        output_field_types = [get_input_field_type(self.get_node(e.node_id), e.field) for e in outputs]
 
         # Input type must be a list
-        if get_origin(input_field) is not list:
-            return False
+        if get_origin(input_field_type) is not list:
+            return "Iterator input must be a collection"
 
         # Validate that all outputs match the input type
-        input_field_item_type = get_args(input_field)[0]
-        if not all((are_connection_types_compatible(input_field_item_type, f) for f in output_fields)):
-            return False
+        input_field_item_type = get_args(input_field_type)[0]
+        if not all((are_connection_types_compatible(input_field_item_type, t) for t in output_field_types)):
+            return "Iterator outputs must connect to an input with a matching type"
 
-        return True
+        # Collector input type must match all iterator output types
+        if isinstance(input_node, CollectInvocation):
+            # Traverse the graph to find the first collector input edge. Collectors validate that their collection
+            # inputs are all of the same type, so we can use the first input edge to determine the collector's type
+            first_collector_input_edge = self._get_input_edges(input_node.id, "item")[0]
+            first_collector_input_type = get_output_field_type(
+                self.get_node(first_collector_input_edge.source.node_id), first_collector_input_edge.source.field
+            )
+            resolved_collector_type = (
+                first_collector_input_type
+                if get_origin(first_collector_input_type) is None
+                else get_args(first_collector_input_type)
+            )
+            if not all((are_connection_types_compatible(resolved_collector_type, t) for t in output_field_types)):
+                return "Iterator collection type must match all iterator output types"
+
+        return None
 
     def _is_collector_connection_valid(
         self,
         node_id: str,
         new_input: Optional[EdgeConnection] = None,
         new_output: Optional[EdgeConnection] = None,
-    ) -> bool:
+    ) -> str | None:
         inputs = [e.source for e in self._get_input_edges(node_id, "item")]
         outputs = [e.destination for e in self._get_output_edges(node_id, "collection")]
 
@@ -677,38 +682,42 @@ class Graph(BaseModel):
             outputs.append(new_output)
 
         # Get input and output fields (the fields linked to the iterator's input/output)
-        input_fields = [get_output_field(self.get_node(e.node_id), e.field) for e in inputs]
-        output_fields = [get_input_field(self.get_node(e.node_id), e.field) for e in outputs]
+        input_field_types = [get_output_field_type(self.get_node(e.node_id), e.field) for e in inputs]
+        output_field_types = [get_input_field_type(self.get_node(e.node_id), e.field) for e in outputs]
 
         # Validate that all inputs are derived from or match a single type
         input_field_types = {
-            t
-            for input_field in input_fields
-            for t in ([input_field] if get_origin(input_field) is None else get_args(input_field))
-            if t != NoneType
+            resolved_type
+            for input_field_type in input_field_types
+            for resolved_type in (
+                [input_field_type] if get_origin(input_field_type) is None else get_args(input_field_type)
+            )
+            if resolved_type != NoneType
         }  # Get unique types
         type_tree = nx.DiGraph()
         type_tree.add_nodes_from(input_field_types)
         type_tree.add_edges_from([e for e in itertools.permutations(input_field_types, 2) if issubclass(e[1], e[0])])
         type_degrees = type_tree.in_degree(type_tree.nodes)
         if sum((t[1] == 0 for t in type_degrees)) != 1:  # type: ignore
-            return False  # There is more than one root type
+            return "Collector input collection items must be of a single type"
 
         # Get the input root type
         input_root_type = next(t[0] for t in type_degrees if t[1] == 0)  # type: ignore
 
         # Verify that all outputs are lists
-        if not all(is_list_or_contains_list(f) for f in output_fields):
-            return False
+        if not all(is_list_or_contains_list(t) or is_any(t) for t in output_field_types):
+            return "Collector output must connect to a collection input"
 
         # Verify that all outputs match the input type (are a base class or the same class)
         if not all(
-            is_union_subtype(input_root_type, get_args(f)[0]) or issubclass(input_root_type, get_args(f)[0])
-            for f in output_fields
+            is_any(t)
+            or is_union_subtype(input_root_type, get_args(t)[0])
+            or issubclass(input_root_type, get_args(t)[0])
+            for t in output_field_types
         ):
-            return False
+            return "Collector outputs must connect to a collection input with a matching type"
 
-        return True
+        return None
 
     def nx_graph(self) -> nx.DiGraph:
         """Returns a NetworkX DiGraph representing the layout of this graph"""

invokeai/app/services/shared/invocation_context.py

@@ -9,6 +9,7 @@ from torch import Tensor
 
 from invokeai.app.invocations.constants import IMAGE_MODES
 from invokeai.app.invocations.fields import MetadataField, WithBoard, WithMetadata
+from invokeai.app.services.board_records.board_records_common import BoardRecordOrderBy
 from invokeai.app.services.boards.boards_common import BoardDTO
 from invokeai.app.services.config.config_default import InvokeAIAppConfig
 from invokeai.app.services.image_records.image_records_common import ImageCategory, ResourceOrigin
@@ -16,6 +17,7 @@ from invokeai.app.services.images.images_common import ImageDTO
 from invokeai.app.services.invocation_services import InvocationServices
 from invokeai.app.services.model_records.model_records_base import UnknownModelException
 from invokeai.app.services.session_processor.session_processor_common import ProgressImage
+from invokeai.app.services.shared.sqlite.sqlite_common import SQLiteDirection
 from invokeai.app.util.step_callback import flux_step_callback, stable_diffusion_step_callback
 from invokeai.backend.model_manager.config import (
     AnyModel,
@@ -102,7 +104,9 @@ class BoardsInterface(InvocationContextInterface):
         Returns:
             A list of all boards.
         """
-        return self._services.boards.get_all()
+        return self._services.boards.get_all(
+            order_by=BoardRecordOrderBy.CreatedAt, direction=SQLiteDirection.Descending
+        )
 
     def add_image_to_board(self, board_id: str, image_name: str) -> None:
         """Adds an image to a board.
@@ -122,7 +126,11 @@ class BoardsInterface(InvocationContextInterface):
         Returns:
             A list of all image names for the board.
         """
-        return self._services.board_images.get_all_board_image_names_for_board(board_id)
+        return self._services.board_images.get_all_board_image_names_for_board(
+            board_id,
+            categories=None,
+            is_intermediate=None,
+        )
 
 
 class LoggerInterface(InvocationContextInterface):
@@ -283,7 +291,7 @@ class ImagesInterface(InvocationContextInterface):
         Returns:
             The local path of the image or thumbnail.
         """
-        return self._services.images.get_path(image_name, thumbnail)
+        return Path(self._services.images.get_path(image_name, thumbnail))
 
 
 class TensorsInterface(InvocationContextInterface):

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

@@ -1,5 +1,4 @@
 import sqlite3
-import threading
 from logging import Logger
 from pathlib import Path
 
@@ -38,14 +37,20 @@ class SqliteDatabase:
             self.logger.info(f"Initializing database at {self.db_path}")
 
         self.conn = sqlite3.connect(database=self.db_path or sqlite_memory, check_same_thread=False)
-        self.lock = threading.RLock()
         self.conn.row_factory = sqlite3.Row
 
         if self.verbose:
             self.conn.set_trace_callback(self.logger.debug)
 
+        # Enable foreign key constraints
         self.conn.execute("PRAGMA foreign_keys = ON;")
 
+        # Enable Write-Ahead Logging (WAL) mode for better concurrency
+        self.conn.execute("PRAGMA journal_mode = WAL;")
+
+        # Set a busy timeout to prevent database lockups during writes
+        self.conn.execute("PRAGMA busy_timeout = 5000;")  # 5 seconds
+
     def clean(self) -> None:
         """
         Cleans the database by running the VACUUM command, reporting on the freed space.
@@ -53,15 +58,14 @@ class SqliteDatabase:
         # No need to clean in-memory database
         if not self.db_path:
             return
-        with self.lock:
-            try:
-                initial_db_size = Path(self.db_path).stat().st_size
-                self.conn.execute("VACUUM;")
-                self.conn.commit()
-                final_db_size = Path(self.db_path).stat().st_size
-                freed_space_in_mb = round((initial_db_size - final_db_size) / 1024 / 1024, 2)
-                if freed_space_in_mb > 0:
-                    self.logger.info(f"Cleaned database (freed {freed_space_in_mb}MB)")
-            except Exception as e:
-                self.logger.error(f"Error cleaning database: {e}")
-                raise
+        try:
+            initial_db_size = Path(self.db_path).stat().st_size
+            self.conn.execute("VACUUM;")
+            self.conn.commit()
+            final_db_size = Path(self.db_path).stat().st_size
+            freed_space_in_mb = round((initial_db_size - final_db_size) / 1024 / 1024, 2)
+            if freed_space_in_mb > 0:
+                self.logger.info(f"Cleaned database (freed {freed_space_in_mb}MB)")
+        except Exception as e:
+            self.logger.error(f"Error cleaning database: {e}")
+            raise

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

@@ -18,6 +18,7 @@ from invokeai.app.services.shared.sqlite_migrator.migrations.migration_12 import
 from invokeai.app.services.shared.sqlite_migrator.migrations.migration_13 import build_migration_13
 from invokeai.app.services.shared.sqlite_migrator.migrations.migration_14 import build_migration_14
 from invokeai.app.services.shared.sqlite_migrator.migrations.migration_15 import build_migration_15
+from invokeai.app.services.shared.sqlite_migrator.migrations.migration_16 import build_migration_16
 from invokeai.app.services.shared.sqlite_migrator.sqlite_migrator_impl import SqliteMigrator
 
 
@@ -53,6 +54,7 @@ def init_db(config: InvokeAIAppConfig, logger: Logger, image_files: ImageFileSto
     migrator.register_migration(build_migration_13())
     migrator.register_migration(build_migration_14())
     migrator.register_migration(build_migration_15())
+    migrator.register_migration(build_migration_16())
     migrator.run_migrations()
 
     return db

invokeai/app/services/shared/sqlite_migrator/migrations/migration_16.py

@@ -0,0 +1,31 @@
+import sqlite3
+
+from invokeai.app.services.shared.sqlite_migrator.sqlite_migrator_common import Migration
+
+
+class Migration16Callback:
+    def __call__(self, cursor: sqlite3.Cursor) -> None:
+        self._add_retried_from_item_id_col(cursor)
+
+    def _add_retried_from_item_id_col(self, cursor: sqlite3.Cursor) -> None:
+        """
+        - Adds `retried_from_item_id` column to the session queue table.
+        """
+
+        cursor.execute("ALTER TABLE session_queue ADD COLUMN retried_from_item_id INTEGER;")
+
+
+def build_migration_16() -> Migration:
+    """
+    Build the migration from database version 15 to 16.
+
+    This migration does the following:
+        - Adds `retried_from_item_id` column to the session queue table.
+    """
+    migration_16 = Migration(
+        from_version=15,
+        to_version=16,
+        callback=Migration16Callback(),
+    )
+
+    return migration_16

invokeai/app/services/shared/sqlite_migrator/sqlite_migrator_impl.py

@@ -43,46 +43,45 @@ class SqliteMigrator:
 
     def run_migrations(self) -> bool:
         """Migrates the database to the latest version."""
-        with self._db.lock:
-            # This throws if there is a problem.
-            self._migration_set.validate_migration_chain()
-            cursor = self._db.conn.cursor()
-            self._create_migrations_table(cursor=cursor)
+        # This throws if there is a problem.
+        self._migration_set.validate_migration_chain()
+        cursor = self._db.conn.cursor()
+        self._create_migrations_table(cursor=cursor)
 
-            if self._migration_set.count == 0:
-                self._logger.debug("No migrations registered")
-                return False
+        if self._migration_set.count == 0:
+            self._logger.debug("No migrations registered")
+            return False
 
-            if self._get_current_version(cursor=cursor) == self._migration_set.latest_version:
-                self._logger.debug("Database is up to date, no migrations to run")
-                return False
+        if self._get_current_version(cursor=cursor) == self._migration_set.latest_version:
+            self._logger.debug("Database is up to date, no migrations to run")
+            return False
 
-            self._logger.info("Database update needed")
+        self._logger.info("Database update needed")
 
-            # Make a backup of the db if it needs to be updated and is a file db
-            if self._db.db_path is not None:
-                timestamp = datetime.now().strftime("%Y%m%d-%H%M%S")
-                self._backup_path = self._db.db_path.parent / f"{self._db.db_path.stem}_backup_{timestamp}.db"
-                self._logger.info(f"Backing up database to {str(self._backup_path)}")
-                # Use SQLite to do the backup
-                with closing(sqlite3.connect(self._backup_path)) as backup_conn:
-                    self._db.conn.backup(backup_conn)
-            else:
-                self._logger.info("Using in-memory database, no backup needed")
+        # Make a backup of the db if it needs to be updated and is a file db
+        if self._db.db_path is not None:
+            timestamp = datetime.now().strftime("%Y%m%d-%H%M%S")
+            self._backup_path = self._db.db_path.parent / f"{self._db.db_path.stem}_backup_{timestamp}.db"
+            self._logger.info(f"Backing up database to {str(self._backup_path)}")
+            # Use SQLite to do the backup
+            with closing(sqlite3.connect(self._backup_path)) as backup_conn:
+                self._db.conn.backup(backup_conn)
+        else:
+            self._logger.info("Using in-memory database, no backup needed")
 
-            next_migration = self._migration_set.get(from_version=self._get_current_version(cursor))
-            while next_migration is not None:
-                self._run_migration(next_migration)
-                next_migration = self._migration_set.get(self._get_current_version(cursor))
-            self._logger.info("Database updated successfully")
-            return True
+        next_migration = self._migration_set.get(from_version=self._get_current_version(cursor))
+        while next_migration is not None:
+            self._run_migration(next_migration)
+            next_migration = self._migration_set.get(self._get_current_version(cursor))
+        self._logger.info("Database updated successfully")
+        return True
 
     def _run_migration(self, migration: Migration) -> None:
         """Runs a single migration."""
         try:
             # Using sqlite3.Connection as a context manager commits a the transaction on exit, or rolls it back if an
             # exception is raised.
-            with self._db.lock, self._db.conn as conn:
+            with self._db.conn as conn:
                 cursor = conn.cursor()
                 if self._get_current_version(cursor) != migration.from_version:
                     raise MigrationError(
@@ -108,27 +107,26 @@ class SqliteMigrator:
 
     def _create_migrations_table(self, cursor: sqlite3.Cursor) -> None:
         """Creates the migrations table for the database, if one does not already exist."""
-        with self._db.lock:
-            try:
-                cursor.execute("SELECT name FROM sqlite_master WHERE type='table' AND name='migrations';")
-                if cursor.fetchone() is not None:
-                    return
-                cursor.execute(
-                    """--sql
-                    CREATE TABLE migrations (
-                        version INTEGER PRIMARY KEY,
-                        migrated_at DATETIME NOT NULL DEFAULT(STRFTIME('%Y-%m-%d %H:%M:%f', 'NOW'))
-                    );
-                    """
-                )
-                cursor.execute("INSERT INTO migrations (version) VALUES (0);")
-                cursor.connection.commit()
-                self._logger.debug("Created migrations table")
-            except sqlite3.Error as e:
-                msg = f"Problem creating migrations table: {e}"
-                self._logger.error(msg)
-                cursor.connection.rollback()
-                raise MigrationError(msg) from e
+        try:
+            cursor.execute("SELECT name FROM sqlite_master WHERE type='table' AND name='migrations';")
+            if cursor.fetchone() is not None:
+                return
+            cursor.execute(
+                """--sql
+                CREATE TABLE migrations (
+                    version INTEGER PRIMARY KEY,
+                    migrated_at DATETIME NOT NULL DEFAULT(STRFTIME('%Y-%m-%d %H:%M:%f', 'NOW'))
+                );
+                """
+            )
+            cursor.execute("INSERT INTO migrations (version) VALUES (0);")
+            cursor.connection.commit()
+            self._logger.debug("Created migrations table")
+        except sqlite3.Error as e:
+            msg = f"Problem creating migrations table: {e}"
+            self._logger.error(msg)
+            cursor.connection.rollback()
+            raise MigrationError(msg) from e
 
     @classmethod
     def _get_current_version(cls, cursor: sqlite3.Cursor) -> int:

invokeai/app/services/style_preset_records/style_preset_records_sqlite.py

@@ -17,9 +17,7 @@ from invokeai.app.util.misc import uuid_string
 class SqliteStylePresetRecordsStorage(StylePresetRecordsStorageBase):
     def __init__(self, db: SqliteDatabase) -> None:
         super().__init__()
-        self._lock = db.lock
         self._conn = db.conn
-        self._cursor = self._conn.cursor()
 
     def start(self, invoker: Invoker) -> None:
         self._invoker = invoker
@@ -27,31 +25,25 @@ class SqliteStylePresetRecordsStorage(StylePresetRecordsStorageBase):
 
     def get(self, style_preset_id: str) -> StylePresetRecordDTO:
         """Gets a style preset by ID."""
-        try:
-            self._lock.acquire()
-            self._cursor.execute(
-                """--sql
-                SELECT *
-                FROM style_presets
-                WHERE id = ?;
-                """,
-                (style_preset_id,),
-            )
-            row = self._cursor.fetchone()
-            if row is None:
-                raise StylePresetNotFoundError(f"Style preset with id {style_preset_id} not found")
-            return StylePresetRecordDTO.from_dict(dict(row))
-        except Exception:
-            self._conn.rollback()
-            raise
-        finally:
-            self._lock.release()
+        cursor = self._conn.cursor()
+        cursor.execute(
+            """--sql
+            SELECT *
+            FROM style_presets
+            WHERE id = ?;
+            """,
+            (style_preset_id,),
+        )
+        row = cursor.fetchone()
+        if row is None:
+            raise StylePresetNotFoundError(f"Style preset with id {style_preset_id} not found")
+        return StylePresetRecordDTO.from_dict(dict(row))
 
     def create(self, style_preset: StylePresetWithoutId) -> StylePresetRecordDTO:
         style_preset_id = uuid_string()
         try:
-            self._lock.acquire()
-            self._cursor.execute(
+            cursor = self._conn.cursor()
+            cursor.execute(
                 """--sql
                 INSERT OR IGNORE INTO style_presets (
                     id,
@@ -72,18 +64,16 @@ class SqliteStylePresetRecordsStorage(StylePresetRecordsStorageBase):
         except Exception:
             self._conn.rollback()
             raise
-        finally:
-            self._lock.release()
         return self.get(style_preset_id)
 
     def create_many(self, style_presets: list[StylePresetWithoutId]) -> None:
         style_preset_ids = []
         try:
-            self._lock.acquire()
+            cursor = self._conn.cursor()
             for style_preset in style_presets:
                 style_preset_id = uuid_string()
                 style_preset_ids.append(style_preset_id)
-                self._cursor.execute(
+                cursor.execute(
                     """--sql
                     INSERT OR IGNORE INTO style_presets (
                         id,
@@ -104,17 +94,15 @@ class SqliteStylePresetRecordsStorage(StylePresetRecordsStorageBase):
         except Exception:
             self._conn.rollback()
             raise
-        finally:
-            self._lock.release()
 
         return None
 
     def update(self, style_preset_id: str, changes: StylePresetChanges) -> StylePresetRecordDTO:
         try:
-            self._lock.acquire()
+            cursor = self._conn.cursor()
             # Change the name of a style preset
             if changes.name is not None:
-                self._cursor.execute(
+                cursor.execute(
                     """--sql
                     UPDATE style_presets
                     SET name = ?
@@ -125,7 +113,7 @@ class SqliteStylePresetRecordsStorage(StylePresetRecordsStorageBase):
 
             # Change the preset data for a style preset
             if changes.preset_data is not None:
-                self._cursor.execute(
+                cursor.execute(
                     """--sql
                     UPDATE style_presets
                     SET preset_data = ?
@@ -138,14 +126,12 @@ class SqliteStylePresetRecordsStorage(StylePresetRecordsStorageBase):
         except Exception:
             self._conn.rollback()
             raise
-        finally:
-            self._lock.release()
         return self.get(style_preset_id)
 
     def delete(self, style_preset_id: str) -> None:
         try:
-            self._lock.acquire()
-            self._cursor.execute(
+            cursor = self._conn.cursor()
+            cursor.execute(
                 """--sql
                 DELETE from style_presets
                 WHERE id = ?;
@@ -156,46 +142,38 @@ class SqliteStylePresetRecordsStorage(StylePresetRecordsStorageBase):
         except Exception:
             self._conn.rollback()
             raise
-        finally:
-            self._lock.release()
         return None
 
     def get_many(self, type: PresetType | None = None) -> list[StylePresetRecordDTO]:
-        try:
-            self._lock.acquire()
-            main_query = """
-                SELECT
-                    *
-                FROM style_presets
-                """
+        main_query = """
+            SELECT
+                *
+            FROM style_presets
+            """
 
-            if type is not None:
-                main_query += "WHERE type = ? "
+        if type is not None:
+            main_query += "WHERE type = ? "
 
-            main_query += "ORDER BY LOWER(name) ASC"
+        main_query += "ORDER BY LOWER(name) ASC"
 
-            if type is not None:
-                self._cursor.execute(main_query, (type,))
-            else:
-                self._cursor.execute(main_query)
+        cursor = self._conn.cursor()
+        if type is not None:
+            cursor.execute(main_query, (type,))
+        else:
+            cursor.execute(main_query)
 
-            rows = self._cursor.fetchall()
-            style_presets = [StylePresetRecordDTO.from_dict(dict(row)) for row in rows]
+        rows = cursor.fetchall()
+        style_presets = [StylePresetRecordDTO.from_dict(dict(row)) for row in rows]
 
-            return style_presets
-        except Exception:
-            self._conn.rollback()
-            raise
-        finally:
-            self._lock.release()
+        return style_presets
 
     def _sync_default_style_presets(self) -> None:
         """Syncs default style presets to the database. Internal use only."""
 
         # First delete all existing default style presets
         try:
-            self._lock.acquire()
-            self._cursor.execute(
+            cursor = self._conn.cursor()
+            cursor.execute(
                 """--sql
                 DELETE FROM style_presets
                 WHERE type = "default";
@@ -205,10 +183,8 @@ class SqliteStylePresetRecordsStorage(StylePresetRecordsStorageBase):
         except Exception:
             self._conn.rollback()
             raise
-        finally:
-            self._lock.release()
         # Next, parse and create the default style presets
-        with self._lock, open(Path(__file__).parent / Path("default_style_presets.json"), "r") as file:
+        with open(Path(__file__).parent / Path("default_style_presets.json"), "r") as file:
             presets = json.load(file)
             for preset in presets:
                 style_preset = StylePresetWithoutId.model_validate(preset)

invokeai/app/services/workflow_records/workflow_records_common.py

@@ -62,9 +62,13 @@ class WorkflowWithoutID(BaseModel):
     notes: str = Field(description="The notes of the workflow.")
     exposedFields: list[ExposedField] = Field(description="The exposed fields of the workflow.")
     meta: WorkflowMeta = Field(description="The meta of the workflow.")
-    # TODO: nodes and edges are very loosely typed
+    # TODO(psyche): nodes, edges and form are very loosely typed - they are strictly modeled and checked on the frontend.
     nodes: list[dict[str, JsonValue]] = Field(description="The nodes of the workflow.")
     edges: list[dict[str, JsonValue]] = Field(description="The edges of the workflow.")
+    # TODO(psyche): We have a crapload of workflows that have no form, bc it was added after we introduced workflows.
+    # This is typed as optional to prevent errors when pulling workflows from the DB. The frontend adds a default form if
+    # it is None.
+    form: dict[str, JsonValue] | None = Field(default=None, description="The form of the workflow.")
 
     model_config = ConfigDict(extra="ignore")
 

invokeai/app/services/workflow_records/workflow_records_sqlite.py

@@ -23,9 +23,7 @@ from invokeai.app.util.misc import uuid_string
 class SqliteWorkflowRecordsStorage(WorkflowRecordsStorageBase):
     def __init__(self, db: SqliteDatabase) -> None:
         super().__init__()
-        self._lock = db.lock
         self._conn = db.conn
-        self._cursor = self._conn.cursor()
 
     def start(self, invoker: Invoker) -> None:
         self._invoker = invoker
@@ -33,42 +31,36 @@ class SqliteWorkflowRecordsStorage(WorkflowRecordsStorageBase):
 
     def get(self, workflow_id: str) -> WorkflowRecordDTO:
         """Gets a workflow by ID. Updates the opened_at column."""
-        try:
-            self._lock.acquire()
-            self._cursor.execute(
-                """--sql
-                UPDATE workflow_library
-                SET opened_at = STRFTIME('%Y-%m-%d %H:%M:%f', 'NOW')
-                WHERE workflow_id = ?;
-                """,
-                (workflow_id,),
-            )
-            self._conn.commit()
-            self._cursor.execute(
-                """--sql
-                SELECT workflow_id, workflow, name, created_at, updated_at, opened_at
-                FROM workflow_library
-                WHERE workflow_id = ?;
-                """,
-                (workflow_id,),
-            )
-            row = self._cursor.fetchone()
-            if row is None:
-                raise WorkflowNotFoundError(f"Workflow with id {workflow_id} not found")
-            return WorkflowRecordDTO.from_dict(dict(row))
-        except Exception:
-            self._conn.rollback()
-            raise
-        finally:
-            self._lock.release()
+        cursor = self._conn.cursor()
+        cursor.execute(
+            """--sql
+            UPDATE workflow_library
+            SET opened_at = STRFTIME('%Y-%m-%d %H:%M:%f', 'NOW')
+            WHERE workflow_id = ?;
+            """,
+            (workflow_id,),
+        )
+        self._conn.commit()
+        cursor.execute(
+            """--sql
+            SELECT workflow_id, workflow, name, created_at, updated_at, opened_at
+            FROM workflow_library
+            WHERE workflow_id = ?;
+            """,
+            (workflow_id,),
+        )
+        row = cursor.fetchone()
+        if row is None:
+            raise WorkflowNotFoundError(f"Workflow with id {workflow_id} not found")
+        return WorkflowRecordDTO.from_dict(dict(row))
 
     def create(self, workflow: WorkflowWithoutID) -> WorkflowRecordDTO:
         try:
             # Only user workflows may be created by this method
             assert workflow.meta.category is WorkflowCategory.User
             workflow_with_id = Workflow(**workflow.model_dump(), id=uuid_string())
-            self._lock.acquire()
-            self._cursor.execute(
+            cursor = self._conn.cursor()
+            cursor.execute(
                 """--sql
                 INSERT OR IGNORE INTO workflow_library (
                     workflow_id,
@@ -82,14 +74,12 @@ class SqliteWorkflowRecordsStorage(WorkflowRecordsStorageBase):
         except Exception:
             self._conn.rollback()
             raise
-        finally:
-            self._lock.release()
         return self.get(workflow_with_id.id)
 
     def update(self, workflow: Workflow) -> WorkflowRecordDTO:
         try:
-            self._lock.acquire()
-            self._cursor.execute(
+            cursor = self._conn.cursor()
+            cursor.execute(
                 """--sql
                 UPDATE workflow_library
                 SET workflow = ?
@@ -101,14 +91,12 @@ class SqliteWorkflowRecordsStorage(WorkflowRecordsStorageBase):
         except Exception:
             self._conn.rollback()
             raise
-        finally:
-            self._lock.release()
         return self.get(workflow.id)
 
     def delete(self, workflow_id: str) -> None:
         try:
-            self._lock.acquire()
-            self._cursor.execute(
+            cursor = self._conn.cursor()
+            cursor.execute(
                 """--sql
                 DELETE from workflow_library
                 WHERE workflow_id = ? AND category = 'user';
@@ -119,8 +107,6 @@ class SqliteWorkflowRecordsStorage(WorkflowRecordsStorageBase):
         except Exception:
             self._conn.rollback()
             raise
-        finally:
-            self._lock.release()
         return None
 
     def get_many(
@@ -132,66 +118,60 @@ class SqliteWorkflowRecordsStorage(WorkflowRecordsStorageBase):
         per_page: Optional[int] = None,
         query: Optional[str] = None,
     ) -> PaginatedResults[WorkflowRecordListItemDTO]:
-        try:
-            self._lock.acquire()
-            # sanitize!
-            assert order_by in WorkflowRecordOrderBy
-            assert direction in SQLiteDirection
-            assert category in WorkflowCategory
-            count_query = "SELECT COUNT(*) FROM workflow_library WHERE category = ?"
-            main_query = """
-                SELECT
-                    workflow_id,
-                    category,
-                    name,
-                    description,
-                    created_at,
-                    updated_at,
-                    opened_at
-                FROM workflow_library
-                WHERE category = ?
-                """
-            main_params: list[int | str] = [category.value]
-            count_params: list[int | str] = [category.value]
+        # sanitize!
+        assert order_by in WorkflowRecordOrderBy
+        assert direction in SQLiteDirection
+        assert category in WorkflowCategory
+        count_query = "SELECT COUNT(*) FROM workflow_library WHERE category = ?"
+        main_query = """
+            SELECT
+                workflow_id,
+                category,
+                name,
+                description,
+                created_at,
+                updated_at,
+                opened_at
+            FROM workflow_library
+            WHERE category = ?
+            """
+        main_params: list[int | str] = [category.value]
+        count_params: list[int | str] = [category.value]
 
-            stripped_query = query.strip() if query else None
-            if stripped_query:
-                wildcard_query = "%" + stripped_query + "%"
-                main_query += " AND name LIKE ? OR description LIKE ? "
-                count_query += " AND name LIKE ? OR description LIKE ?;"
-                main_params.extend([wildcard_query, wildcard_query])
-                count_params.extend([wildcard_query, wildcard_query])
+        stripped_query = query.strip() if query else None
+        if stripped_query:
+            wildcard_query = "%" + stripped_query + "%"
+            main_query += " AND name LIKE ? OR description LIKE ? "
+            count_query += " AND name LIKE ? OR description LIKE ?;"
+            main_params.extend([wildcard_query, wildcard_query])
+            count_params.extend([wildcard_query, wildcard_query])
 
-            main_query += f" ORDER BY {order_by.value} {direction.value}"
+        main_query += f" ORDER BY {order_by.value} {direction.value}"
 
-            if per_page:
-                main_query += " LIMIT ? OFFSET ?"
-                main_params.extend([per_page, page * per_page])
+        if per_page:
+            main_query += " LIMIT ? OFFSET ?"
+            main_params.extend([per_page, page * per_page])
 
-            self._cursor.execute(main_query, main_params)
-            rows = self._cursor.fetchall()
-            workflows = [WorkflowRecordListItemDTOValidator.validate_python(dict(row)) for row in rows]
+        cursor = self._conn.cursor()
+        cursor.execute(main_query, main_params)
+        rows = cursor.fetchall()
+        workflows = [WorkflowRecordListItemDTOValidator.validate_python(dict(row)) for row in rows]
 
-            self._cursor.execute(count_query, count_params)
-            total = self._cursor.fetchone()[0]
+        cursor.execute(count_query, count_params)
+        total = cursor.fetchone()[0]
 
-            if per_page:
-                pages = total // per_page + (total % per_page > 0)
-            else:
-                pages = 1  # If no pagination, there is only one page
+        if per_page:
+            pages = total // per_page + (total % per_page > 0)
+        else:
+            pages = 1  # If no pagination, there is only one page
 
-            return PaginatedResults(
-                items=workflows,
-                page=page,
-                per_page=per_page if per_page else total,
-                pages=pages,
-                total=total,
-            )
-        except Exception:
-            self._conn.rollback()
-            raise
-        finally:
-            self._lock.release()
+        return PaginatedResults(
+            items=workflows,
+            page=page,
+            per_page=per_page if per_page else total,
+            pages=pages,
+            total=total,
+        )
 
     def _sync_default_workflows(self) -> None:
         """Syncs default workflows to the database. Internal use only."""
@@ -207,7 +187,6 @@ class SqliteWorkflowRecordsStorage(WorkflowRecordsStorageBase):
         """
 
         try:
-            self._lock.acquire()
             workflows: list[Workflow] = []
             workflows_dir = Path(__file__).parent / Path("default_workflows")
             workflow_paths = workflows_dir.glob("*.json")
@@ -218,14 +197,15 @@ class SqliteWorkflowRecordsStorage(WorkflowRecordsStorageBase):
                 workflows.append(workflow)
             # Only default workflows may be managed by this method
             assert all(w.meta.category is WorkflowCategory.Default for w in workflows)
-            self._cursor.execute(
+            cursor = self._conn.cursor()
+            cursor.execute(
                 """--sql
                 DELETE FROM workflow_library
                 WHERE category = 'default';
                 """
             )
             for w in workflows:
-                self._cursor.execute(
+                cursor.execute(
                     """--sql
                     INSERT OR REPLACE INTO workflow_library (
                         workflow_id,
@@ -239,5 +219,3 @@ class SqliteWorkflowRecordsStorage(WorkflowRecordsStorageBase):
         except Exception:
             self._conn.rollback()
             raise
-        finally:
-            self._lock.release()

invokeai/app/util/startup_utils.py

@@ -0,0 +1,64 @@
+import logging
+import mimetypes
+import socket
+
+import torch
+
+
+def find_open_port(port: int) -> int:
+    """Find a port not in use starting at given port"""
+    # Taken from https://waylonwalker.com/python-find-available-port/, thanks Waylon!
+    # https://github.com/WaylonWalker
+    with socket.socket(socket.AF_INET, socket.SOCK_STREAM) as s:
+        s.settimeout(1)
+        if s.connect_ex(("localhost", port)) == 0:
+            return find_open_port(port=port + 1)
+        else:
+            return port
+
+
+def check_cudnn(logger: logging.Logger) -> None:
+    """Check for cuDNN issues that could be causing degraded performance."""
+    if torch.backends.cudnn.is_available():
+        try:
+            # Note: At the time of writing (torch 2.2.1), torch.backends.cudnn.version() only raises an error the first
+            # time it is called. Subsequent calls will return the version number without complaining about a mismatch.
+            cudnn_version = torch.backends.cudnn.version()
+            logger.info(f"cuDNN version: {cudnn_version}")
+        except RuntimeError as e:
+            logger.warning(
+                "Encountered a cuDNN version issue. This may result in degraded performance. This issue is usually "
+                "caused by an incompatible cuDNN version installed in your python environment, or on the host "
+                f"system. Full error message:\n{e}"
+            )
+
+
+def enable_dev_reload() -> None:
+    """Enable hot reloading on python file changes during development."""
+    from invokeai.backend.util.logging import InvokeAILogger
+
+    try:
+        import jurigged
+    except ImportError as e:
+        raise RuntimeError(
+            'Can\'t start `--dev_reload` because jurigged is not found; `pip install -e ".[dev]"` to include development dependencies.'
+        ) from e
+    else:
+        jurigged.watch(logger=InvokeAILogger.get_logger(name="jurigged").info)
+
+
+def apply_monkeypatches() -> None:
+    """Apply monkeypatches to fix issues with third-party libraries."""
+
+    import invokeai.backend.util.hotfixes  # noqa: F401 (monkeypatching on import)
+
+    if torch.backends.mps.is_available():
+        import invokeai.backend.util.mps_fixes  # noqa: F401 (monkeypatching on import)
+
+
+def register_mime_types() -> None:
+    """Register additional mime types for windows."""
+    # Fix for windows mimetypes registry entries being borked.
+    # see https://github.com/invoke-ai/InvokeAI/discussions/3684#discussioncomment-6391352
+    mimetypes.add_type("application/javascript", ".js")
+    mimetypes.add_type("text/css", ".css")

invokeai/app/util/t5_model_identifier.py

@@ -0,0 +1,26 @@
+from invokeai.app.invocations.model import ModelIdentifierField
+from invokeai.backend.model_manager.config import BaseModelType, SubModelType
+
+
+def preprocess_t5_encoder_model_identifier(model_identifier: ModelIdentifierField) -> ModelIdentifierField:
+    """A helper function to normalize a T5 encoder model identifier so that T5 models associated with FLUX
+    or SD3 models can be used interchangeably.
+    """
+    if model_identifier.base == BaseModelType.Any:
+        return model_identifier.model_copy(update={"submodel_type": SubModelType.TextEncoder2})
+    elif model_identifier.base == BaseModelType.StableDiffusion3:
+        return model_identifier.model_copy(update={"submodel_type": SubModelType.TextEncoder3})
+    else:
+        raise ValueError(f"Unsupported model base: {model_identifier.base}")
+
+
+def preprocess_t5_tokenizer_model_identifier(model_identifier: ModelIdentifierField) -> ModelIdentifierField:
+    """A helper function to normalize a T5 tokenizer model identifier so that T5 models associated with FLUX
+    or SD3 models can be used interchangeably.
+    """
+    if model_identifier.base == BaseModelType.Any:
+        return model_identifier.model_copy(update={"submodel_type": SubModelType.Tokenizer2})
+    elif model_identifier.base == BaseModelType.StableDiffusion3:
+        return model_identifier.model_copy(update={"submodel_type": SubModelType.Tokenizer3})
+    else:
+        raise ValueError(f"Unsupported model base: {model_identifier.base}")

invokeai/app/util/torch_cuda_allocator.py

@@ -0,0 +1,52 @@
+import logging
+import os
+import sys
+
+
+def configure_torch_cuda_allocator(pytorch_cuda_alloc_conf: str, logger: logging.Logger):
+    """Configure the PyTorch CUDA memory allocator. See
+    https://pytorch.org/docs/stable/notes/cuda.html#optimizing-memory-usage-with-pytorch-cuda-alloc-conf for supported
+    configurations.
+    """
+
+    if "torch" in sys.modules:
+        raise RuntimeError("configure_torch_cuda_allocator() must be called before importing torch.")
+
+    # Log a warning if the PYTORCH_CUDA_ALLOC_CONF environment variable is already set.
+    prev_cuda_alloc_conf = os.environ.get("PYTORCH_CUDA_ALLOC_CONF", None)
+    if prev_cuda_alloc_conf is not None:
+        if prev_cuda_alloc_conf == pytorch_cuda_alloc_conf:
+            logger.info(
+                f"PYTORCH_CUDA_ALLOC_CONF is already set to '{pytorch_cuda_alloc_conf}'. Skipping configuration."
+            )
+            return
+        else:
+            logger.warning(
+                f"Attempted to configure the PyTorch CUDA memory allocator with '{pytorch_cuda_alloc_conf}', but PYTORCH_CUDA_ALLOC_CONF is already set to "
+                f"'{prev_cuda_alloc_conf}'. Skipping configuration."
+            )
+            return
+
+    # Configure the PyTorch CUDA memory allocator.
+    # NOTE: It is important that this happens before torch is imported.
+    os.environ["PYTORCH_CUDA_ALLOC_CONF"] = pytorch_cuda_alloc_conf
+
+    import torch
+
+    # Relevant docs: https://pytorch.org/docs/stable/notes/cuda.html#optimizing-memory-usage-with-pytorch-cuda-alloc-conf
+    if not torch.cuda.is_available():
+        raise RuntimeError(
+            "Attempted to configure the PyTorch CUDA memory allocator, but no CUDA devices are available."
+        )
+
+    # Verify that the torch allocator was properly configured.
+    allocator_backend = torch.cuda.get_allocator_backend()
+    expected_backend = "cudaMallocAsync" if "cudaMallocAsync" in pytorch_cuda_alloc_conf else "native"
+    if allocator_backend != expected_backend:
+        raise RuntimeError(
+            f"Failed to configure the PyTorch CUDA memory allocator. Expected backend: '{expected_backend}', but got "
+            f"'{allocator_backend}'. Verify that 1) the pytorch_cuda_alloc_conf is set correctly, and 2) that torch is "
+            "not imported before calling configure_torch_cuda_allocator()."
+        )
+
+    logger.info(f"PyTorch CUDA memory allocator: {torch.cuda.get_allocator_backend()}")

invokeai/backend/flux/modules/conditioner.py

@@ -1,13 +1,19 @@
 # Initially pulled from https://github.com/black-forest-labs/flux
 
 from torch import Tensor, nn
-from transformers import PreTrainedModel, PreTrainedTokenizer
+from transformers import PreTrainedModel, PreTrainedTokenizer, PreTrainedTokenizerFast
 
 from invokeai.backend.util.devices import TorchDevice
 
 
 class HFEncoder(nn.Module):
-    def __init__(self, encoder: PreTrainedModel, tokenizer: PreTrainedTokenizer, is_clip: bool, max_length: int):
+    def __init__(
+        self,
+        encoder: PreTrainedModel,
+        tokenizer: PreTrainedTokenizer | PreTrainedTokenizerFast,
+        is_clip: bool,
+        max_length: int,
+    ):
         super().__init__()
         self.max_length = max_length
         self.is_clip = is_clip

invokeai/backend/model_manager/load/model_cache/cached_model/cached_model_only_full_load.py

@@ -9,12 +9,17 @@ class CachedModelOnlyFullLoad:
     MPS memory, etc.
     """
 
-    def __init__(self, model: torch.nn.Module | Any, compute_device: torch.device, total_bytes: int):
+    def __init__(
+        self, model: torch.nn.Module | Any, compute_device: torch.device, total_bytes: int, keep_ram_copy: bool = False
+    ):
         """Initialize a CachedModelOnlyFullLoad.
         Args:
             model (torch.nn.Module | Any): The model to wrap. Should be on the CPU.
             compute_device (torch.device): The compute device to move the model to.
             total_bytes (int): The total size (in bytes) of all the weights in the model.
+            keep_ram_copy (bool): Whether to keep a read-only copy of the model's state dict in RAM. Keeping a RAM copy
+                increases RAM usage, but speeds up model offload from VRAM and LoRA patching (assuming there is
+                sufficient RAM).
         """
         # model is often a torch.nn.Module, but could be any model type. Throughout this class, we handle both cases.
         self._model = model
@@ -23,7 +28,7 @@ class CachedModelOnlyFullLoad:
 
         # A CPU read-only copy of the model's state dict.
         self._cpu_state_dict: dict[str, torch.Tensor] | None = None
-        if isinstance(model, torch.nn.Module):
+        if isinstance(model, torch.nn.Module) and keep_ram_copy:
             self._cpu_state_dict = model.state_dict()
 
         self._total_bytes = total_bytes

invokeai/backend/model_manager/load/model_cache/cached_model/cached_model_with_partial_load.py

@@ -14,33 +14,38 @@ class CachedModelWithPartialLoad:
     MPS memory, etc.
     """
 
-    def __init__(self, model: torch.nn.Module, compute_device: torch.device):
+    def __init__(self, model: torch.nn.Module, compute_device: torch.device, keep_ram_copy: bool = False):
         self._model = model
         self._compute_device = compute_device
 
-        # A CPU read-only copy of the model's state dict.
-        self._cpu_state_dict: dict[str, torch.Tensor] = model.state_dict()
+        model_state_dict = model.state_dict()
+        # A CPU read-only copy of the model's state dict. Used for faster model unloads from VRAM, and to speed up LoRA
+        # patching. Set to `None` if keep_ram_copy is False.
+        self._cpu_state_dict: dict[str, torch.Tensor] | None = model_state_dict if keep_ram_copy else None
 
         # A dictionary of the size of each tensor in the state dict.
         # HACK(ryand): We use this dictionary any time we are doing byte tracking calculations. We do this for
         # consistency in case the application code has modified the model's size (e.g. by casting to a different
         # precision). Of course, this means that we are making model cache load/unload decisions based on model size
         # data that may not be fully accurate.
-        self._state_dict_bytes = {k: calc_tensor_size(v) for k, v in self._cpu_state_dict.items()}
+        self._state_dict_bytes = {k: calc_tensor_size(v) for k, v in model_state_dict.items()}
 
         self._total_bytes = sum(self._state_dict_bytes.values())
         self._cur_vram_bytes: int | None = None
 
         self._modules_that_support_autocast = self._find_modules_that_support_autocast()
-        self._keys_in_modules_that_do_not_support_autocast = self._find_keys_in_modules_that_do_not_support_autocast()
+        self._keys_in_modules_that_do_not_support_autocast = self._find_keys_in_modules_that_do_not_support_autocast(
+            model_state_dict
+        )
+        self._state_dict_keys_by_module_prefix = self._group_state_dict_keys_by_module_prefix(model_state_dict)
 
     def _find_modules_that_support_autocast(self) -> dict[str, torch.nn.Module]:
         """Find all modules that support autocasting."""
         return {n: m for n, m in self._model.named_modules() if isinstance(m, CustomModuleMixin)}  # type: ignore
 
-    def _find_keys_in_modules_that_do_not_support_autocast(self) -> set[str]:
+    def _find_keys_in_modules_that_do_not_support_autocast(self, state_dict: dict[str, torch.Tensor]) -> set[str]:
         keys_in_modules_that_do_not_support_autocast: set[str] = set()
-        for key in self._cpu_state_dict.keys():
+        for key in state_dict.keys():
             for module_name in self._modules_that_support_autocast.keys():
                 if key.startswith(module_name):
                     break
@@ -48,6 +53,47 @@ class CachedModelWithPartialLoad:
                 keys_in_modules_that_do_not_support_autocast.add(key)
         return keys_in_modules_that_do_not_support_autocast
 
+    def _group_state_dict_keys_by_module_prefix(self, state_dict: dict[str, torch.Tensor]) -> dict[str, list[str]]:
+        """A helper function that groups state dict keys by module prefix.
+
+        Example:
+        ```
+        state_dict = {
+            "weight": ...,
+            "module.submodule.weight": ...,
+            "module.submodule.bias": ...,
+            "module.other_submodule.weight": ...,
+            "module.other_submodule.bias": ...,
+        }
+
+        output = group_state_dict_keys_by_module_prefix(state_dict)
+
+        # The output will be:
+        output = {
+            "": [
+                "weight",
+            ],
+            "module.submodule": [
+                "module.submodule.weight",
+                "module.submodule.bias",
+            ],
+            "module.other_submodule": [
+                "module.other_submodule.weight",
+                "module.other_submodule.bias",
+            ],
+        }
+        ```
+        """
+        state_dict_keys_by_module_prefix: dict[str, list[str]] = {}
+        for key in state_dict.keys():
+            split = key.rsplit(".", 1)
+            # `split` will have length 1 if the root module has parameters.
+            module_name = split[0] if len(split) > 1 else ""
+            if module_name not in state_dict_keys_by_module_prefix:
+                state_dict_keys_by_module_prefix[module_name] = []
+            state_dict_keys_by_module_prefix[module_name].append(key)
+        return state_dict_keys_by_module_prefix
+
     def _move_non_persistent_buffers_to_device(self, device: torch.device):
         """Move the non-persistent buffers to the target device. These buffers are not included in the state dict,
         so we need to move them manually.
@@ -98,6 +144,82 @@ class CachedModelWithPartialLoad:
         """Unload all weights from VRAM."""
         return self.partial_unload_from_vram(self.total_bytes())
 
+    def _load_state_dict_with_device_conversion(
+        self, state_dict: dict[str, torch.Tensor], keys_to_convert: set[str], target_device: torch.device
+    ):
+        if self._cpu_state_dict is not None:
+            # Run the fast version.
+            self._load_state_dict_with_fast_device_conversion(
+                state_dict=state_dict,
+                keys_to_convert=keys_to_convert,
+                target_device=target_device,
+                cpu_state_dict=self._cpu_state_dict,
+            )
+        else:
+            # Run the low-virtual-memory version.
+            self._load_state_dict_with_jit_device_conversion(
+                state_dict=state_dict,
+                keys_to_convert=keys_to_convert,
+                target_device=target_device,
+            )
+
+    def _load_state_dict_with_jit_device_conversion(
+        self,
+        state_dict: dict[str, torch.Tensor],
+        keys_to_convert: set[str],
+        target_device: torch.device,
+    ):
+        """A custom state dict loading implementation with good peak memory properties.
+
+        This implementation has the important property that it copies parameters to the target device one module at a time
+        rather than applying all of the device conversions and then calling load_state_dict(). This is done to minimize the
+        peak virtual memory usage. Specifically, we want to avoid a case where we hold references to all of the CPU weights
+        and CUDA weights simultaneously, because Windows will reserve virtual memory for both.
+        """
+        for module_name, module in self._model.named_modules():
+            module_keys = self._state_dict_keys_by_module_prefix.get(module_name, [])
+            # Calculate the length of the module name prefix.
+            prefix_len = len(module_name)
+            if prefix_len > 0:
+                prefix_len += 1
+
+            module_state_dict = {}
+            for key in module_keys:
+                if key in keys_to_convert:
+                    # It is important that we overwrite `state_dict[key]` to avoid keeping two copies of the same
+                    # parameter.
+                    state_dict[key] = state_dict[key].to(target_device)
+                # Note that we keep parameters that have not been moved to a new device in case the module implements
+                # weird custom state dict loading logic that requires all parameters to be present.
+                module_state_dict[key[prefix_len:]] = state_dict[key]
+
+            if len(module_state_dict) > 0:
+                # We set strict=False, because if `module` has both parameters and child modules, then we are loading a
+                # state dict that only contains the parameters of `module` (not its children).
+                # We assume that it is rare for non-leaf modules to have parameters. Calling load_state_dict() on non-leaf
+                # modules will recurse through all of the children, so is a bit wasteful.
+                incompatible_keys = module.load_state_dict(module_state_dict, strict=False, assign=True)
+                # Missing keys are ok, unexpected keys are not.
+                assert len(incompatible_keys.unexpected_keys) == 0
+
+    def _load_state_dict_with_fast_device_conversion(
+        self,
+        state_dict: dict[str, torch.Tensor],
+        keys_to_convert: set[str],
+        target_device: torch.device,
+        cpu_state_dict: dict[str, torch.Tensor],
+    ):
+        """Convert parameters to the target device and load them into the model. Leverages the `cpu_state_dict` to speed
+        up transfers of weights to the CPU.
+        """
+        for key in keys_to_convert:
+            if target_device.type == "cpu":
+                state_dict[key] = cpu_state_dict[key]
+            else:
+                state_dict[key] = state_dict[key].to(target_device)
+
+        self._model.load_state_dict(state_dict, assign=True)
+
     @torch.no_grad()
     def partial_load_to_vram(self, vram_bytes_to_load: int) -> int:
         """Load more weights into VRAM without exceeding vram_bytes_to_load.
@@ -112,26 +234,33 @@ class CachedModelWithPartialLoad:
 
         cur_state_dict = self._model.state_dict()
 
+        # Identify the keys that will be loaded into VRAM.
+        keys_to_load: set[str] = set()
+
         # First, process the keys that *must* be loaded into VRAM.
         for key in self._keys_in_modules_that_do_not_support_autocast:
             param = cur_state_dict[key]
             if param.device.type == self._compute_device.type:
                 continue
 
+            keys_to_load.add(key)
             param_size = self._state_dict_bytes[key]
-            cur_state_dict[key] = param.to(self._compute_device, copy=True)
             vram_bytes_loaded += param_size
 
         if vram_bytes_loaded > vram_bytes_to_load:
             logger = InvokeAILogger.get_logger()
             logger.warning(
-                f"Loaded {vram_bytes_loaded / 2**20} MB into VRAM, but only {vram_bytes_to_load / 2**20} MB were "
+                f"Loading {vram_bytes_loaded / 2**20} MB into VRAM, but only {vram_bytes_to_load / 2**20} MB were "
                 "requested. This is the minimum set of weights in VRAM required to run the model."
             )
 
         # Next, process the keys that can optionally be loaded into VRAM.
         fully_loaded = True
         for key, param in cur_state_dict.items():
+            # Skip the keys that have already been processed above.
+            if key in keys_to_load:
+                continue
+
             if param.device.type == self._compute_device.type:
                 continue
 
@@ -142,14 +271,14 @@ class CachedModelWithPartialLoad:
                 fully_loaded = False
                 continue
 
-            cur_state_dict[key] = param.to(self._compute_device, copy=True)
+            keys_to_load.add(key)
             vram_bytes_loaded += param_size
 
-        if vram_bytes_loaded > 0:
+        if len(keys_to_load) > 0:
             # We load the entire state dict, not just the parameters that changed, in case there are modules that
             # override _load_from_state_dict() and do some funky stuff that requires the entire state dict.
             # Alternatively, in the future, grouping parameters by module could probably solve this problem.
-            self._model.load_state_dict(cur_state_dict, assign=True)
+            self._load_state_dict_with_device_conversion(cur_state_dict, keys_to_load, self._compute_device)
 
         if self._cur_vram_bytes is not None:
             self._cur_vram_bytes += vram_bytes_loaded
@@ -180,6 +309,10 @@ class CachedModelWithPartialLoad:
 
         offload_device = "cpu"
         cur_state_dict = self._model.state_dict()
+
+        # Identify the keys that will be offloaded to CPU.
+        keys_to_offload: set[str] = set()
+
         for key, param in cur_state_dict.items():
             if vram_bytes_freed >= vram_bytes_to_free:
                 break
@@ -191,11 +324,11 @@ class CachedModelWithPartialLoad:
                 required_weights_in_vram += self._state_dict_bytes[key]
                 continue
 
-            cur_state_dict[key] = self._cpu_state_dict[key]
+            keys_to_offload.add(key)
             vram_bytes_freed += self._state_dict_bytes[key]
 
-        if vram_bytes_freed > 0:
-            self._model.load_state_dict(cur_state_dict, assign=True)
+        if len(keys_to_offload) > 0:
+            self._load_state_dict_with_device_conversion(cur_state_dict, keys_to_offload, torch.device("cpu"))
 
         if self._cur_vram_bytes is not None:
             self._cur_vram_bytes -= vram_bytes_freed

invokeai/backend/model_manager/load/model_cache/model_cache.py

@@ -1,8 +1,10 @@
 import gc
 import logging
+import threading
 import time
+from functools import wraps
 from logging import Logger
-from typing import Dict, List, Optional
+from typing import Any, Callable, Dict, List, Optional
 
 import psutil
 import torch
@@ -41,6 +43,17 @@ def get_model_cache_key(model_key: str, submodel_type: Optional[SubModelType] =
         return model_key
 
 
+def synchronized(method: Callable[..., Any]) -> Callable[..., Any]:
+    """A decorator that applies the class's self._lock to the method."""
+
+    @wraps(method)
+    def wrapper(self, *args, **kwargs):
+        with self._lock:  # Automatically acquire and release the lock
+            return method(self, *args, **kwargs)
+
+    return wrapper
+
+
 class ModelCache:
     """A cache for managing models in memory.
 
@@ -78,6 +91,7 @@ class ModelCache:
         self,
         execution_device_working_mem_gb: float,
         enable_partial_loading: bool,
+        keep_ram_copy_of_weights: bool,
         max_ram_cache_size_gb: float | None = None,
         max_vram_cache_size_gb: float | None = None,
         execution_device: torch.device | str = "cuda",
@@ -105,6 +119,7 @@ class ModelCache:
         :param logger: InvokeAILogger to use (otherwise creates one)
         """
         self._enable_partial_loading = enable_partial_loading
+        self._keep_ram_copy_of_weights = keep_ram_copy_of_weights
         self._execution_device_working_mem_gb = execution_device_working_mem_gb
         self._execution_device: torch.device = torch.device(execution_device)
         self._storage_device: torch.device = torch.device(storage_device)
@@ -121,16 +136,27 @@ class ModelCache:
         self._cached_models: Dict[str, CacheRecord] = {}
         self._cache_stack: List[str] = []
 
+        self._ram_cache_size_bytes = self._calc_ram_available_to_model_cache()
+
+        # A lock applied to all public method calls to make the ModelCache thread-safe.
+        # At the time of writing, the ModelCache should only be accessed from two threads:
+        # - The graph execution thread
+        # - Requests to empty the cache from a separate thread
+        self._lock = threading.RLock()
+
     @property
+    @synchronized
     def stats(self) -> Optional[CacheStats]:
         """Return collected CacheStats object."""
         return self._stats
 
     @stats.setter
+    @synchronized
     def stats(self, stats: CacheStats) -> None:
         """Set the CacheStats object for collecting cache statistics."""
         self._stats = stats
 
+    @synchronized
     def put(self, key: str, model: AnyModel) -> None:
         """Add a model to the cache."""
         if key in self._cached_models:
@@ -154,9 +180,13 @@ class ModelCache:
 
         # Wrap model.
         if isinstance(model, torch.nn.Module) and running_with_cuda and self._enable_partial_loading:
-            wrapped_model = CachedModelWithPartialLoad(model, self._execution_device)
+            wrapped_model = CachedModelWithPartialLoad(
+                model, self._execution_device, keep_ram_copy=self._keep_ram_copy_of_weights
+            )
         else:
-            wrapped_model = CachedModelOnlyFullLoad(model, self._execution_device, size)
+            wrapped_model = CachedModelOnlyFullLoad(
+                model, self._execution_device, size, keep_ram_copy=self._keep_ram_copy_of_weights
+            )
 
         cache_record = CacheRecord(key=key, cached_model=wrapped_model)
         self._cached_models[key] = cache_record
@@ -165,6 +195,7 @@ class ModelCache:
             f"Added model {key} (Type: {model.__class__.__name__}, Wrap mode: {wrapped_model.__class__.__name__}, Model size: {size/MB:.2f}MB)"
         )
 
+    @synchronized
     def get(self, key: str, stats_name: Optional[str] = None) -> CacheRecord:
         """Retrieve a model from the cache.
 
@@ -200,6 +231,7 @@ class ModelCache:
         self._logger.debug(f"Cache hit: {key} (Type: {cache_entry.cached_model.model.__class__.__name__})")
         return cache_entry
 
+    @synchronized
     def lock(self, cache_entry: CacheRecord, working_mem_bytes: Optional[int]) -> None:
         """Lock a model for use and move it into VRAM."""
         if cache_entry.key not in self._cached_models:
@@ -235,6 +267,7 @@ class ModelCache:
 
         self._log_cache_state()
 
+    @synchronized
     def unlock(self, cache_entry: CacheRecord) -> None:
         """Unlock a model."""
         if cache_entry.key not in self._cached_models:
@@ -382,41 +415,77 @@ class ModelCache:
         # Alternative definition of VRAM in use:
         # return sum(ce.cached_model.cur_vram_bytes() for ce in self._cached_models.values())
 
-    def _get_ram_available(self) -> int:
-        """Get the amount of RAM available for the cache to use, while keeping memory pressure under control."""
+    def _calc_ram_available_to_model_cache(self) -> int:
+        """Calculate the amount of RAM available for the cache to use."""
         # If self._max_ram_cache_size_gb is set, then it overrides the default logic.
         if self._max_ram_cache_size_gb is not None:
-            ram_total_available_to_cache = int(self._max_ram_cache_size_gb * GB)
-            return ram_total_available_to_cache - self._get_ram_in_use()
+            self._logger.info(f"Using user-defined RAM cache size: {self._max_ram_cache_size_gb} GB.")
+            return int(self._max_ram_cache_size_gb * GB)
 
-        virtual_memory = psutil.virtual_memory()
-        ram_total = virtual_memory.total
-        ram_available = virtual_memory.available
-        ram_used = ram_total - ram_available
+        # Heuristics for dynamically calculating the RAM cache size, **in order of increasing priority**:
+        # 1. As an initial default, use 50% of the total RAM for InvokeAI.
+        #   - Assume a 2GB baseline for InvokeAI's non-model RAM usage, and use the rest of the RAM for the model cache.
+        # 2. On a system with a lot of RAM, users probably don't want InvokeAI to eat up too much RAM.
+        #    There are diminishing returns to storing more and more models. So, we apply an upper bound. (Keep in mind
+        #    that most OSes have some amount of disk caching, which we still benefit from if there is excess memory,
+        #    even if we drop models from the cache.)
+        #    - On systems without a CUDA device, the upper bound is 32GB.
+        #    - On systems with a CUDA device, the upper bound is 1x the amount of VRAM (less the working memory).
+        # 3. Absolute minimum of 4GB.
 
-        # The total size of all the models in the cache will often be larger than the amount of RAM reported by psutil
-        # (due to lazy-loading and OS RAM caching behaviour). We could just rely on the psutil values, but it feels
-        # like a bad idea to over-fill the model cache. So, for now, we'll try to keep the total size of models in the
-        # cache under the total amount of system RAM.
-        cache_ram_used = self._get_ram_in_use()
-        ram_used = max(cache_ram_used, ram_used)
+        # NOTE(ryand): We explored dynamically adjusting the RAM cache size based on memory pressure (using psutil), but
+        # decided against it for now, for the following reasons:
+        # - It was surprisingly difficult to get memory metrics with consistent definitions across OSes. (If you go
+        #   down this path again, don't underestimate the amount of complexity here and be sure to test rigorously on all
+        #   OSes.)
+        # - Making the RAM cache size dynamic opens the door for performance regressions that are hard to diagnose and
+        #   hard for users to understand. It is better for users to see that their RAM is maxed out, and then override
+        #   the default value if desired.
 
-        # Aim to keep 10% of RAM free.
-        ram_available_based_on_memory_usage = int(ram_total * 0.9) - ram_used
+        # Lookup the total VRAM size for the CUDA execution device.
+        total_cuda_vram_bytes: int | None = None
+        if self._execution_device.type == "cuda":
+            _, total_cuda_vram_bytes = torch.cuda.mem_get_info(self._execution_device)
 
-        # If we are running out of RAM, then there's an increased likelihood that we will run into this issue:
-        # https://github.com/invoke-ai/InvokeAI/issues/7513
-        # To keep things running smoothly, there's a minimum RAM cache size that we always allow (even if this means
-        # using swap).
-        min_ram_cache_size_bytes = 4 * GB
-        ram_available_based_on_min_cache_size = min_ram_cache_size_bytes - cache_ram_used
+        # Apply heuristic 1.
+        # ------------------
+        heuristics_applied = [1]
+        total_system_ram_bytes = psutil.virtual_memory().total
+        # Assumed baseline RAM used by InvokeAI for non-model stuff.
+        baseline_ram_used_by_invokeai = 2 * GB
+        ram_available_to_model_cache = int(total_system_ram_bytes * 0.5 - baseline_ram_used_by_invokeai)
 
-        return max(ram_available_based_on_memory_usage, ram_available_based_on_min_cache_size)
+        # Apply heuristic 2.
+        # ------------------
+        max_ram_cache_size_bytes = 32 * GB
+        if total_cuda_vram_bytes is not None:
+            if self._max_vram_cache_size_gb is not None:
+                max_ram_cache_size_bytes = int(self._max_vram_cache_size_gb * GB)
+            else:
+                max_ram_cache_size_bytes = total_cuda_vram_bytes - int(self._execution_device_working_mem_gb * GB)
+        if ram_available_to_model_cache > max_ram_cache_size_bytes:
+            heuristics_applied.append(2)
+            ram_available_to_model_cache = max_ram_cache_size_bytes
+
+        # Apply heuristic 3.
+        # ------------------
+        if ram_available_to_model_cache < 4 * GB:
+            heuristics_applied.append(3)
+            ram_available_to_model_cache = 4 * GB
+
+        self._logger.info(
+            f"Calculated model RAM cache size: {ram_available_to_model_cache / MB:.2f} MB. Heuristics applied: {heuristics_applied}."
+        )
+        return ram_available_to_model_cache
 
     def _get_ram_in_use(self) -> int:
         """Get the amount of RAM currently in use."""
         return sum(ce.cached_model.total_bytes() for ce in self._cached_models.values())
 
+    def _get_ram_available(self) -> int:
+        """Get the amount of RAM available for the cache to use."""
+        return self._ram_cache_size_bytes - self._get_ram_in_use()
+
     def _capture_memory_snapshot(self) -> Optional[MemorySnapshot]:
         if self._log_memory_usage:
             return MemorySnapshot.capture()
@@ -532,6 +601,7 @@ class ModelCache:
 
         self._logger.debug(log)
 
+    @synchronized
     def make_room(self, bytes_needed: int) -> None:
         """Make enough room in the cache to accommodate a new model of indicated size.
 

invokeai/backend/model_manager/load/model_cache/torch_module_autocast/custom_modules/custom_linear.py

@@ -7,7 +7,6 @@ from invokeai.backend.model_manager.load.model_cache.torch_module_autocast.custo
     CustomModuleMixin,
 )
 from invokeai.backend.patches.layers.base_layer_patch import BaseLayerPatch
-from invokeai.backend.patches.layers.concatenated_lora_layer import ConcatenatedLoRALayer
 from invokeai.backend.patches.layers.flux_control_lora_layer import FluxControlLoRALayer
 from invokeai.backend.patches.layers.lora_layer import LoRALayer
 
@@ -22,25 +21,6 @@ def linear_lora_forward(input: torch.Tensor, lora_layer: LoRALayer, lora_weight:
     return x
 
 
-def concatenated_lora_forward(
-    input: torch.Tensor, concatenated_lora_layer: ConcatenatedLoRALayer, lora_weight: float
-) -> torch.Tensor:
-    """An optimized implementation of the residual calculation for a sidecar ConcatenatedLoRALayer."""
-    x_chunks: list[torch.Tensor] = []
-    for lora_layer in concatenated_lora_layer.lora_layers:
-        x_chunk = torch.nn.functional.linear(input, lora_layer.down)
-        if lora_layer.mid is not None:
-            x_chunk = torch.nn.functional.linear(x_chunk, lora_layer.mid)
-        x_chunk = torch.nn.functional.linear(x_chunk, lora_layer.up, bias=lora_layer.bias)
-        x_chunk *= lora_weight * lora_layer.scale()
-        x_chunks.append(x_chunk)
-
-    # TODO(ryand): Generalize to support concat_axis != 0.
-    assert concatenated_lora_layer.concat_axis == 0
-    x = torch.cat(x_chunks, dim=-1)
-    return x
-
-
 def autocast_linear_forward_sidecar_patches(
     orig_module: torch.nn.Linear, input: torch.Tensor, patches_and_weights: list[tuple[BaseLayerPatch, float]]
 ) -> torch.Tensor:
@@ -66,8 +46,6 @@ def autocast_linear_forward_sidecar_patches(
             output += linear_lora_forward(orig_input, patch, patch_weight)
         elif isinstance(patch, LoRALayer):
             output += linear_lora_forward(input, patch, patch_weight)
-        elif isinstance(patch, ConcatenatedLoRALayer):
-            output += concatenated_lora_forward(input, patch, patch_weight)
         else:
             unprocessed_patches_and_weights.append((patch, patch_weight))
 

invokeai/backend/model_manager/load/model_cache/torch_module_autocast/custom_modules/custom_module_mixin.py

@@ -3,6 +3,8 @@ import copy
 import torch
 
 from invokeai.backend.patches.layers.base_layer_patch import BaseLayerPatch
+from invokeai.backend.patches.layers.param_shape_utils import get_param_shape
+from invokeai.backend.quantization.gguf.ggml_tensor import GGMLTensor
 
 
 class CustomModuleMixin:
@@ -42,6 +44,20 @@ class CustomModuleMixin:
         device: torch.device | None = None,
     ):
         """Helper function that aggregates the parameters from all patches into a single dict."""
+        # HACK(ryand): If the original parameters are in a quantized format whose weights can't be accessed, we replace
+        # them with dummy tensors on the 'meta' device. This allows patch layers to access the shapes of the original
+        # parameters. But, of course, any sub-layers that need to access the actual values of the parameters will fail.
+        for param_name in orig_params.keys():
+            param = orig_params[param_name]
+            if type(param) is torch.nn.Parameter and type(param.data) is torch.Tensor:
+                pass
+            elif type(param) is GGMLTensor:
+                # Move to device and dequantize here. Doing it in the patch layer can result in redundant casts /
+                # dequantizations.
+                orig_params[param_name] = param.to(device=device).get_dequantized_tensor()
+            else:
+                orig_params[param_name] = torch.empty(get_param_shape(param), device="meta")
+
         params: dict[str, torch.Tensor] = {}
 
         for patch, patch_weight in patches_and_weights:

invokeai/backend/model_manager/load/model_loaders/flux.py

@@ -80,19 +80,19 @@ class FluxVAELoader(ModelLoader):
             raise ValueError("Only VAECheckpointConfig models are currently supported here.")
         model_path = Path(config.path)
 
-        with SilenceWarnings():
+        with accelerate.init_empty_weights():
             model = AutoEncoder(ae_params[config.config_path])
-            sd = load_file(model_path)
-            model.load_state_dict(sd, assign=True)
-            # VAE is broken in float16, which mps defaults to
-            if self._torch_dtype == torch.float16:
-                try:
-                    vae_dtype = torch.tensor([1.0], dtype=torch.bfloat16, device=self._torch_device).dtype
-                except TypeError:
-                    vae_dtype = torch.float32
-            else:
-                vae_dtype = self._torch_dtype
-            model.to(vae_dtype)
+        sd = load_file(model_path)
+        model.load_state_dict(sd, assign=True)
+        # VAE is broken in float16, which mps defaults to
+        if self._torch_dtype == torch.float16:
+            try:
+                vae_dtype = torch.tensor([1.0], dtype=torch.bfloat16, device=self._torch_device).dtype
+            except TypeError:
+                vae_dtype = torch.float32
+        else:
+            vae_dtype = self._torch_dtype
+        model.to(vae_dtype)
 
         return model
 
@@ -183,7 +183,9 @@ class T5EncoderCheckpointModel(ModelLoader):
             case SubModelType.Tokenizer2 | SubModelType.Tokenizer3:
                 return T5Tokenizer.from_pretrained(Path(config.path) / "tokenizer_2", max_length=512)
             case SubModelType.TextEncoder2 | SubModelType.TextEncoder3:
-                return T5EncoderModel.from_pretrained(Path(config.path) / "text_encoder_2", torch_dtype="auto")
+                return T5EncoderModel.from_pretrained(
+                    Path(config.path) / "text_encoder_2", torch_dtype="auto", low_cpu_mem_usage=True
+                )
 
         raise ValueError(
             f"Only Tokenizer and TextEncoder submodels are currently supported. Received: {submodel_type.value if submodel_type else 'None'}"
@@ -217,17 +219,18 @@ class FluxCheckpointModel(ModelLoader):
         assert isinstance(config, MainCheckpointConfig)
         model_path = Path(config.path)
 
-        with SilenceWarnings():
+        with accelerate.init_empty_weights():
             model = Flux(params[config.config_path])
-            sd = load_file(model_path)
-            if "model.diffusion_model.double_blocks.0.img_attn.norm.key_norm.scale" in sd:
-                sd = convert_bundle_to_flux_transformer_checkpoint(sd)
-            new_sd_size = sum([ten.nelement() * torch.bfloat16.itemsize for ten in sd.values()])
-            self._ram_cache.make_room(new_sd_size)
-            for k in sd.keys():
-                # We need to cast to bfloat16 due to it being the only currently supported dtype for inference
-                sd[k] = sd[k].to(torch.bfloat16)
-            model.load_state_dict(sd, assign=True)
+
+        sd = load_file(model_path)
+        if "model.diffusion_model.double_blocks.0.img_attn.norm.key_norm.scale" in sd:
+            sd = convert_bundle_to_flux_transformer_checkpoint(sd)
+        new_sd_size = sum([ten.nelement() * torch.bfloat16.itemsize for ten in sd.values()])
+        self._ram_cache.make_room(new_sd_size)
+        for k in sd.keys():
+            # We need to cast to bfloat16 due to it being the only currently supported dtype for inference
+            sd[k] = sd[k].to(torch.bfloat16)
+        model.load_state_dict(sd, assign=True)
         return model
 
 
@@ -258,11 +261,11 @@ class FluxGGUFCheckpointModel(ModelLoader):
         assert isinstance(config, MainGGUFCheckpointConfig)
         model_path = Path(config.path)
 
-        with SilenceWarnings():
+        with accelerate.init_empty_weights():
             model = Flux(params[config.config_path])
 
-            # HACK(ryand): We shouldn't be hard-coding the compute_dtype here.
-            sd = gguf_sd_loader(model_path, compute_dtype=torch.bfloat16)
+        # HACK(ryand): We shouldn't be hard-coding the compute_dtype here.
+        sd = gguf_sd_loader(model_path, compute_dtype=torch.bfloat16)
 
         # HACK(ryand): There are some broken GGUF models in circulation that have the wrong shape for img_in.weight.
         # We override the shape here to fix the issue.

invokeai/backend/model_manager/load/model_loaders/lora.py

@@ -31,6 +31,10 @@ from invokeai.backend.patches.lora_conversions.flux_kohya_lora_conversion_utils
     is_state_dict_likely_in_flux_kohya_format,
     lora_model_from_flux_kohya_state_dict,
 )
+from invokeai.backend.patches.lora_conversions.flux_onetrainer_lora_conversion_utils import (
+    is_state_dict_likely_in_flux_onetrainer_format,
+    lora_model_from_flux_onetrainer_state_dict,
+)
 from invokeai.backend.patches.lora_conversions.sd_lora_conversion_utils import lora_model_from_sd_state_dict
 from invokeai.backend.patches.lora_conversions.sdxl_lora_conversion_utils import convert_sdxl_keys_to_diffusers_format
 
@@ -84,8 +88,12 @@ class LoRALoader(ModelLoader):
             elif config.format == ModelFormat.LyCORIS:
                 if is_state_dict_likely_in_flux_kohya_format(state_dict=state_dict):
                     model = lora_model_from_flux_kohya_state_dict(state_dict=state_dict)
+                elif is_state_dict_likely_in_flux_onetrainer_format(state_dict=state_dict):
+                    model = lora_model_from_flux_onetrainer_state_dict(state_dict=state_dict)
                 elif is_state_dict_likely_flux_control(state_dict=state_dict):
                     model = lora_model_from_flux_control_state_dict(state_dict=state_dict)
+                else:
+                    raise ValueError(f"LoRA model is in unsupported FLUX format: {config.format}")
             else:
                 raise ValueError(f"LoRA model is in unsupported FLUX format: {config.format}")
         elif self._model_base in [BaseModelType.StableDiffusion1, BaseModelType.StableDiffusion2]:

invokeai/backend/model_manager/probe.py

@@ -46,6 +46,9 @@ from invokeai.backend.patches.lora_conversions.flux_diffusers_lora_conversion_ut
 from invokeai.backend.patches.lora_conversions.flux_kohya_lora_conversion_utils import (
     is_state_dict_likely_in_flux_kohya_format,
 )
+from invokeai.backend.patches.lora_conversions.flux_onetrainer_lora_conversion_utils import (
+    is_state_dict_likely_in_flux_onetrainer_format,
+)
 from invokeai.backend.quantization.gguf.ggml_tensor import GGMLTensor
 from invokeai.backend.quantization.gguf.loaders import gguf_sd_loader
 from invokeai.backend.spandrel_image_to_image_model import SpandrelImageToImageModel
@@ -283,7 +286,7 @@ class ModelProbe(object):
                 return ModelType.Main
             elif key.startswith(("encoder.conv_in", "decoder.conv_in")):
                 return ModelType.VAE
-            elif key.startswith(("lora_te_", "lora_unet_")):
+            elif key.startswith(("lora_te_", "lora_unet_", "lora_te1_", "lora_te2_", "lora_transformer_")):
                 return ModelType.LoRA
             # "lora_A.weight" and "lora_B.weight" are associated with models in PEFT format. We don't support all PEFT
             # LoRA models, but as of the time of writing, we support Diffusers FLUX PEFT LoRA models.
@@ -632,6 +635,7 @@ class LoRACheckpointProbe(CheckpointProbeBase):
     def get_base_type(self) -> BaseModelType:
         if (
             is_state_dict_likely_in_flux_kohya_format(self.checkpoint)
+            or is_state_dict_likely_in_flux_onetrainer_format(self.checkpoint)
             or is_state_dict_likely_in_flux_diffusers_format(self.checkpoint)
             or is_state_dict_likely_flux_control(self.checkpoint)
         ):

invokeai/backend/patches/layers/concatenated_lora_layer.py

@@ -1,55 +0,0 @@
-from typing import Optional, Sequence
-
-import torch
-
-from invokeai.backend.patches.layers.lora_layer import LoRALayer
-from invokeai.backend.patches.layers.lora_layer_base import LoRALayerBase
-
-
-class ConcatenatedLoRALayer(LoRALayerBase):
-    """A LoRA layer that is composed of multiple LoRA layers concatenated along a specified axis.
-
-    This class was created to handle a special case with FLUX LoRA models. In the BFL FLUX model format, the attention
-    Q, K, V matrices are concatenated along the first dimension. In the diffusers LoRA format, the Q, K, V matrices are
-    stored as separate tensors. This class enables diffusers LoRA layers to be used in BFL FLUX models.
-    """
-
-    def __init__(self, lora_layers: Sequence[LoRALayer], concat_axis: int = 0):
-        super().__init__(alpha=None, bias=None)
-
-        self.lora_layers = lora_layers
-        self.concat_axis = concat_axis
-
-    def _rank(self) -> int | None:
-        return None
-
-    def get_weight(self, orig_weight: torch.Tensor) -> torch.Tensor:
-        # TODO(ryand): Currently, we pass orig_weight=None to the sub-layers. If we want to support sub-layers that
-        # require this value, we will need to implement chunking of the original weight tensor here.
-        # Note that we must apply the sub-layer scales here.
-        layer_weights = [lora_layer.get_weight(None) * lora_layer.scale() for lora_layer in self.lora_layers]  # pyright: ignore[reportArgumentType]
-        return torch.cat(layer_weights, dim=self.concat_axis)
-
-    def get_bias(self, orig_bias: torch.Tensor | None) -> Optional[torch.Tensor]:
-        # TODO(ryand): Currently, we pass orig_bias=None to the sub-layers. If we want to support sub-layers that
-        # require this value, we will need to implement chunking of the original bias tensor here.
-        # Note that we must apply the sub-layer scales here.
-        layer_biases: list[torch.Tensor] = []
-        for lora_layer in self.lora_layers:
-            layer_bias = lora_layer.get_bias(None)
-            if layer_bias is not None:
-                layer_biases.append(layer_bias * lora_layer.scale())
-
-        if len(layer_biases) == 0:
-            return None
-
-        assert len(layer_biases) == len(self.lora_layers)
-        return torch.cat(layer_biases, dim=self.concat_axis)
-
-    def to(self, device: torch.device | None = None, dtype: torch.dtype | None = None):
-        super().to(device=device, dtype=dtype)
-        for lora_layer in self.lora_layers:
-            lora_layer.to(device=device, dtype=dtype)
-
-    def calc_size(self) -> int:
-        return super().calc_size() + sum(lora_layer.calc_size() for lora_layer in self.lora_layers)

invokeai/backend/patches/layers/dora_layer.py

@@ -0,0 +1,115 @@
+from typing import Dict, Optional
+
+import torch
+
+from invokeai.backend.model_manager.load.model_cache.torch_module_autocast.cast_to_device import cast_to_device
+from invokeai.backend.patches.layers.lora_layer_base import LoRALayerBase
+from invokeai.backend.util.calc_tensor_size import calc_tensors_size
+
+
+class DoRALayer(LoRALayerBase):
+    """A DoRA layer. As defined in https://arxiv.org/pdf/2402.09353."""
+
+    def __init__(
+        self,
+        up: torch.Tensor,
+        down: torch.Tensor,
+        dora_scale: torch.Tensor,
+        alpha: float | None,
+        bias: Optional[torch.Tensor],
+    ):
+        super().__init__(alpha, bias)
+        self.up = up
+        self.down = down
+        self.dora_scale = dora_scale
+
+    @classmethod
+    def from_state_dict_values(cls, values: Dict[str, torch.Tensor]):
+        alpha = cls._parse_alpha(values.get("alpha", None))
+        bias = cls._parse_bias(
+            values.get("bias_indices", None), values.get("bias_values", None), values.get("bias_size", None)
+        )
+
+        layer = cls(
+            up=values["lora_up.weight"],
+            down=values["lora_down.weight"],
+            dora_scale=values["dora_scale"],
+            alpha=alpha,
+            bias=bias,
+        )
+
+        cls.warn_on_unhandled_keys(
+            values=values,
+            handled_keys={
+                # Default keys.
+                "alpha",
+                "bias_indices",
+                "bias_values",
+                "bias_size",
+                # Layer-specific keys.
+                "lora_up.weight",
+                "lora_down.weight",
+                "dora_scale",
+            },
+        )
+
+        return layer
+
+    def _rank(self) -> int:
+        return self.down.shape[0]
+
+    def get_weight(self, orig_weight: torch.Tensor) -> torch.Tensor:
+        orig_weight = cast_to_device(orig_weight, self.up.device)
+
+        # Note: Variable names (e.g. delta_v) are based on the paper.
+        delta_v = self.up.reshape(self.up.shape[0], -1) @ self.down.reshape(self.down.shape[0], -1)
+        delta_v = delta_v.reshape(orig_weight.shape)
+
+        delta_v = delta_v * self.scale()
+
+        # At this point, out_weight is the unnormalized direction matrix.
+        out_weight = orig_weight + delta_v
+
+        # TODO(ryand): Simplify this logic.
+        direction_norm = (
+            out_weight.transpose(0, 1)
+            .reshape(out_weight.shape[1], -1)
+            .norm(dim=1, keepdim=True)
+            .reshape(out_weight.shape[1], *[1] * (out_weight.dim() - 1))
+            .transpose(0, 1)
+        )
+
+        out_weight *= self.dora_scale / direction_norm
+
+        return out_weight - orig_weight
+
+    def to(self, device: torch.device | None = None, dtype: torch.dtype | None = None):
+        super().to(device=device, dtype=dtype)
+        self.up = self.up.to(device=device, dtype=dtype)
+        self.down = self.down.to(device=device, dtype=dtype)
+        self.dora_scale = self.dora_scale.to(device=device, dtype=dtype)
+
+    def calc_size(self) -> int:
+        return super().calc_size() + calc_tensors_size([self.up, self.down, self.dora_scale])
+
+    def get_parameters(self, orig_parameters: dict[str, torch.Tensor], weight: float) -> dict[str, torch.Tensor]:
+        if any(p.device.type == "meta" for p in orig_parameters.values()):
+            # If any of the original parameters are on the 'meta' device, we assume this is because the base model is in
+            # a quantization format that doesn't allow easy dequantization.
+            raise RuntimeError(
+                "The base model quantization format (likely bitsandbytes) is not compatible with DoRA patches."
+            )
+
+        scale = self.scale()
+        params = {"weight": self.get_weight(orig_parameters["weight"]) * weight}
+        bias = self.get_bias(orig_parameters.get("bias", None))
+        if bias is not None:
+            params["bias"] = bias * (weight * scale)
+
+        # Reshape all params to match the original module's shape.
+        for param_name, param_weight in params.items():
+            orig_param = orig_parameters[param_name]
+            if param_weight.shape != orig_param.shape:
+                params[param_name] = param_weight.reshape(orig_param.shape)
+
+        return params

invokeai/backend/patches/layers/lora_layer_base.py

@@ -4,6 +4,7 @@ import torch
 
 import invokeai.backend.util.logging as logger
 from invokeai.backend.patches.layers.base_layer_patch import BaseLayerPatch
+from invokeai.backend.patches.layers.param_shape_utils import get_param_shape
 from invokeai.backend.util.calc_tensor_size import calc_tensors_size
 
 
@@ -67,8 +68,8 @@ class LoRALayerBase(BaseLayerPatch):
         # Reshape all params to match the original module's shape.
         for param_name, param_weight in params.items():
             orig_param = orig_parameters[param_name]
-            if param_weight.shape != orig_param.shape:
-                params[param_name] = param_weight.reshape(orig_param.shape)
+            if param_weight.shape != get_param_shape(orig_param):
+                params[param_name] = param_weight.reshape(get_param_shape(orig_param))
 
         return params
 

invokeai/backend/patches/layers/merged_layer_patch.py

@@ -0,0 +1,65 @@
+from dataclasses import dataclass
+from typing import Sequence
+
+import torch
+
+from invokeai.backend.patches.layers.base_layer_patch import BaseLayerPatch
+from invokeai.backend.patches.layers.param_shape_utils import get_param_shape
+
+
+@dataclass
+class Range:
+    start: int
+    end: int
+
+
+class MergedLayerPatch(BaseLayerPatch):
+    """A patch layer that is composed of multiple sub-layers merged together.
+
+    This class was created to handle a special case with FLUX LoRA models. In the BFL FLUX model format, the attention
+    Q, K, V matrices are concatenated along the first dimension. In the diffusers LoRA format, the Q, K, V matrices are
+    stored as separate tensors. This class enables diffusers LoRA layers to be used in BFL FLUX models.
+    """
+
+    def __init__(
+        self,
+        lora_layers: Sequence[BaseLayerPatch],
+        ranges: Sequence[Range],
+    ):
+        super().__init__()
+
+        self.lora_layers = lora_layers
+        # self.ranges[i] is the range for the i'th lora layer along the 0'th weight dimension.
+        self.ranges = ranges
+        assert len(self.ranges) == len(self.lora_layers)
+
+    def get_parameters(self, orig_parameters: dict[str, torch.Tensor], weight: float) -> dict[str, torch.Tensor]:
+        out_parameters: dict[str, torch.Tensor] = {}
+
+        for lora_layer, range in zip(self.lora_layers, self.ranges, strict=True):
+            sliced_parameters: dict[str, torch.Tensor] = {
+                n: p[range.start : range.end] for n, p in orig_parameters.items()
+            }
+
+            # Note that `weight` is applied in the sub-layers, no need to apply it in this function.
+            layer_out_parameters = lora_layer.get_parameters(sliced_parameters, weight)
+
+            for out_param_name, out_param in layer_out_parameters.items():
+                if out_param_name not in out_parameters:
+                    # If not already in the output dict, initialize an output tensor with the same shape as the full
+                    # original parameter.
+                    out_parameters[out_param_name] = torch.zeros(
+                        get_param_shape(orig_parameters[out_param_name]),
+                        dtype=out_param.dtype,
+                        device=out_param.device,
+                    )
+                out_parameters[out_param_name][range.start : range.end] += out_param
+
+        return out_parameters
+
+    def to(self, device: torch.device | None = None, dtype: torch.dtype | None = None):
+        for lora_layer in self.lora_layers:
+            lora_layer.to(device=device, dtype=dtype)
+
+    def calc_size(self) -> int:
+        return sum(lora_layer.calc_size() for lora_layer in self.lora_layers)

invokeai/backend/patches/layers/param_shape_utils.py

@@ -0,0 +1,19 @@
+import torch
+
+try:
+    from bitsandbytes.nn.modules import Params4bit
+
+    bnb_available: bool = True
+except ImportError:
+    bnb_available: bool = False
+
+
+def get_param_shape(param: torch.Tensor) -> torch.Size:
+    """A helper function to get the shape of a parameter that handles `bitsandbytes.nn.Params4Bit` correctly."""
+    # Accessing the `.shape` attribute of `bitsandbytes.nn.Params4Bit` will return an incorrect result. Instead, we must
+    # access the `.quant_state.shape` attribute.
+    if bnb_available and type(param) is Params4bit:  # type: ignore
+        quant_state = param.quant_state
+        if quant_state is not None:
+            return quant_state.shape
+    return param.shape

invokeai/backend/patches/layers/utils.py

@@ -3,6 +3,7 @@ from typing import Dict
 import torch
 
 from invokeai.backend.patches.layers.base_layer_patch import BaseLayerPatch
+from invokeai.backend.patches.layers.dora_layer import DoRALayer
 from invokeai.backend.patches.layers.full_layer import FullLayer
 from invokeai.backend.patches.layers.ia3_layer import IA3Layer
 from invokeai.backend.patches.layers.loha_layer import LoHALayer
@@ -14,8 +15,9 @@ from invokeai.backend.patches.layers.norm_layer import NormLayer
 def any_lora_layer_from_state_dict(state_dict: Dict[str, torch.Tensor]) -> BaseLayerPatch:
     # Detect layers according to LyCORIS detection logic(`weight_list_det`)
     # https://github.com/KohakuBlueleaf/LyCORIS/tree/8ad8000efb79e2b879054da8c9356e6143591bad/lycoris/modules
-
-    if "lora_up.weight" in state_dict:
+    if "dora_scale" in state_dict:
+        return DoRALayer.from_state_dict_values(state_dict)
+    elif "lora_up.weight" in state_dict:
         # LoRA a.k.a LoCon
         return LoRALayer.from_state_dict_values(state_dict)
     elif "hada_w1_a" in state_dict:

invokeai/backend/patches/lora_conversions/flux_diffusers_lora_conversion_utils.py

@@ -3,8 +3,8 @@ from typing import Dict
 import torch
 
 from invokeai.backend.patches.layers.base_layer_patch import BaseLayerPatch
-from invokeai.backend.patches.layers.concatenated_lora_layer import ConcatenatedLoRALayer
-from invokeai.backend.patches.layers.lora_layer import LoRALayer
+from invokeai.backend.patches.layers.merged_layer_patch import MergedLayerPatch, Range
+from invokeai.backend.patches.layers.utils import any_lora_layer_from_state_dict
 from invokeai.backend.patches.lora_conversions.flux_lora_constants import FLUX_LORA_TRANSFORMER_PREFIX
 from invokeai.backend.patches.model_patch_raw import ModelPatchRaw
 
@@ -33,13 +33,21 @@ def is_state_dict_likely_in_flux_diffusers_format(state_dict: Dict[str, torch.Te
 def lora_model_from_flux_diffusers_state_dict(
     state_dict: Dict[str, torch.Tensor], alpha: float | None
 ) -> ModelPatchRaw:
-    """Loads a state dict in the Diffusers FLUX LoRA format into a LoRAModelRaw object.
+    # Group keys by layer.
+    grouped_state_dict: dict[str, dict[str, torch.Tensor]] = _group_by_layer(state_dict)
+    layers = lora_layers_from_flux_diffusers_grouped_state_dict(grouped_state_dict, alpha)
+    return ModelPatchRaw(layers=layers)
+
+
+def lora_layers_from_flux_diffusers_grouped_state_dict(
+    grouped_state_dict: Dict[str, Dict[str, torch.Tensor]], alpha: float | None
+) -> dict[str, BaseLayerPatch]:
+    """Converts a grouped state dict with Diffusers FLUX LoRA keys to LoRA layers with BFL keys (i.e. the module key
+    format used by Invoke).
 
     This function is based on:
     https://github.com/huggingface/diffusers/blob/55ac421f7bb12fd00ccbef727be4dc2f3f920abb/scripts/convert_flux_to_diffusers.py
     """
-    # Group keys by layer.
-    grouped_state_dict: dict[str, dict[str, torch.Tensor]] = _group_by_layer(state_dict)
 
     # Remove the "transformer." prefix from all keys.
     grouped_state_dict = {k.replace("transformer.", ""): v for k, v in grouped_state_dict.items()}
@@ -53,17 +61,26 @@ def lora_model_from_flux_diffusers_state_dict(
 
     layers: dict[str, BaseLayerPatch] = {}
 
-    def add_lora_layer_if_present(src_key: str, dst_key: str) -> None:
-        if src_key in grouped_state_dict:
-            src_layer_dict = grouped_state_dict.pop(src_key)
-            value = {
+    def get_lora_layer_values(src_layer_dict: dict[str, torch.Tensor]) -> dict[str, torch.Tensor]:
+        if "lora_A.weight" in src_layer_dict:
+            # The LoRA keys are in PEFT format.
+            values = {
                 "lora_down.weight": src_layer_dict.pop("lora_A.weight"),
                 "lora_up.weight": src_layer_dict.pop("lora_B.weight"),
             }
             if alpha is not None:
-                value["alpha"] = torch.tensor(alpha)
-            layers[dst_key] = LoRALayer.from_state_dict_values(values=value)
+                values["alpha"] = torch.tensor(alpha)
             assert len(src_layer_dict) == 0
+            return values
+        else:
+            # Assume that the LoRA keys are in Kohya format.
+            return src_layer_dict
+
+    def add_lora_layer_if_present(src_key: str, dst_key: str) -> None:
+        if src_key in grouped_state_dict:
+            src_layer_dict = grouped_state_dict.pop(src_key)
+            values = get_lora_layer_values(src_layer_dict)
+            layers[dst_key] = any_lora_layer_from_state_dict(values)
 
     def add_qkv_lora_layer_if_present(
         src_keys: list[str],
@@ -79,29 +96,24 @@ def lora_model_from_flux_diffusers_state_dict(
         if not any(keys_present):
             return
 
-        sub_layers: list[LoRALayer] = []
+        dim_0_offset = 0
+        sub_layers: list[BaseLayerPatch] = []
+        sub_layer_ranges: list[Range] = []
         for src_key, src_weight_shape in zip(src_keys, src_weight_shapes, strict=True):
             src_layer_dict = grouped_state_dict.pop(src_key, None)
             if src_layer_dict is not None:
-                values = {
-                    "lora_down.weight": src_layer_dict.pop("lora_A.weight"),
-                    "lora_up.weight": src_layer_dict.pop("lora_B.weight"),
-                }
-                if alpha is not None:
-                    values["alpha"] = torch.tensor(alpha)
-                assert values["lora_down.weight"].shape[1] == src_weight_shape[1]
-                assert values["lora_up.weight"].shape[0] == src_weight_shape[0]
-                sub_layers.append(LoRALayer.from_state_dict_values(values=values))
-                assert len(src_layer_dict) == 0
+                values = get_lora_layer_values(src_layer_dict)
+                # assert values["lora_down.weight"].shape[1] == src_weight_shape[1]
+                # assert values["lora_up.weight"].shape[0] == src_weight_shape[0]
+                sub_layers.append(any_lora_layer_from_state_dict(values))
+                sub_layer_ranges.append(Range(dim_0_offset, dim_0_offset + src_weight_shape[0]))
             else:
                 if not allow_missing_keys:
                     raise ValueError(f"Missing LoRA layer: '{src_key}'.")
-                values = {
-                    "lora_up.weight": torch.zeros((src_weight_shape[0], 1)),
-                    "lora_down.weight": torch.zeros((1, src_weight_shape[1])),
-                }
-                sub_layers.append(LoRALayer.from_state_dict_values(values=values))
-        layers[dst_qkv_key] = ConcatenatedLoRALayer(lora_layers=sub_layers)
+
+            dim_0_offset += src_weight_shape[0]
+
+        layers[dst_qkv_key] = MergedLayerPatch(sub_layers, sub_layer_ranges)
 
     # time_text_embed.timestep_embedder -> time_in.
     add_lora_layer_if_present("time_text_embed.timestep_embedder.linear_1", "time_in.in_layer")
@@ -217,7 +229,7 @@ def lora_model_from_flux_diffusers_state_dict(
 
     layers_with_prefix = {f"{FLUX_LORA_TRANSFORMER_PREFIX}{k}": v for k, v in layers.items()}
 
-    return ModelPatchRaw(layers=layers_with_prefix)
+    return layers_with_prefix
 
 
 def _group_by_layer(state_dict: Dict[str, torch.Tensor]) -> dict[str, dict[str, torch.Tensor]]:

invokeai/backend/patches/lora_conversions/flux_kohya_lora_conversion_utils.py

@@ -7,6 +7,7 @@ from invokeai.backend.patches.layers.base_layer_patch import BaseLayerPatch
 from invokeai.backend.patches.layers.utils import any_lora_layer_from_state_dict
 from invokeai.backend.patches.lora_conversions.flux_lora_constants import (
     FLUX_LORA_CLIP_PREFIX,
+    FLUX_LORA_T5_PREFIX,
     FLUX_LORA_TRANSFORMER_PREFIX,
 )
 from invokeai.backend.patches.model_patch_raw import ModelPatchRaw
@@ -26,6 +27,14 @@ FLUX_KOHYA_TRANSFORMER_KEY_REGEX = (
 #   lora_te1_text_model_encoder_layers_0_mlp_fc1.lora_up.weight
 FLUX_KOHYA_CLIP_KEY_REGEX = r"lora_te1_text_model_encoder_layers_(\d+)_(mlp|self_attn)_(\w+)\.?.*"
 
+# A regex pattern that matches all of the T5 keys in the Kohya FLUX LoRA format.
+# Example keys:
+#   lora_te2_encoder_block_0_layer_0_SelfAttention_k.alpha
+#   lora_te2_encoder_block_0_layer_0_SelfAttention_k.dora_scale
+#   lora_te2_encoder_block_0_layer_0_SelfAttention_k.lora_down.weight
+#   lora_te2_encoder_block_0_layer_0_SelfAttention_k.lora_up.weight
+FLUX_KOHYA_T5_KEY_REGEX = r"lora_te2_encoder_block_(\d+)_layer_(\d+)_(DenseReluDense|SelfAttention)_(\w+)_?(\w+)?\.?.*"
+
 
 def is_state_dict_likely_in_flux_kohya_format(state_dict: Dict[str, Any]) -> bool:
     """Checks if the provided state dict is likely in the Kohya FLUX LoRA format.
@@ -34,7 +43,9 @@ def is_state_dict_likely_in_flux_kohya_format(state_dict: Dict[str, Any]) -> boo
     perfect-precision detector would require checking all keys against a whitelist and verifying tensor shapes.)
     """
     return all(
-        re.match(FLUX_KOHYA_TRANSFORMER_KEY_REGEX, k) or re.match(FLUX_KOHYA_CLIP_KEY_REGEX, k)
+        re.match(FLUX_KOHYA_TRANSFORMER_KEY_REGEX, k)
+        or re.match(FLUX_KOHYA_CLIP_KEY_REGEX, k)
+        or re.match(FLUX_KOHYA_T5_KEY_REGEX, k)
         for k in state_dict.keys()
     )
 
@@ -48,27 +59,34 @@ def lora_model_from_flux_kohya_state_dict(state_dict: Dict[str, torch.Tensor]) -
             grouped_state_dict[layer_name] = {}
         grouped_state_dict[layer_name][param_name] = value
 
-    # Split the grouped state dict into transformer and CLIP state dicts.
+    # Split the grouped state dict into transformer, CLIP, and T5 state dicts.
     transformer_grouped_sd: dict[str, dict[str, torch.Tensor]] = {}
     clip_grouped_sd: dict[str, dict[str, torch.Tensor]] = {}
+    t5_grouped_sd: dict[str, dict[str, torch.Tensor]] = {}
     for layer_name, layer_state_dict in grouped_state_dict.items():
         if layer_name.startswith("lora_unet"):
             transformer_grouped_sd[layer_name] = layer_state_dict
         elif layer_name.startswith("lora_te1"):
             clip_grouped_sd[layer_name] = layer_state_dict
+        elif layer_name.startswith("lora_te2"):
+            t5_grouped_sd[layer_name] = layer_state_dict
         else:
             raise ValueError(f"Layer '{layer_name}' does not match the expected pattern for FLUX LoRA weights.")
 
     # Convert the state dicts to the InvokeAI format.
     transformer_grouped_sd = _convert_flux_transformer_kohya_state_dict_to_invoke_format(transformer_grouped_sd)
     clip_grouped_sd = _convert_flux_clip_kohya_state_dict_to_invoke_format(clip_grouped_sd)
+    t5_grouped_sd = _convert_flux_t5_kohya_state_dict_to_invoke_format(t5_grouped_sd)
 
     # Create LoRA layers.
     layers: dict[str, BaseLayerPatch] = {}
-    for layer_key, layer_state_dict in transformer_grouped_sd.items():
-        layers[FLUX_LORA_TRANSFORMER_PREFIX + layer_key] = any_lora_layer_from_state_dict(layer_state_dict)
-    for layer_key, layer_state_dict in clip_grouped_sd.items():
-        layers[FLUX_LORA_CLIP_PREFIX + layer_key] = any_lora_layer_from_state_dict(layer_state_dict)
+    for model_prefix, grouped_sd in [
+        (FLUX_LORA_TRANSFORMER_PREFIX, transformer_grouped_sd),
+        (FLUX_LORA_CLIP_PREFIX, clip_grouped_sd),
+        (FLUX_LORA_T5_PREFIX, t5_grouped_sd),
+    ]:
+        for layer_key, layer_state_dict in grouped_sd.items():
+            layers[model_prefix + layer_key] = any_lora_layer_from_state_dict(layer_state_dict)
 
     # Create and return the LoRAModelRaw.
     return ModelPatchRaw(layers=layers)
@@ -123,3 +141,31 @@ def _convert_flux_transformer_kohya_state_dict_to_invoke_format(state_dict: Dict
             raise ValueError(f"Key '{k}' does not match the expected pattern for FLUX LoRA weights.")
 
     return converted_dict
+
+
+def _convert_flux_t5_kohya_state_dict_to_invoke_format(state_dict: Dict[str, T]) -> Dict[str, T]:
+    """Converts a T5 LoRA state dict from the Kohya FLUX LoRA format to LoRA weight format used internally by
+    InvokeAI.
+
+    Example key conversions:
+
+    "lora_te2_encoder_block_0_layer_0_SelfAttention_k" -> "encoder.block.0.layer.0.SelfAttention.k"
+    "lora_te2_encoder_block_0_layer_1_DenseReluDense_wi_0" -> "encoder.block.0.layer.1.DenseReluDense.wi.0"
+    """
+
+    def replace_func(match: re.Match[str]) -> str:
+        s = f"encoder.block.{match.group(1)}.layer.{match.group(2)}.{match.group(3)}.{match.group(4)}"
+        if match.group(5):
+            s += f".{match.group(5)}"
+        return s
+
+    converted_dict: dict[str, T] = {}
+    for k, v in state_dict.items():
+        match = re.match(FLUX_KOHYA_T5_KEY_REGEX, k)
+        if match:
+            new_key = re.sub(FLUX_KOHYA_T5_KEY_REGEX, replace_func, k)
+            converted_dict[new_key] = v
+        else:
+            raise ValueError(f"Key '{k}' does not match the expected pattern for FLUX LoRA weights.")
+
+    return converted_dict

invokeai/backend/patches/lora_conversions/flux_lora_constants.py

@@ -1,3 +1,4 @@
 # Prefixes used to distinguish between transformer and CLIP text encoder keys in the FLUX InvokeAI LoRA format.
 FLUX_LORA_TRANSFORMER_PREFIX = "lora_transformer-"
 FLUX_LORA_CLIP_PREFIX = "lora_clip-"
+FLUX_LORA_T5_PREFIX = "lora_t5-"

invokeai/backend/patches/lora_conversions/flux_onetrainer_lora_conversion_utils.py

@@ -0,0 +1,163 @@
+import re
+from typing import Any, Dict
+
+import torch
+
+from invokeai.backend.patches.layers.base_layer_patch import BaseLayerPatch
+from invokeai.backend.patches.layers.utils import any_lora_layer_from_state_dict
+from invokeai.backend.patches.lora_conversions.flux_diffusers_lora_conversion_utils import (
+    lora_layers_from_flux_diffusers_grouped_state_dict,
+)
+from invokeai.backend.patches.lora_conversions.flux_kohya_lora_conversion_utils import (
+    FLUX_KOHYA_CLIP_KEY_REGEX,
+    FLUX_KOHYA_T5_KEY_REGEX,
+    _convert_flux_clip_kohya_state_dict_to_invoke_format,
+    _convert_flux_t5_kohya_state_dict_to_invoke_format,
+)
+from invokeai.backend.patches.lora_conversions.flux_lora_constants import (
+    FLUX_LORA_CLIP_PREFIX,
+    FLUX_LORA_T5_PREFIX,
+)
+from invokeai.backend.patches.lora_conversions.kohya_key_utils import (
+    INDEX_PLACEHOLDER,
+    ParsingTree,
+    insert_periods_into_kohya_key,
+)
+from invokeai.backend.patches.model_patch_raw import ModelPatchRaw
+
+# A regex pattern that matches all of the transformer keys in the OneTrainer FLUX LoRA format.
+# The OneTrainer format uses a mix of the Kohya and Diffusers formats:
+#   - The base model keys are in Diffusers format.
+#   - Periods are replaced with underscores, to match Kohya.
+#   - The LoRA key suffixes (e.g. .alpha, .lora_down.weight, .lora_up.weight) match Kohya.
+# Example keys:
+# - "lora_transformer_single_transformer_blocks_0_attn_to_k.alpha"
+# - "lora_transformer_single_transformer_blocks_0_attn_to_k.dora_scale"
+# - "lora_transformer_single_transformer_blocks_0_attn_to_k.lora_down.weight"
+# - "lora_transformer_single_transformer_blocks_0_attn_to_k.lora_up.weight"
+FLUX_ONETRAINER_TRANSFORMER_KEY_REGEX = (
+    r"lora_transformer_(single_transformer_blocks|transformer_blocks)_(\d+)_(\w+)\.(.*)"
+)
+
+
+def is_state_dict_likely_in_flux_onetrainer_format(state_dict: Dict[str, Any]) -> bool:
+    """Checks if the provided state dict is likely in the OneTrainer FLUX LoRA format.
+
+    This is intended to be a high-precision detector, but it is not guaranteed to have perfect precision. (A
+    perfect-precision detector would require checking all keys against a whitelist and verifying tensor shapes.)
+
+    Note that OneTrainer matches the Kohya format for the CLIP and T5 models.
+    """
+    return all(
+        re.match(FLUX_ONETRAINER_TRANSFORMER_KEY_REGEX, k)
+        or re.match(FLUX_KOHYA_CLIP_KEY_REGEX, k)
+        or re.match(FLUX_KOHYA_T5_KEY_REGEX, k)
+        for k in state_dict.keys()
+    )
+
+
+def lora_model_from_flux_onetrainer_state_dict(state_dict: Dict[str, torch.Tensor]) -> ModelPatchRaw:  # type: ignore
+    # Group keys by layer.
+    grouped_state_dict: dict[str, dict[str, torch.Tensor]] = {}
+    for key, value in state_dict.items():
+        layer_name, param_name = key.split(".", 1)
+        if layer_name not in grouped_state_dict:
+            grouped_state_dict[layer_name] = {}
+        grouped_state_dict[layer_name][param_name] = value
+
+    # Split the grouped state dict into transformer, CLIP, and T5 state dicts.
+    transformer_grouped_sd: dict[str, dict[str, torch.Tensor]] = {}
+    clip_grouped_sd: dict[str, dict[str, torch.Tensor]] = {}
+    t5_grouped_sd: dict[str, dict[str, torch.Tensor]] = {}
+    for layer_name, layer_state_dict in grouped_state_dict.items():
+        if layer_name.startswith("lora_transformer"):
+            transformer_grouped_sd[layer_name] = layer_state_dict
+        elif layer_name.startswith("lora_te1"):
+            clip_grouped_sd[layer_name] = layer_state_dict
+        elif layer_name.startswith("lora_te2"):
+            t5_grouped_sd[layer_name] = layer_state_dict
+        else:
+            raise ValueError(f"Layer '{layer_name}' does not match the expected pattern for FLUX LoRA weights.")
+
+    # Convert the state dicts to the InvokeAI format.
+    clip_grouped_sd = _convert_flux_clip_kohya_state_dict_to_invoke_format(clip_grouped_sd)
+    t5_grouped_sd = _convert_flux_t5_kohya_state_dict_to_invoke_format(t5_grouped_sd)
+
+    # Create LoRA layers.
+    layers: dict[str, BaseLayerPatch] = {}
+    for model_prefix, grouped_sd in [
+        # (FLUX_LORA_TRANSFORMER_PREFIX, transformer_grouped_sd),
+        (FLUX_LORA_CLIP_PREFIX, clip_grouped_sd),
+        (FLUX_LORA_T5_PREFIX, t5_grouped_sd),
+    ]:
+        for layer_key, layer_state_dict in grouped_sd.items():
+            layers[model_prefix + layer_key] = any_lora_layer_from_state_dict(layer_state_dict)
+
+    # Handle the transformer.
+    transformer_layers = _convert_flux_transformer_onetrainer_state_dict_to_invoke_format(transformer_grouped_sd)
+    layers.update(transformer_layers)
+
+    # Create and return the LoRAModelRaw.
+    return ModelPatchRaw(layers=layers)
+
+
+# This parsing tree was generated by calling `generate_kohya_parsing_tree_from_keys()` on the keys in
+# flux_lora_diffusers_format.py.
+flux_transformer_kohya_parsing_tree: ParsingTree = {
+    "transformer": {
+        "single_transformer_blocks": {
+            INDEX_PLACEHOLDER: {
+                "attn": {"to_k": {}, "to_q": {}, "to_v": {}},
+                "norm": {"linear": {}},
+                "proj_mlp": {},
+                "proj_out": {},
+            }
+        },
+        "transformer_blocks": {
+            INDEX_PLACEHOLDER: {
+                "attn": {
+                    "add_k_proj": {},
+                    "add_q_proj": {},
+                    "add_v_proj": {},
+                    "to_add_out": {},
+                    "to_k": {},
+                    "to_out": {INDEX_PLACEHOLDER: {}},
+                    "to_q": {},
+                    "to_v": {},
+                },
+                "ff": {"net": {INDEX_PLACEHOLDER: {"proj": {}}}},
+                "ff_context": {"net": {INDEX_PLACEHOLDER: {"proj": {}}}},
+                "norm1": {"linear": {}},
+                "norm1_context": {"linear": {}},
+            }
+        },
+    }
+}
+
+
+def _convert_flux_transformer_onetrainer_state_dict_to_invoke_format(
+    state_dict: Dict[str, Dict[str, torch.Tensor]],
+) -> dict[str, BaseLayerPatch]:
+    """Converts a FLUX transformer LoRA state dict from the OneTrainer FLUX LoRA format to the LoRA weight format used
+    internally by InvokeAI.
+    """
+
+    # Step 1: Convert the Kohya-style keys with underscores to classic keys with periods.
+    # Example:
+    # "lora_transformer_single_transformer_blocks_0_attn_to_k.lora_down.weight" -> "transformer.single_transformer_blocks.0.attn.to_k.lora_down.weight"
+    lora_prefix = "lora_"
+    lora_prefix_length = len(lora_prefix)
+    kohya_state_dict: dict[str, Dict[str, torch.Tensor]] = {}
+    for key in state_dict.keys():
+        # Remove the "lora_" prefix.
+        assert key.startswith(lora_prefix)
+        new_key = key[lora_prefix_length:]
+
+        # Add periods to the Kohya-style module keys.
+        new_key = insert_periods_into_kohya_key(new_key, flux_transformer_kohya_parsing_tree)
+
+        # Replace the old key with the new key.
+        kohya_state_dict[new_key] = state_dict[key]
+
+    # Step 2: Convert diffusers module names to the BFL module names.
+    return lora_layers_from_flux_diffusers_grouped_state_dict(kohya_state_dict, alpha=None)

invokeai/backend/patches/lora_conversions/kohya_key_utils.py

@@ -0,0 +1,102 @@
+from typing import Iterable
+
+INDEX_PLACEHOLDER = "index_placeholder"
+
+
+# Type alias for a 'ParsingTree', which is a recursive dict with string keys.
+ParsingTree = dict[str, "ParsingTree"]
+
+
+def insert_periods_into_kohya_key(key: str, parsing_tree: ParsingTree) -> str:
+    """Insert periods into a Kohya key based on a parsing tree.
+
+    Kohya format keys are produced by replacing periods with underscores in the original key.
+
+    Example:
+    ```
+    key = "module_a_module_b_0_attn_to_k"
+    parsing_tree = {
+        "module_a": {
+            "module_b": {
+                INDEX_PLACEHOLDER: {
+                    "attn": {},
+                },
+            },
+        },
+    }
+    result = insert_periods_into_kohya_key(key, parsing_tree)
+    > "module_a.module_b.0.attn.to_k"
+    ```
+    """
+    # Split key into parts by underscore.
+    parts = key.split("_")
+
+    # Build up result by walking through parsing tree and parts.
+    result_parts: list[str] = []
+    current_part = ""
+    current_tree = parsing_tree
+
+    for part in parts:
+        if len(current_part) > 0:
+            current_part = current_part + "_"
+        current_part += part
+
+        if current_part in current_tree:
+            # Match found.
+            current_tree = current_tree[current_part]
+            result_parts.append(current_part)
+            current_part = ""
+        elif current_part.isnumeric() and INDEX_PLACEHOLDER in current_tree:
+            # Match found with index placeholder.
+            current_tree = current_tree[INDEX_PLACEHOLDER]
+            result_parts.append(current_part)
+            current_part = ""
+
+    if len(current_part) > 0:
+        raise ValueError(f"Key {key} does not match parsing tree {parsing_tree}.")
+
+    return ".".join(result_parts)
+
+
+def generate_kohya_parsing_tree_from_keys(keys: Iterable[str]) -> ParsingTree:
+    """Generate a parsing tree from a list of keys.
+
+    Example:
+    ```
+    keys = [
+        "module_a.module_b.0.attn.to_k",
+        "module_a.module_b.1.attn.to_k",
+        "module_a.module_c.proj",
+    ]
+
+    tree = generate_kohya_parsing_tree_from_keys(keys)
+    > {
+    >     "module_a": {
+    >         "module_b": {
+    >             INDEX_PLACEHOLDER: {
+    >                 "attn": {
+    >                     "to_k": {},
+    >                     "to_q": {},
+    >                 },
+    >             }
+    >         },
+    >         "module_c": {
+    >             "proj": {},
+    >         }
+    >     }
+    > }
+    ```
+    """
+    tree: ParsingTree = {}
+    for key in keys:
+        subtree: ParsingTree = tree
+        for module_name in key.split("."):
+            key = module_name
+            if module_name.isnumeric():
+                key = INDEX_PLACEHOLDER
+
+            if key not in subtree:
+                subtree[key] = {}
+
+            subtree = subtree[key]
+    return tree

invokeai/backend/quantization/gguf/ggml_tensor.py

@@ -54,7 +54,9 @@ GGML_TENSOR_OP_TABLE = {
     torch.ops.aten.addmm.default: dequantize_and_run,  # pyright: ignore
     torch.ops.aten.mul.Tensor: dequantize_and_run,  # pyright: ignore
     torch.ops.aten.add.Tensor: dequantize_and_run,  # pyright: ignore
+    torch.ops.aten.sub.Tensor: dequantize_and_run,  # pyright: ignore
     torch.ops.aten.allclose.default: dequantize_and_run,  # pyright: ignore
+    torch.ops.aten.slice.Tensor: dequantize_and_run,  # pyright: ignore
 }
 
 if torch.backends.mps.is_available():

invokeai/frontend/web/.eslintrc.js

@@ -23,6 +23,12 @@ module.exports = {
         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.',
+      },
     ],
   },
   overrides: [

invokeai/frontend/web/index.html

@@ -11,9 +11,11 @@
   <link id="invoke-favicon" rel="icon" type="icon" href="assets/images/invoke-favicon.svg" />
   <style>
     html,
-    body {
+    body,
+    #root {
       padding: 0;
       margin: 0;
+      overflow: hidden;
     }
   </style>
 </head>
@@ -23,4 +25,4 @@
   <script type="module" src="/src/main.tsx"></script>
 </body>
 
-</html>
+</html>

invokeai/frontend/web/package.json

@@ -58,10 +58,11 @@
     "@dagrejs/dagre": "^1.1.4",
     "@dagrejs/graphlib": "^2.2.4",
     "@fontsource-variable/inter": "^5.1.0",
-    "@invoke-ai/ui-library": "^0.0.44",
+    "@invoke-ai/ui-library": "^0.0.46",
     "@nanostores/react": "^0.7.3",
-    "@reduxjs/toolkit": "2.2.3",
+    "@reduxjs/toolkit": "2.5.1",
     "@roarr/browser-log-writer": "^1.3.0",
+    "@xyflow/react": "^12.4.2",
     "async-mutex": "^0.5.0",
     "chakra-react-select": "^4.9.2",
     "cmdk": "^1.0.0",
@@ -74,8 +75,11 @@
     "idb-keyval": "^6.2.1",
     "jsondiffpatch": "^0.6.0",
     "konva": "^9.3.15",
+    "linkify-react": "^4.2.0",
+    "linkifyjs": "^4.2.0",
     "lodash-es": "^4.17.21",
     "lru-cache": "^11.0.1",
+    "mtwist": "^1.0.2",
     "nanoid": "^5.0.7",
     "nanostores": "^0.11.3",
     "new-github-issue-url": "^1.0.0",
@@ -95,9 +99,9 @@
     "react-icons": "^5.3.0",
     "react-redux": "9.1.2",
     "react-resizable-panels": "^2.1.4",
+    "react-textarea-autosize": "^8.5.7",
     "react-use": "^17.5.1",
     "react-virtuoso": "^4.10.4",
-    "reactflow": "^11.11.4",
     "redux-dynamic-middlewares": "^2.2.0",
     "redux-remember": "^5.1.0",
     "redux-undo": "^1.1.0",
@@ -125,7 +129,7 @@
     "@storybook/addon-storysource": "^8.3.4",
     "@storybook/manager-api": "^8.3.4",
     "@storybook/react": "^8.3.4",
-    "@storybook/react-vite": "^8.3.4",
+    "@storybook/react-vite": "^8.5.5",
     "@storybook/theming": "^8.3.4",
     "@types/dateformat": "^5.0.2",
     "@types/lodash-es": "^4.17.12",
@@ -133,9 +137,9 @@
     "@types/react": "^18.3.11",
     "@types/react-dom": "^18.3.0",
     "@types/uuid": "^10.0.0",
-    "@vitejs/plugin-react-swc": "^3.7.1",
-    "@vitest/coverage-v8": "^1.6.0",
-    "@vitest/ui": "^1.6.0",
+    "@vitejs/plugin-react-swc": "^3.8.0",
+    "@vitest/coverage-v8": "^3.0.6",
+    "@vitest/ui": "^3.0.6",
     "concurrently": "^8.2.2",
     "csstype": "^3.1.3",
     "dpdm": "^3.14.0",
@@ -151,12 +155,12 @@
     "tsafe": "^1.7.5",
     "type-fest": "^4.26.1",
     "typescript": "^5.6.2",
-    "vite": "^5.4.8",
+    "vite": "^6.1.0",
     "vite-plugin-css-injected-by-js": "^3.5.2",
-    "vite-plugin-dts": "^3.9.1",
+    "vite-plugin-dts": "^4.5.0",
     "vite-plugin-eslint": "^1.8.1",
-    "vite-tsconfig-paths": "^4.3.2",
-    "vitest": "^1.6.0"
+    "vite-tsconfig-paths": "^5.1.4",
+    "vitest": "^3.0.6"
   },
   "engines": {
     "pnpm": "8"

invokeai/frontend/web/public/locales/de.json

@@ -99,7 +99,21 @@
         "clipboard": "Zwischenablage",
         "generating": "Generieren",
         "loadingModel": "Lade Modell",
-        "warnings": "Warnungen"
+        "warnings": "Warnungen",
+        "start": "Starten",
+        "count": "Anzahl",
+        "step": "Schritt",
+        "values": "Werte",
+        "min": "Min",
+        "max": "Max",
+        "resetToDefaults": "Auf Standard zurücksetzen",
+        "seed": "Seed",
+        "row": "Reihe",
+        "column": "Spalte",
+        "end": "Ende",
+        "layout": "Layout",
+        "board": "Ordner",
+        "combinatorial": "Kombinatorisch"
     },
     "gallery": {
         "galleryImageSize": "Bildgröße",
@@ -119,7 +133,6 @@
         "autoAssignBoardOnClick": "Board per Klick automatisch zuweisen",
         "noImageSelected": "Kein Bild ausgewählt",
         "starImage": "Bild markieren",
-        "assets": "Ressourcen",
         "unstarImage": "Markierung entfernen",
         "image": "Bild",
         "deleteSelection": "Lösche Auswahl",
@@ -609,7 +622,9 @@
         "hfTokenUnableToVerify": "HF-Token kann nicht überprüft werden",
         "hfTokenUnableToVerifyErrorMessage": "HuggingFace-Token kann nicht überprüft werden. Dies ist wahrscheinlich auf einen Netzwerkfehler zurückzuführen. Bitte versuchen Sie es später erneut.",
         "hfTokenSaved": "HF-Token gespeichert",
-        "hfTokenRequired": "Sie versuchen, ein Modell herunterzuladen, für das ein gültiges HuggingFace-Token erforderlich ist."
+        "hfTokenRequired": "Sie versuchen, ein Modell herunterzuladen, für das ein gültiges HuggingFace-Token erforderlich ist.",
+        "urlUnauthorizedErrorMessage2": "Hier erfahren wie.",
+        "urlForbidden": "Sie haben keinen Zugriff auf dieses Modell"
     },
     "parameters": {
         "images": "Bilder",
@@ -676,7 +691,8 @@
         "iterations": "Iterationen",
         "guidance": "Führung",
         "coherenceMode": "Modus",
-        "recallMetadata": "Metadaten abrufen"
+        "recallMetadata": "Metadaten abrufen",
+        "gaussianBlur": "Gaußsche Unschärfe"
     },
     "settings": {
         "displayInProgress": "Zwischenbilder anzeigen",
@@ -876,7 +892,8 @@
         "canvas": "Leinwand",
         "prompts_one": "Prompt",
         "prompts_other": "Prompts",
-        "batchSize": "Stapelgröße"
+        "batchSize": "Stapelgröße",
+        "confirm": "Bestätigen"
     },
     "metadata": {
         "negativePrompt": "Negativ Beschreibung",
@@ -1282,7 +1299,22 @@
         "unknownFieldType": "$t(nodes.unknownField) Typ: {{type}}",
         "unknownField": "Unbekanntes Feld",
         "unableToUpdateNodes_one": "{{count}} Knoten kann nicht aktualisiert werden",
-        "unableToUpdateNodes_other": "{{count}} Knoten können nicht aktualisiert werden"
+        "unableToUpdateNodes_other": "{{count}} Knoten können nicht aktualisiert werden",
+        "uniformRandomDistribution": "Uniforme Zufallsverteilung",
+        "linearDistribution": "Lineare Verteilung",
+        "generatorNRandomValues_one": "{{count}} Zufallswert",
+        "generatorNRandomValues_other": "{{count}} Zufallswerte",
+        "arithmeticSequence": "Arithmetische Folge",
+        "noBatchGroup": "keine Gruppe",
+        "generatorNoValues": "leer",
+        "generatorLoading": "wird geladen",
+        "generatorLoadFromFile": "Aus Datei laden",
+        "showEdgeLabels": "Kantenbeschriftungen anzeigen",
+        "downloadWorkflowError": "Fehler beim Herunterladen des Arbeitsablaufs",
+        "nodeName": "Knotenname",
+        "description": "Beschreibung",
+        "loadWorkflowDesc": "Arbeitsablauf laden?",
+        "loadWorkflowDesc2": "Ihr aktueller Arbeitsablauf enthält nicht gespeicherte Änderungen."
     },
     "hrf": {
         "enableHrf": "Korrektur für hohe Auflösungen",

invokeai/frontend/web/public/locales/en.json

@@ -90,6 +90,7 @@
         "back": "Back",
         "batch": "Batch Manager",
         "beta": "Beta",
+        "board": "Board",
         "cancel": "Cancel",
         "close": "Close",
         "copy": "Copy",
@@ -177,7 +178,20 @@
         "none": "None",
         "new": "New",
         "generating": "Generating",
-        "warnings": "Warnings"
+        "warnings": "Warnings",
+        "start": "Start",
+        "count": "Count",
+        "step": "Step",
+        "end": "End",
+        "min": "Min",
+        "max": "Max",
+        "values": "Values",
+        "resetToDefaults": "Reset to Defaults",
+        "seed": "Seed",
+        "combinatorial": "Combinatorial",
+        "layout": "Layout",
+        "row": "Row",
+        "column": "Column"
     },
     "hrf": {
         "hrf": "High Resolution Fix",
@@ -209,9 +223,15 @@
         "pauseSucceeded": "Processor Paused",
         "pauseFailed": "Problem Pausing Processor",
         "cancel": "Cancel",
+        "cancelAllExceptCurrentQueueItemAlertDialog": "Canceling all queue items except the current one will stop pending items but allow the in-progress one to finish.",
+        "cancelAllExceptCurrentQueueItemAlertDialog2": "Are you sure you want to cancel all pending queue items?",
+        "cancelAllExceptCurrentTooltip": "Cancel All Except Current Item",
         "cancelTooltip": "Cancel Current Item",
         "cancelSucceeded": "Item Canceled",
         "cancelFailed": "Problem Canceling Item",
+        "retrySucceeded": "Item Retried",
+        "retryFailed": "Problem Retrying Item",
+        "confirm": "Confirm",
         "prune": "Prune",
         "pruneTooltip": "Prune {{item_count}} Completed Items",
         "pruneSucceeded": "Pruned {{item_count}} Completed Items from Queue",
@@ -222,6 +242,7 @@
         "clearFailed": "Problem Clearing Queue",
         "cancelBatch": "Cancel Batch",
         "cancelItem": "Cancel Item",
+        "retryItem": "Retry Item",
         "cancelBatchSucceeded": "Batch Canceled",
         "cancelBatchFailed": "Problem Canceling Batch",
         "clearQueueAlertDialog": "Clearing the queue immediately cancels any processing items and clears the queue entirely. Pending filters will be canceled.",
@@ -284,10 +305,16 @@
         "disableFailed": "Problem Disabling Invocation Cache",
         "useCache": "Use Cache"
     },
+    "modelCache": {
+        "clear": "Clear Model Cache",
+        "clearSucceeded": "Model Cache Cleared",
+        "clearFailed": "Problem Clearing Model Cache"
+    },
     "gallery": {
         "gallery": "Gallery",
-        "alwaysShowImageSizeBadge": "Always Show Image Size Badge",
+        "images": "Images",
         "assets": "Assets",
+        "alwaysShowImageSizeBadge": "Always Show Image Size Badge",
         "assetsTab": "Files you’ve uploaded for use in your projects.",
         "autoAssignBoardOnClick": "Auto-Assign Board on Click",
         "autoSwitchNewImages": "Auto-Switch to New Images",
@@ -850,6 +877,23 @@
         "defaultVAE": "Default VAE"
     },
     "nodes": {
+        "arithmeticSequence": "Arithmetic Sequence",
+        "linearDistribution": "Linear Distribution",
+        "uniformRandomDistribution": "Uniform Random Distribution",
+        "parseString": "Parse String",
+        "splitOn": "Split On",
+        "noBatchGroup": "no group",
+        "generatorImagesCategory": "Category",
+        "generatorImages_one": "{{count}} image",
+        "generatorImages_other": "{{count}} images",
+        "generatorNRandomValues_one": "{{count}} random value",
+        "generatorNRandomValues_other": "{{count}} random values",
+        "generatorNoValues": "empty",
+        "generatorLoading": "loading",
+        "generatorLoadFromFile": "Load from File",
+        "generatorImagesFromBoard": "Images from Board",
+        "dynamicPromptsRandom": "Dynamic Prompts (Random)",
+        "dynamicPromptsCombinatorial": "Dynamic Prompts (Combinatorial)",
         "addNode": "Add Node",
         "addNodeToolTip": "Add Node (Shift+A, Space)",
         "addLinearView": "Add to Linear View",
@@ -864,6 +908,8 @@
         "missingNode": "Missing invocation node",
         "missingInvocationTemplate": "Missing invocation template",
         "missingFieldTemplate": "Missing field template",
+        "missingSourceOrTargetNode": "Missing source or target node",
+        "missingSourceOrTargetHandle": "Missing source or target handle",
         "nodePack": "Node pack",
         "collection": "Collection",
         "singleFieldType": "{{name}} (Single)",
@@ -875,6 +921,7 @@
         "currentImage": "Current Image",
         "currentImageDescription": "Displays the current image in the Node Editor",
         "downloadWorkflow": "Download Workflow JSON",
+        "downloadWorkflowError": "Error downloading workflow",
         "edge": "Edge",
         "edit": "Edit",
         "editMode": "Edit in Workflow Editor",
@@ -899,6 +946,7 @@
         "noWorkflows": "No Workflows",
         "noMatchingWorkflows": "No Matching Workflows",
         "noWorkflow": "No Workflow",
+        "unableToUpdateNode": "Node update failed: node {{node}} of type {{type}} (may require deleting and recreating)",
         "mismatchedVersion": "Invalid node: node {{node}} of type {{type}} has mismatched version (try updating?)",
         "missingTemplate": "Invalid node: node {{node}} of type {{type}} missing template (not installed?)",
         "sourceNodeDoesNotExist": "Invalid edge: source/output node {{node}} does not exist",
@@ -906,12 +954,14 @@
         "sourceNodeFieldDoesNotExist": "Invalid edge: source/output field {{node}}.{{field}} does not exist",
         "targetNodeFieldDoesNotExist": "Invalid edge: target/input field {{node}}.{{field}} does not exist",
         "deletedInvalidEdge": "Deleted invalid edge {{source}} -> {{target}}",
+        "deletedMissingNodeFieldFormElement": "Deleted missing form field: node {{nodeId}} field {{fieldName}}",
         "noConnectionInProgress": "No connection in progress",
         "node": "Node",
         "nodeOutputs": "Node Outputs",
         "nodeSearch": "Search for nodes",
         "nodeTemplate": "Node Template",
         "nodeType": "Node Type",
+        "nodeName": "Node Name",
         "noFieldsLinearview": "No fields added to Linear View",
         "noFieldsViewMode": "This workflow has no selected fields to display. View the full workflow to configure values.",
         "workflowHelpText": "Need Help? Check out our guide to <LinkComponent>Getting Started with Workflows</LinkComponent>.",
@@ -920,6 +970,7 @@
         "nodeVersion": "Node Version",
         "noOutputRecorded": "No outputs recorded",
         "notes": "Notes",
+        "description": "Description",
         "notesDescription": "Add notes about your workflow",
         "problemSettingTitle": "Problem Setting Title",
         "resetToDefaultValue": "Reset to default value",
@@ -929,6 +980,8 @@
         "newWorkflow": "New Workflow",
         "newWorkflowDesc": "Create a new workflow?",
         "newWorkflowDesc2": "Your current workflow has unsaved changes.",
+        "loadWorkflowDesc": "Load workflow?",
+        "loadWorkflowDesc2": "Your current workflow has unsaved changes.",
         "clearWorkflow": "Clear Workflow",
         "clearWorkflowDesc": "Clear this workflow and start a new one?",
         "clearWorkflowDesc2": "Your current workflow has unsaved changes.",
@@ -958,6 +1011,7 @@
         "unknownOutput": "Unknown output: {{name}}",
         "updateNode": "Update Node",
         "updateApp": "Update App",
+        "loadingTemplates": "Loading {{name}}",
         "updateAllNodes": "Update Nodes",
         "allNodesUpdated": "All Nodes Updated",
         "unableToUpdateNodes_one": "Unable to update {{count}} node",
@@ -989,7 +1043,11 @@
         "imageAccessError": "Unable to find image {{image_name}}, resetting to default",
         "boardAccessError": "Unable to find board {{board_id}}, resetting to default",
         "modelAccessError": "Unable to find model {{key}}, resetting to default",
-        "saveToGallery": "Save To Gallery"
+        "saveToGallery": "Save To Gallery",
+        "addItem": "Add Item",
+        "generateValues": "Generate Values",
+        "floatRangeGenerator": "Float Range Generator",
+        "integerRangeGenerator": "Integer Range Generator"
     },
     "parameters": {
         "aspect": "Aspect",
@@ -1024,11 +1082,23 @@
             "addingImagesTo": "Adding images to",
             "invoke": "Invoke",
             "missingFieldTemplate": "Missing field template",
-            "missingInputForField": "{{nodeLabel}} -> {{fieldLabel}}: missing input",
+            "missingInputForField": "missing input",
             "missingNodeTemplate": "Missing node template",
-            "collectionEmpty": "{{nodeLabel}} -> {{fieldLabel}} empty collection",
-            "collectionTooFewItems": "{{nodeLabel}} -> {{fieldLabel}}: too few items, minimum {{minItems}}",
-            "collectionTooManyItems": "{{nodeLabel}} -> {{fieldLabel}}: too many items, maximum {{maxItems}}",
+            "emptyBatches": "empty batches",
+            "batchNodeNotConnected": "Batch node not connected: {{label}}",
+            "batchNodeEmptyCollection": "Some batch nodes have empty collections",
+            "collectionEmpty": "empty collection",
+            "collectionTooFewItems": "too few items, minimum {{minItems}}",
+            "collectionTooManyItems": "too many items, maximum {{maxItems}}",
+            "collectionStringTooLong": "too long, max {{maxLength}}",
+            "collectionStringTooShort": "too short, min {{minLength}}",
+            "collectionNumberGTMax": "{{value}} > {{maximum}} (inc max)",
+            "collectionNumberLTMin": "{{value}} < {{minimum}} (inc min)",
+            "collectionNumberGTExclusiveMax": "{{value}} >= {{exclusiveMaximum}} (exc max)",
+            "collectionNumberLTExclusiveMin": "{{value}} <= {{exclusiveMinimum}} (exc min)",
+            "collectionNumberNotMultipleOf": "{{value}} not multiple of {{multipleOf}}",
+            "batchNodeCollectionSizeMismatchNoGroupId": "Batch group collection size mismatch",
+            "batchNodeCollectionSizeMismatch": "Collection size mismatch on Batch {{batchGroupId}}",
             "noModelSelected": "No model selected",
             "noT5EncoderModelSelected": "No T5 Encoder model selected for FLUX generation",
             "noFLUXVAEModelSelected": "No VAE model selected for FLUX generation",
@@ -1100,7 +1170,8 @@
             "perPromptLabel": "Seed per Image",
             "perPromptDesc": "Use a different seed for each image"
         },
-        "loading": "Generating Dynamic Prompts..."
+        "loading": "Generating Dynamic Prompts...",
+        "promptsToGenerate": "Prompts to Generate"
     },
     "sdxl": {
         "cfgScale": "CFG Scale",
@@ -1200,6 +1271,8 @@
         "problemCopyingLayer": "Unable to Copy Layer",
         "problemSavingLayer": "Unable to Save Layer",
         "problemDownloadingImage": "Unable to Download Image",
+        "pasteSuccess": "Pasted to {{destination}}",
+        "pasteFailed": "Paste Failed",
         "prunedQueue": "Pruned Queue",
         "sentToCanvas": "Sent to Canvas",
         "sentToUpscale": "Sent to Upscale",
@@ -1216,7 +1289,10 @@
         "workflowLoaded": "Workflow Loaded",
         "problemRetrievingWorkflow": "Problem Retrieving Workflow",
         "workflowDeleted": "Workflow Deleted",
-        "problemDeletingWorkflow": "Problem Deleting Workflow"
+        "problemDeletingWorkflow": "Problem Deleting Workflow",
+        "unableToCopy": "Unable to Copy",
+        "unableToCopyDesc": "Your browser does not support clipboard access. Firefox users may be able to fix this by following ",
+        "unableToCopyDesc_theseSteps": "these steps"
     },
     "popovers": {
         "clipSkip": {
@@ -1641,7 +1717,38 @@
         "download": "Download",
         "copyShareLink": "Copy Share Link",
         "copyShareLinkForWorkflow": "Copy Share Link for Workflow",
-        "delete": "Delete"
+        "delete": "Delete",
+        "openLibrary": "Open Library",
+        "builder": {
+            "deleteAllElements": "Delete All Form Elements",
+            "resetAllNodeFields": "Reset All Node Fields",
+            "builder": "Form Builder",
+            "layout": "Layout",
+            "row": "Row",
+            "column": "Column",
+            "container": "Container",
+            "heading": "Heading",
+            "text": "Text",
+            "divider": "Divider",
+            "nodeField": "Node Field",
+            "zoomToNode": "Zoom to Node",
+            "nodeFieldTooltip": "To add a node field, click the small plus sign button on the field in the Workflow Editor, or drag the field by its name into the form.",
+            "addToForm": "Add to Form",
+            "label": "Label",
+            "showDescription": "Show Description",
+            "component": "Component",
+            "numberInput": "Number Input",
+            "singleLine": "Single Line",
+            "multiLine": "Multi Line",
+            "slider": "Slider",
+            "both": "Both",
+            "emptyRootPlaceholderViewMode": "Click Edit to start building a form for this workflow.",
+            "emptyRootPlaceholderEditMode": "Drag a form element or node field here to get started.",
+            "containerPlaceholder": "Empty Container",
+            "headingPlaceholder": "Empty Heading",
+            "textPlaceholder": "Empty Text",
+            "workflowBuilderAlphaWarning": "The workflow builder is currently in alpha. There may be breaking changes before the stable release."
+        }
     },
     "controlLayers": {
         "regional": "Regional",
@@ -1656,6 +1763,8 @@
         "cropLayerToBbox": "Crop Layer to Bbox",
         "savedToGalleryOk": "Saved to Gallery",
         "savedToGalleryError": "Error saving to gallery",
+        "regionCopiedToClipboard": "{{region}} Copied to Clipboard",
+        "copyRegionError": "Error copying {{region}}",
         "newGlobalReferenceImageOk": "Created Global Reference Image",
         "newGlobalReferenceImageError": "Problem Creating Global Reference Image",
         "newRegionalReferenceImageOk": "Created Regional Reference Image",
@@ -1766,6 +1875,14 @@
         "newControlLayer": "New $t(controlLayers.controlLayer)",
         "newInpaintMask": "New $t(controlLayers.inpaintMask)",
         "newRegionalGuidance": "New $t(controlLayers.regionalGuidance)",
+        "pasteTo": "Paste To",
+        "pasteToAssets": "Assets",
+        "pasteToAssetsDesc": "Paste to Assets",
+        "pasteToBbox": "Bbox",
+        "pasteToBboxDesc": "New Layer (in Bbox)",
+        "pasteToCanvas": "Canvas",
+        "pasteToCanvasDesc": "New Layer (in Canvas)",
+        "pastedTo": "Pasted to {{destination}}",
         "transparency": "Transparency",
         "enableTransparencyEffect": "Enable Transparency Effect",
         "disableTransparencyEffect": "Disable Transparency Effect",
@@ -1794,7 +1911,7 @@
         "resetGenerationSettings": "Reset Generation Settings",
         "replaceCurrent": "Replace Current",
         "controlLayerEmptyState": "<UploadButton>Upload an image</UploadButton>, drag an image from the <GalleryButton>gallery</GalleryButton> onto this layer, or draw on the canvas to get started.",
-        "referenceImageEmptyState": "<UploadButton>Upload an image</UploadButton> or drag an image from the <GalleryButton>gallery</GalleryButton> onto this layer to get started.",
+        "referenceImageEmptyState": "<UploadButton>Upload an image</UploadButton>, drag an image from the <GalleryButton>gallery</GalleryButton> onto this layer, or <PullBboxButton>pull the bounding box into this layer</PullBboxButton> to get started.",
         "warnings": {
             "problemsFound": "Problems found",
             "unsupportedModel": "layer not supported for selected base model",
@@ -1810,6 +1927,10 @@
             "rgAutoNegativeNotSupported": "Auto-Negative not supported for selected base model",
             "rgNoRegion": "no region drawn"
         },
+        "errors": {
+            "unableToFindImage": "Unable to find image",
+            "unableToLoadImage": "Unable to Load Image"
+        },
         "controlMode": {
             "controlMode": "Control Mode",
             "balanced": "Balanced (recommended)",
@@ -1932,6 +2053,48 @@
                 "description": "Generates an edge map from the selected layer using the PiDiNet edge detection model.",
                 "scribble": "Scribble",
                 "quantize_edges": "Quantize Edges"
+            },
+            "img_blur": {
+                "label": "Blur Image",
+                "description": "Blurs the selected layer.",
+                "blur_type": "Blur Type",
+                "blur_radius": "Radius",
+                "gaussian_type": "Gaussian",
+                "box_type": "Box"
+            },
+            "img_noise": {
+                "label": "Noise Image",
+                "description": "Adds noise to the selected layer.",
+                "noise_type": "Noise Type",
+                "noise_amount": "Amount",
+                "gaussian_type": "Gaussian",
+                "salt_and_pepper_type": "Salt and Pepper",
+                "noise_color": "Colored Noise",
+                "size": "Noise Size"
+            },
+            "adjust_image": {
+                "label": "Adjust Image",
+                "description": "Adjusts the selected channel of an image.",
+                "channel": "Channel",
+                "value_setting": "Value",
+                "scale_values": "Scale Values",
+                "red": "Red (RGBA)",
+                "green": "Green (RGBA)",
+                "blue": "Blue (RGBA)",
+                "alpha": "Alpha (RGBA)",
+                "cyan": "Cyan (CMYK)",
+                "magenta": "Magenta (CMYK)",
+                "yellow": "Yellow (CMYK)",
+                "black": "Black (CMYK)",
+                "hue": "Hue (HSV)",
+                "saturation": "Saturation (HSV)",
+                "value": "Value (HSV)",
+                "luminosity": "Luminosity (LAB)",
+                "a": "A (LAB)",
+                "b": "B (LAB)",
+                "y": "Y (YCbCr)",
+                "cb": "Cb (YCbCr)",
+                "cr": "Cr (YCbCr)"
             }
         },
         "transform": {
@@ -2005,7 +2168,10 @@
             "newRasterLayer": "New Raster Layer",
             "newInpaintMask": "New Inpaint Mask",
             "newRegionalGuidance": "New Regional Guidance",
-            "cropCanvasToBbox": "Crop Canvas to Bbox"
+            "cropCanvasToBbox": "Crop Canvas to Bbox",
+            "copyToClipboard": "Copy to Clipboard",
+            "copyCanvasToClipboard": "Copy Canvas to Clipboard",
+            "copyBboxToClipboard": "Copy Bbox to Clipboard"
         },
         "stagingArea": {
             "accept": "Accept",
@@ -2139,7 +2305,10 @@
     },
     "whatsNew": {
         "whatsNewInInvoke": "What's New in Invoke",
-        "items": ["Low-VRAM mode", "Dynamic memory management", "Faster model loading times", "Fewer memory errors"],
+        "items": [
+            "Memory Management: New setting for users with Nvidia GPUs to reduce VRAM usage.",
+            "Performance: Continued improvements to overall application performance and responsiveness."
+        ],
         "readReleaseNotes": "Read Release Notes",
         "watchRecentReleaseVideos": "Watch Recent Release Videos",
         "watchUiUpdatesOverview": "Watch UI Updates Overview"

invokeai/frontend/web/public/locales/es.json

@@ -109,7 +109,6 @@
         "deleteImage_many": "Eliminar {{count}} Imágenes",
         "deleteImage_other": "Eliminar {{count}} Imágenes",
         "deleteImagePermanent": "Las imágenes eliminadas no se pueden restaurar.",
-        "assets": "Activos",
         "autoAssignBoardOnClick": "Asignar automática tableros al hacer clic",
         "gallery": "Galería",
         "noImageSelected": "Sin imágenes seleccionadas",
@@ -901,9 +900,7 @@
         }
     },
     "newUserExperience": {
-        "downloadStarterModels": "Descargar modelos de inicio",
         "toGetStarted": "Para empezar, introduzca un mensaje en el cuadro y haga clic en <StrongComponent>Invocar</StrongComponent> para generar su primera imagen. Seleccione una plantilla para mejorar los resultados. Puede elegir guardar sus imágenes directamente en <StrongComponent>Galería</StrongComponent> o editarlas en <StrongComponent>Lienzo</StrongComponent>.",
-        "importModels": "Importar modelos",
         "noModelsInstalled": "Parece que no tienes ningún modelo instalado",
         "gettingStartedSeries": "¿Desea más orientación? Consulte nuestra <LinkComponent>Serie de introducción</LinkComponent> para obtener consejos sobre cómo aprovechar todo el potencial de Invoke Studio.",
         "toGetStartedLocal": "Para empezar, asegúrate de descargar o importar los modelos necesarios para ejecutar Invoke. A continuación, introduzca un mensaje en el cuadro y haga clic en <StrongComponent>Invocar</StrongComponent> para generar su primera imagen. Seleccione una plantilla para mejorar los resultados. Puede elegir guardar sus imágenes directamente en <StrongComponent>Galería</StrongComponent> o editarlas en el <StrongComponent>Lienzo</StrongComponent>."

invokeai/frontend/web/public/locales/fr.json

@@ -98,7 +98,22 @@
         "close": "Fermer",
         "clipboard": "Presse-papier",
         "loadingModel": "Chargement du modèle",
-        "generating": "En Génération"
+        "generating": "En Génération",
+        "warnings": "Alertes",
+        "layout": "Disposition",
+        "row": "Ligne",
+        "column": "Colonne",
+        "start": "Commencer",
+        "board": "Planche",
+        "count": "Quantité",
+        "step": "Étape",
+        "end": "Fin",
+        "min": "Min",
+        "max": "Max",
+        "values": "Valeurs",
+        "resetToDefaults": "Réinitialiser par défaut",
+        "seed": "Graine",
+        "combinatorial": "Combinatoire"
     },
     "gallery": {
         "galleryImageSize": "Taille de l'image",
@@ -114,7 +129,6 @@
         "sortDirection": "Direction de tri",
         "sideBySide": "Côte-à-Côte",
         "hover": "Au passage de la souris",
-        "assets": "Ressources",
         "alwaysShowImageSizeBadge": "Toujours montrer le badge de taille de l'Image",
         "gallery": "Galerie",
         "bulkDownloadRequestFailed": "Problème lors de la préparation du téléchargement",
@@ -166,7 +180,9 @@
         "imagesSettings": "Paramètres des images de la galerie",
         "assetsTab": "Fichiers que vous avez importés pour vos projets.",
         "imagesTab": "Images que vous avez créées et enregistrées dans Invoke.",
-        "boardsSettings": "Paramètres des planches"
+        "boardsSettings": "Paramètres des planches",
+        "assets": "Ressources",
+        "images": "Images"
     },
     "modelManager": {
         "modelManager": "Gestionnaire de modèle",
@@ -290,7 +306,7 @@
         "usingDefaultSettings": "Utilisation des paramètres par défaut du modèle",
         "defaultSettingsOutOfSync": "Certain paramètres ne correspondent pas aux valeurs par défaut du modèle :",
         "restoreDefaultSettings": "Cliquez pour utiliser les paramètres par défaut du modèle.",
-        "hfForbiddenErrorMessage": "Nous vous recommandons de visiter la page du modèle sur HuggingFace.com. Le propriétaire peut exiger l'acceptation des conditions pour pouvoir télécharger.",
+        "hfForbiddenErrorMessage": "Nous vous recommandons de visiter la page du modèle. Le propriétaire peut exiger l'acceptation des conditions pour pouvoir télécharger.",
         "hfTokenRequired": "Vous essayez de télécharger un modèle qui nécessite un token HuggingFace valide.",
         "clipLEmbed": "CLIP-L Embed",
         "hfTokenSaved": "Token HF enregistré",
@@ -302,7 +318,12 @@
         "hfTokenHelperText": "Un token HF est requis pour utiliser certains modèles. Cliquez ici pour créer ou obtenir votre token.",
         "hfTokenInvalid": "Token HF invalide ou manquant",
         "hfForbidden": "Vous n'avez pas accès à ce modèle HF.",
-        "hfTokenInvalidErrorMessage2": "Mettre à jour dans le "
+        "hfTokenInvalidErrorMessage2": "Mettre à jour dans le ",
+        "controlLora": "Controle LoRA",
+        "urlUnauthorizedErrorMessage2": "Découvrir comment ici.",
+        "urlUnauthorizedErrorMessage": "Vous devrez peut-être configurer un jeton API pour accéder à ce modèle.",
+        "urlForbidden": "Vous n'avez pas accès à ce modèle",
+        "urlForbiddenErrorMessage": "Vous devrez peut-être demander l'autorisation du site qui distribue le modèle."
     },
     "parameters": {
         "images": "Images",
@@ -333,7 +354,7 @@
         "showOptionsPanel": "Afficher le panneau latéral (O ou T)",
         "invoke": {
             "noPrompts": "Aucun prompts généré",
-            "missingInputForField": "{{nodeLabel}} -> {{fieldLabel}} entrée manquante",
+            "missingInputForField": "entrée manquante",
             "missingFieldTemplate": "Modèle de champ manquant",
             "invoke": "Invoke",
             "addingImagesTo": "Ajouter des images à",
@@ -344,18 +365,31 @@
             "fluxModelIncompatibleBboxHeight": "$t(parameters.invoke.fluxRequiresDimensionsToBeMultipleOf16), la hauteur de la bounding box est {{height}}",
             "fluxModelIncompatibleScaledBboxHeight": "$t(parameters.invoke.fluxRequiresDimensionsToBeMultipleOf16), la hauteur de la bounding box est {{height}}",
             "noFLUXVAEModelSelected": "Aucun modèle VAE sélectionné pour la génération FLUX",
-            "canvasIsTransforming": "La Toile se transforme",
-            "canvasIsRasterizing": "La Toile se rastérise",
+            "canvasIsTransforming": "La Toile est occupée (en transformation)",
+            "canvasIsRasterizing": "La Toile est occupée (en rastérisation)",
             "noCLIPEmbedModelSelected": "Aucun modèle CLIP Embed sélectionné pour la génération FLUX",
-            "canvasIsFiltering": "La Toile est en train de filtrer",
+            "canvasIsFiltering": "La Toile est occupée (en filtration)",
             "fluxModelIncompatibleBboxWidth": "$t(parameters.invoke.fluxRequiresDimensionsToBeMultipleOf16), la largeur de la bounding box est {{width}}",
             "noT5EncoderModelSelected": "Aucun modèle T5 Encoder sélectionné pour la génération FLUX",
             "fluxModelIncompatibleScaledBboxWidth": "$t(parameters.invoke.fluxRequiresDimensionsToBeMultipleOf16), la largeur de la bounding box mise à l'échelle est {{width}}",
-            "canvasIsCompositing": "La toile est en train de composer",
-            "collectionEmpty": "{{nodeLabel}} -> {{fieldLabel}} collection vide",
-            "collectionTooFewItems": "{{nodeLabel}} -> {{fieldLabel}} : trop peu d'éléments, minimum {{minItems}}",
-            "collectionTooManyItems": "{{nodeLabel}} -> {{fieldLabel}} : trop d'éléments, maximum {{maxItems}}",
-            "canvasIsSelectingObject": "La toile est occupée (sélection d'objet)"
+            "canvasIsCompositing": "La Toile est occupée (en composition)",
+            "collectionTooFewItems": "trop peu d'éléments, minimum {{minItems}}",
+            "collectionTooManyItems": "trop d'éléments, maximum {{maxItems}}",
+            "canvasIsSelectingObject": "La toile est occupée (sélection d'objet)",
+            "emptyBatches": "lots vides",
+            "batchNodeNotConnected": "Noeud de lots non connecté : {{label}}",
+            "fluxModelMultipleControlLoRAs": "Vous ne pouvez utiliser qu'un seul Control LoRA à la fois",
+            "collectionNumberLTMin": "{{value}} < {{minimum}} (incl. min)",
+            "collectionNumberGTMax": "{{value}} > {{maximum}} (incl. max)",
+            "collectionNumberGTExclusiveMax": "{{value}} >= {{exclusiveMaximum}} (max exc)",
+            "batchNodeEmptyCollection": "Certains nœuds de lot ont des collections vides",
+            "batchNodeCollectionSizeMismatch": "Non-concordance de taille de collection sur le lot {{batchGroupId}}",
+            "collectionStringTooLong": "trop long, max {{maxLength}}",
+            "collectionNumberNotMultipleOf": "{{value}} n'est pas un multiple de {{multipleOf}}",
+            "collectionEmpty": "collection vide",
+            "collectionStringTooShort": "trop court, min {{minLength}}",
+            "collectionNumberLTExclusiveMin": "{{value}} <= {{exclusiveMinimum}} (min exc)",
+            "batchNodeCollectionSizeMismatchNoGroupId": "Taille de collection de groupe par lot non conforme"
         },
         "negativePromptPlaceholder": "Prompt Négatif",
         "positivePromptPlaceholder": "Prompt Positif",
@@ -499,7 +533,13 @@
         "uploadFailedInvalidUploadDesc_withCount_one": "Doit être au maximum une image PNG ou JPEG.",
         "uploadFailedInvalidUploadDesc_withCount_many": "Doit être au maximum {{count}} images PNG ou JPEG.",
         "uploadFailedInvalidUploadDesc_withCount_other": "Doit être au maximum {{count}} images PNG ou JPEG.",
-        "addedToUncategorized": "Ajouté aux ressources de la planche $t(boards.uncategorized)"
+        "addedToUncategorized": "Ajouté aux ressources de la planche $t(boards.uncategorized)",
+        "pasteSuccess": "Collé à {{destination}}",
+        "pasteFailed": "Échec du collage",
+        "outOfMemoryErrorDescLocal": "Suivez notre <LinkComponent>guide Low VRAM</LinkComponent> pour réduire les OOMs.",
+        "unableToCopy": "Incapable de Copier",
+        "unableToCopyDesc": "Votre navigateur ne prend pas en charge l'accès au presse-papiers. Les utilisateurs de Firefox peuvent peut-être résoudre ce problème en suivant ",
+        "unableToCopyDesc_theseSteps": "ces étapes"
     },
     "accessibility": {
         "uploadImage": "Importer une image",
@@ -657,7 +697,14 @@
         "iterations_many": "Itérations",
         "iterations_other": "Itérations",
         "back": "fin",
-        "batchSize": "Taille de lot"
+        "batchSize": "Taille de lot",
+        "retryFailed": "Problème de nouvelle tentative de l'élément",
+        "retrySucceeded": "Élément Retenté",
+        "retryItem": "Réessayer l'élement",
+        "cancelAllExceptCurrentQueueItemAlertDialog": "Annuler tous les éléments de la file d'attente, sauf celui en cours, arrêtera les éléments en attente mais permettra à celui en cours de se terminer.",
+        "cancelAllExceptCurrentQueueItemAlertDialog2": "Êtes-vous sûr de vouloir annuler tous les éléments en attente dans la file d'attente ?",
+        "cancelAllExceptCurrentTooltip": "Annuler tout sauf l'élément actuel",
+        "confirm": "Confirmer"
     },
     "prompt": {
         "noMatchingTriggers": "Pas de déclancheurs correspondants",
@@ -1029,7 +1076,9 @@
         "controlNetWeight": {
             "heading": "Poids",
             "paragraphs": [
-                "Poids du Control Adapter. Un poids plus élevé aura un impact plus important sur l'image finale."
+                "Poids du Control Adapter. Un poids plus élevé aura un impact plus important sur l'image finale.",
+                "• Poids plus élevé (.75-2) : Crée un impact plus significatif sur le résultat final.",
+                "• Poids inférieur (0-.75) : Crée un impact plus faible sur le résultat final."
             ]
         },
         "compositingMaskAdjustments": {
@@ -1074,8 +1123,9 @@
         "controlNetBeginEnd": {
             "heading": "Pourcentage de début / de fin d'étape",
             "paragraphs": [
-                "La partie du processus de débruitage à laquelle le Control Adapter sera appliqué.",
-                "En général, les Control Adapter appliqués au début du processus guident la composition, tandis que les Control Adapter appliqués à la fin guident les détails."
+                "Ce paramètre détérmine quelle portion du processus de débruitage (génération) utilisera cette couche comme guide.",
+                "En général, les Control Adapter appliqués au début du processus guident la composition, tandis que les Control Adapter appliqués à la fin guident les détails.",
+                "• Étape de fin (%): Spécifie quand arrêter d'appliquer le guide de cette couche et revenir aux guides généraux du modèle et aux autres paramètres."
             ]
         },
         "controlNetControlMode": {
@@ -1440,7 +1490,8 @@
         "showDynamicPrompts": "Afficher les Prompts dynamiques",
         "dynamicPrompts": "Prompts Dynamiques",
         "promptsPreview": "Prévisualisation des Prompts",
-        "loading": "Génération des Pompts Dynamiques..."
+        "loading": "Génération des Pompts Dynamiques...",
+        "promptsToGenerate": "Prompts à générer"
     },
     "metadata": {
         "positivePrompt": "Prompt Positif",
@@ -1632,7 +1683,41 @@
         "boardAccessError": "Impossible de trouver la planche {{board_id}}, réinitialisation à la valeur par défaut",
         "workflowHelpText": "Besoin d'aide ? Consultez notre guide sur <LinkComponent>Comment commencer avec les Workflows</LinkComponent>.",
         "noWorkflows": "Aucun Workflows",
-        "noMatchingWorkflows": "Aucun Workflows correspondant"
+        "noMatchingWorkflows": "Aucun Workflows correspondant",
+        "arithmeticSequence": "Séquence Arithmétique",
+        "uniformRandomDistribution": "Distribution Aléatoire Uniforme",
+        "noBatchGroup": "aucun groupe",
+        "generatorLoading": "chargement",
+        "generatorLoadFromFile": "Charger depuis un Fichier",
+        "dynamicPromptsRandom": "Prompts Dynamiques (Aléatoire)",
+        "integerRangeGenerator": "Générateur d'interval d'entiers",
+        "generateValues": "Générer Valeurs",
+        "linearDistribution": "Distribution Linéaire",
+        "floatRangeGenerator": "Générateur d'interval de nombres décimaux",
+        "generatorNRandomValues_one": "{{count}} valeur aléatoire",
+        "generatorNRandomValues_many": "{{count}} valeurs aléatoires",
+        "generatorNRandomValues_other": "{{count}} valeurs aléatoires",
+        "dynamicPromptsCombinatorial": "Prompts Dynamiques (Combinatoire)",
+        "parseString": "Analyser la chaine de charactères",
+        "internalDesc": "Cette invocation est utilisée internalement par Invoke. En fonction des mises à jours il est possible que des changements y soit effectués ou qu'elle soit supprimé sans prévention.",
+        "splitOn": "Diviser sur",
+        "generatorNoValues": "vide",
+        "addItem": "Ajouter un élément",
+        "specialDesc": "Cette invocation nécessite un traitement spécial dans l'application. Par exemple, les nœuds Batch sont utilisés pour mettre en file d'attente plusieurs graphes à partir d'un seul workflow.",
+        "unableToUpdateNode": "La mise à jour du nœud a échoué : nœud {{node}} de type {{type}} (peut nécessiter la suppression et la recréation).",
+        "deletedMissingNodeFieldFormElement": "Champ de formulaire manquant supprimé : nœud {{nodeId}} champ {{fieldName}}",
+        "nodeName": "Nom du nœud",
+        "description": "Description",
+        "loadWorkflowDesc": "Charger le workflow ?",
+        "missingSourceOrTargetNode": "Nœud source ou cible manquant",
+        "generatorImagesCategory": "Catégorie",
+        "generatorImagesFromBoard": "Images de la Planche",
+        "missingSourceOrTargetHandle": "Manque de gestionnaire source ou cible",
+        "loadingTemplates": "Chargement de {{name}}",
+        "loadWorkflowDesc2": "Votre workflow actuel contient des modifications non enregistrées.",
+        "generatorImages_one": "{{count}} image",
+        "generatorImages_many": "{{count}} images",
+        "generatorImages_other": "{{count}} images"
     },
     "models": {
         "noMatchingModels": "Aucun modèle correspondant",
@@ -1691,13 +1776,41 @@
         "deleteWorkflow2": "Êtes-vous sûr de vouloir supprimer ce Workflow ? Cette action ne peut pas être annulé.",
         "download": "Télécharger",
         "copyShareLinkForWorkflow": "Copier le lien de partage pour le Workflow",
-        "delete": "Supprimer"
+        "delete": "Supprimer",
+        "builder": {
+            "component": "Composant",
+            "numberInput": "Entrée de nombre",
+            "slider": "Curseur",
+            "both": "Les deux",
+            "singleLine": "Ligne unique",
+            "multiLine": "Multi Ligne",
+            "headingPlaceholder": "En-tête vide",
+            "emptyRootPlaceholderEditMode": "Faites glisser un élément de formulaire ou un champ de nœud ici pour commencer.",
+            "emptyRootPlaceholderViewMode": "Cliquez sur Modifier pour commencer à créer un formulaire pour ce workflow.",
+            "containerPlaceholder": "Conteneur Vide",
+            "row": "Ligne",
+            "column": "Colonne",
+            "layout": "Mise en page",
+            "nodeField": "Champ de nœud",
+            "zoomToNode": "Zoomer sur le nœud",
+            "nodeFieldTooltip": "Pour ajouter un champ de nœud, cliquez sur le petit bouton plus sur le champ dans l'Éditeur de Workflow, ou faites glisser le champ par son nom dans le formulaire.",
+            "addToForm": "Ajouter au formulaire",
+            "label": "Étiquette",
+            "textPlaceholder": "Texte vide",
+            "builder": "Constructeur de Formulaire",
+            "resetAllNodeFields": "Réinitialiser tous les champs de nœud",
+            "deleteAllElements": "Supprimer tous les éléments de formulaire",
+            "workflowBuilderAlphaWarning": "Le constructeur de workflow est actuellement en version alpha. Il peut y avoir des changements majeurs avant la version stable.",
+            "showDescription": "Afficher la description"
+        },
+        "openLibrary": "Ouvrir la Bibliothèque"
     },
     "whatsNew": {
         "whatsNewInInvoke": "Quoi de neuf dans Invoke",
         "watchRecentReleaseVideos": "Regarder les vidéos des dernières versions",
         "items": [
-            "<StrongComponent>FLUX Guidage Régional (bêta)</StrongComponent> : Notre version bêta de FLUX Guidage Régional est en ligne pour le contrôle des prompt régionaux."
+            "<StrongComponent>FLUX Guidage Régional (bêta)</StrongComponent> : Notre version bêta de FLUX Guidage Régional est en ligne pour le contrôle des prompt régionaux.",
+            "Autres améliorations : mise en file d'attente par lots plus rapide, meilleur redimensionnement, sélecteur de couleurs amélioré et nœuds de métadonnées."
         ],
         "readReleaseNotes": "Notes de version",
         "watchUiUpdatesOverview": "Aperçu des mises à jour de l'interface utilisateur"
@@ -1811,7 +1924,49 @@
             "cancel": "Annuler",
             "advanced": "Avancé",
             "processingLayerWith": "Calque de traitement avec le filtre {{type}}.",
-            "forMoreControl": "Pour plus de contrôle, cliquez sur Avancé ci-dessous."
+            "forMoreControl": "Pour plus de contrôle, cliquez sur Avancé ci-dessous.",
+            "adjust_image": {
+                "b": "B (LAB)",
+                "blue": "Bleu (RGBA)",
+                "alpha": "Alpha (RGBA)",
+                "magenta": "Magenta (CMJN)",
+                "yellow": "Jaune (CMJN)",
+                "cb": "Cb (YCbCr)",
+                "cr": "Cr (YCbCr)",
+                "cyan": "Cyan (CMJN)",
+                "label": "Ajuster l'image",
+                "description": "Ajuste le canal sélectionné d'une image.",
+                "channel": "Canal",
+                "value_setting": "Valeur",
+                "scale_values": "Valeurs d'échelle",
+                "red": "Rouge (RGBA)",
+                "green": "Vert (RGBA)",
+                "black": "Noir (CMJN)",
+                "hue": "Teinte (HSV)",
+                "saturation": "Saturation (HSV)",
+                "value": "Valeur (HSV)",
+                "luminosity": "Luminosité (LAB)",
+                "a": "A (LAB)",
+                "y": "Y (YCbCr)"
+            },
+            "img_blur": {
+                "label": "Flou de l'image",
+                "blur_type": "Type de flou",
+                "box_type": "Boîte",
+                "description": "Floute la couche sélectionnée.",
+                "blur_radius": "Rayon",
+                "gaussian_type": "Gaussien"
+            },
+            "img_noise": {
+                "label": "Image de bruit",
+                "description": "Ajoute du bruit à la couche sélectionnée.",
+                "gaussian_type": "Gaussien",
+                "size": "Taille du bruit",
+                "noise_amount": "Quantité",
+                "noise_type": "Type de bruit",
+                "salt_and_pepper_type": "Sel et Poivre",
+                "noise_color": "Bruit coloré"
+            }
         },
         "canvasContextMenu": {
             "saveToGalleryGroup": "Enregistrer dans la galerie",
@@ -1825,7 +1980,10 @@
             "newGlobalReferenceImage": "Nouvelle image de référence globale",
             "newControlLayer": "Nouveau couche de contrôle",
             "newInpaintMask": "Nouveau Masque Inpaint",
-            "newRegionalGuidance": "Nouveau Guide Régional"
+            "newRegionalGuidance": "Nouveau Guide Régional",
+            "copyToClipboard": "Copier dans le presse-papiers",
+            "copyBboxToClipboard": "Copier Bbox dans le presse-papiers",
+            "copyCanvasToClipboard": "Copier la Toile dans le presse-papiers"
         },
         "bookmark": "Marque-page pour Changement Rapide",
         "saveLayerToAssets": "Enregistrer la couche dans les ressources",
@@ -1991,7 +2149,10 @@
             "ipAdapterMethod": "Méthode d'IP Adapter",
             "full": "Complet",
             "style": "Style uniquement",
-            "composition": "Composition uniquement"
+            "composition": "Composition uniquement",
+            "fullDesc": "Applique le style visuel (couleurs, textures) et la composition (mise en page, structure).",
+            "styleDesc": "Applique un style visuel (couleurs, textures) sans tenir compte de sa mise en page.",
+            "compositionDesc": "Réplique la mise en page et la structure tout en ignorant le style de la référence."
         },
         "fitBboxToLayers": "Ajuster la bounding box aux calques",
         "regionIsEmpty": "La zone sélectionnée est vide",
@@ -2074,7 +2235,40 @@
         "asRasterLayerResize": "En tant que $t(controlLayers.rasterLayer) (Redimensionner)",
         "asControlLayer": "En tant que $t(controlLayers.controlLayer)",
         "asControlLayerResize": "En $t(controlLayers.controlLayer) (Redimensionner)",
-        "newSession": "Nouvelle session"
+        "newSession": "Nouvelle session",
+        "warnings": {
+            "controlAdapterIncompatibleBaseModel": "modèle de base de la couche de contrôle incompatible",
+            "controlAdapterNoControl": "aucun contrôle sélectionné/dessiné",
+            "rgNoPromptsOrIPAdapters": "pas de textes d'instructions ni d'images de référence",
+            "rgAutoNegativeNotSupported": "Auto-négatif non pris en charge pour le modèle de base sélectionné",
+            "rgNoRegion": "aucune région dessinée",
+            "ipAdapterNoModelSelected": "aucun modèle d'image de référence sélectionné",
+            "rgReferenceImagesNotSupported": "Les images de référence régionales ne sont pas prises en charge pour le modèle de base sélectionné",
+            "problemsFound": "Problèmes trouvés",
+            "unsupportedModel": "couche non prise en charge pour le modèle de base sélectionné",
+            "rgNegativePromptNotSupported": "Prompt négatif non pris en charge pour le modèle de base sélectionné",
+            "ipAdapterIncompatibleBaseModel": "modèle de base d'image de référence incompatible",
+            "controlAdapterNoModelSelected": "aucun modèle de couche de contrôle sélectionné",
+            "ipAdapterNoImageSelected": "Aucune image de référence sélectionnée."
+        },
+        "pasteTo": "Coller vers",
+        "pasteToAssets": "Ressources",
+        "pasteToAssetsDesc": "Coller dans les ressources",
+        "pasteToBbox": "Bbox",
+        "regionCopiedToClipboard": "{{region}} Copié dans le presse-papiers",
+        "copyRegionError": "Erreur de copie {{region}}",
+        "pasteToCanvas": "Toile",
+        "errors": {
+            "unableToFindImage": "Impossible de trouver l'image",
+            "unableToLoadImage": "Impossible de charger l'image"
+        },
+        "referenceImageRegional": "Image de référence (régionale)",
+        "pasteToBboxDesc": "Nouvelle couche (dans Bbox)",
+        "pasteToCanvasDesc": "Nouvelle couche (dans la Toile)",
+        "useImage": "Utiliser l'image",
+        "pastedTo": "Collé à {{destination}}",
+        "referenceImageEmptyState": "<UploadButton>Séléctionner une image</UploadButton> ou faites glisser une image depuis la <GalleryButton>galerie</GalleryButton> sur cette couche pour commencer.",
+        "referenceImageGlobal": "Image de référence (Globale)"
     },
     "upscaling": {
         "exceedsMaxSizeDetails": "La limite maximale d'agrandissement est de {{maxUpscaleDimension}}x{{maxUpscaleDimension}} pixels. Veuillez essayer une image plus petite ou réduire votre sélection d'échelle.",
@@ -2154,7 +2348,8 @@
             "queue": "File d'attente",
             "events": "Événements",
             "metadata": "Métadonnées",
-            "gallery": "Galerie"
+            "gallery": "Galerie",
+            "dnd": "Glisser et déposer"
         },
         "logLevel": {
             "trace": "Trace",
@@ -2171,9 +2366,8 @@
         "toGetStarted": "Pour commencer, saisissez un prompt dans la boîte et cliquez sur <StrongComponent>Invoke</StrongComponent> pour générer votre première image. Sélectionnez un template de prompt pour améliorer les résultats. Vous pouvez choisir de sauvegarder vos images directement dans la <StrongComponent>Galerie</StrongComponent> ou de les modifier sur la <StrongComponent>Toile</StrongComponent>.",
         "gettingStartedSeries": "Vous souhaitez plus de conseils ? Consultez notre <LinkComponent>Série de démarrage</LinkComponent> pour des astuces sur l'exploitation du plein potentiel de l'Invoke Studio.",
         "noModelsInstalled": "Il semble qu'aucun modèle ne soit installé",
-        "downloadStarterModels": "Télécharger les modèles de démarrage",
-        "importModels": "Importer des Modèles",
-        "toGetStartedLocal": "Pour commencer, assurez-vous de télécharger ou d'importer des modèles nécessaires pour exécuter Invoke. Ensuite, saisissez le prompt dans la boîte et cliquez sur <StrongComponent>Invoke</StrongComponent> pour générer votre première image. Sélectionnez un template de prompt pour améliorer les résultats. Vous pouvez choisir de sauvegarder vos images directement sur <StrongComponent>Galerie</StrongComponent> ou les modifier sur la <StrongComponent>Toile</StrongComponent>."
+        "toGetStartedLocal": "Pour commencer, assurez-vous de télécharger ou d'importer des modèles nécessaires pour exécuter Invoke. Ensuite, saisissez le prompt dans la boîte et cliquez sur <StrongComponent>Invoke</StrongComponent> pour générer votre première image. Sélectionnez un template de prompt pour améliorer les résultats. Vous pouvez choisir de sauvegarder vos images directement sur <StrongComponent>Galerie</StrongComponent> ou les modifier sur la <StrongComponent>Toile</StrongComponent>.",
+        "lowVRAMMode": "Pour de meilleures performances, suivez notre <LinkComponent>guide Low VRAM</LinkComponent>."
     },
     "upsell": {
         "shareAccess": "Partager l'accès",
@@ -2221,7 +2415,8 @@
                 "description": "Introduction à l'ajout d'images de référence et IP Adapters globaux."
             },
             "howDoIUseInpaintMasks": {
-                "title": "Comment utiliser les masques d'inpainting ?"
+                "title": "Comment utiliser les masques d'inpainting ?",
+                "description": "Comment appliquer des masques de retourche pour la correction et la variation d'image."
             },
             "creatingYourFirstImage": {
                 "title": "Créer votre première image",
@@ -2230,6 +2425,10 @@
             "understandingImageToImageAndDenoising": {
                 "title": "Comprendre l'Image-à-Image et le Débruitage",
                 "description": "Aperçu des transformations d'image à image et du débruitage dans Invoke."
+            },
+            "howDoIOutpaint": {
+                "title": "Comment effectuer un outpainting ?",
+                "description": "Guide pour l'extension au-delà des bordures de l'image originale."
             }
         },
         "gettingStarted": "Commencer",
@@ -2237,5 +2436,10 @@
         "studioSessionsDesc2": "Rejoignez notre <DiscordLink /> pour participer aux sessions en direct et poser vos questions. Les sessions sont ajoutée dans la playlist la semaine suivante.",
         "supportVideos": "Vidéos d'assistance",
         "controlCanvas": "Contrôler la toile"
+    },
+    "modelCache": {
+        "clear": "Effacer le cache du modèle",
+        "clearSucceeded": "Cache du modèle effacée",
+        "clearFailed": "Problème de nettoyage du cache du modèle"
     }
 }

invokeai/frontend/web/public/locales/it.json

@@ -97,7 +97,19 @@
         "ok": "Ok",
         "generating": "Generazione",
         "loadingModel": "Caricamento del modello",
-        "warnings": "Avvisi"
+        "warnings": "Avvisi",
+        "step": "Passo",
+        "values": "Valori",
+        "start": "Inizio",
+        "end": "Fine",
+        "resetToDefaults": "Ripristina le impostazioni predefinite",
+        "seed": "Seme",
+        "combinatorial": "Combinatorio",
+        "count": "Quantità",
+        "board": "Bacheca",
+        "layout": "Schema",
+        "row": "Riga",
+        "column": "Colonna"
     },
     "gallery": {
         "galleryImageSize": "Dimensione dell'immagine",
@@ -108,7 +120,6 @@
         "deleteImage_many": "Elimina {{count}} immagini",
         "deleteImage_other": "Elimina {{count}} immagini",
         "deleteImagePermanent": "Le immagini eliminate non possono essere ripristinate.",
-        "assets": "Risorse",
         "autoAssignBoardOnClick": "Assegna automaticamente la bacheca al clic",
         "featuresWillReset": "Se elimini questa immagine, quelle funzionalità verranno immediatamente ripristinate.",
         "loading": "Caricamento in corso",
@@ -165,7 +176,9 @@
         "imagesTab": "Immagini create e salvate in Invoke.",
         "assetsTab": "File che hai caricato per usarli nei tuoi progetti.",
         "boardsSettings": "Impostazioni Bacheche",
-        "imagesSettings": "Impostazioni Immagini Galleria"
+        "imagesSettings": "Impostazioni Immagini Galleria",
+        "assets": "Risorse",
+        "images": "Immagini"
     },
     "hotkeys": {
         "searchHotkeys": "Cerca tasti di scelta rapida",
@@ -668,7 +681,7 @@
             "addingImagesTo": "Aggiungi immagini a",
             "systemDisconnected": "Sistema disconnesso",
             "missingNodeTemplate": "Modello di nodo mancante",
-            "missingInputForField": "{{nodeLabel}} -> {{fieldLabel}}: ingresso mancante",
+            "missingInputForField": "ingresso mancante",
             "missingFieldTemplate": "Modello di campo mancante",
             "fluxModelIncompatibleBboxHeight": "$t(parameters.invoke.fluxRequiresDimensionsToBeMultipleOf16), altezza riquadro è {{height}}",
             "fluxModelIncompatibleBboxWidth": "$t(parameters.invoke.fluxRequiresDimensionsToBeMultipleOf16), larghezza riquadro è {{width}}",
@@ -681,11 +694,22 @@
             "canvasIsRasterizing": "La tela è occupata (sta rasterizzando)",
             "canvasIsCompositing": "La tela è occupata (in composizione)",
             "canvasIsFiltering": "La tela è occupata (sta filtrando)",
-            "collectionTooManyItems": "{{nodeLabel}} -> {{fieldLabel}}: troppi elementi, massimo {{maxItems}}",
+            "collectionTooManyItems": "troppi elementi, massimo {{maxItems}}",
             "canvasIsSelectingObject": "La tela è occupata (selezione dell'oggetto)",
-            "collectionTooFewItems": "{{nodeLabel}} -> {{fieldLabel}}: troppi pochi elementi, minimo {{minItems}}",
-            "collectionEmpty": "{{nodeLabel}} -> {{fieldLabel}} raccolta vuota",
-            "fluxModelMultipleControlLoRAs": "È possibile utilizzare solo 1 Controllo LoRA alla volta"
+            "collectionTooFewItems": "troppi pochi elementi, minimo {{minItems}}",
+            "fluxModelMultipleControlLoRAs": "È possibile utilizzare solo 1 Controllo LoRA alla volta",
+            "collectionNumberGTMax": "{{value}} > {{maximum}} (incr max)",
+            "collectionStringTooLong": "troppo lungo, massimo {{maxLength}}",
+            "batchNodeNotConnected": "Nodo Lotto non connesso: {{label}}",
+            "batchNodeEmptyCollection": "Alcuni nodi lotto hanno raccolte vuote",
+            "emptyBatches": "lotti vuoti",
+            "batchNodeCollectionSizeMismatch": "Le dimensioni della raccolta nel Lotto {{batchGroupId}} non corrispondono",
+            "collectionStringTooShort": "troppo corto, minimo {{minLength}}",
+            "collectionNumberNotMultipleOf": "{{value}} non è multiplo di {{multipleOf}}",
+            "collectionNumberLTMin": "{{value}} < {{minimum}} (incr min)",
+            "collectionNumberGTExclusiveMax": "{{value}} >= {{exclusiveMaximum}} (excl max)",
+            "collectionNumberLTExclusiveMin": "{{value}} <= {{exclusiveMinimum}} (excl min)",
+            "collectionEmpty": "raccolta vuota"
         },
         "useCpuNoise": "Usa la CPU per generare rumore",
         "iterations": "Iterazioni",
@@ -813,7 +837,13 @@
         "imagesWillBeAddedTo": "Le immagini caricate verranno aggiunte alle risorse della bacheca {{boardName}}.",
         "uploadFailedInvalidUploadDesc_withCount_one": "Devi caricare al massimo 1 immagine PNG o JPEG.",
         "uploadFailedInvalidUploadDesc_withCount_many": "Devi caricare al massimo {{count}} immagini PNG o JPEG.",
-        "uploadFailedInvalidUploadDesc_withCount_other": "Devi caricare al massimo {{count}} immagini PNG o JPEG."
+        "uploadFailedInvalidUploadDesc_withCount_other": "Devi caricare al massimo {{count}} immagini PNG o JPEG.",
+        "outOfMemoryErrorDescLocal": "Segui la nostra <LinkComponent>guida per bassa VRAM</LinkComponent> per ridurre gli OOM.",
+        "pasteFailed": "Incolla non riuscita",
+        "pasteSuccess": "Incollato su {{destination}}",
+        "unableToCopy": "Impossibile copiare",
+        "unableToCopyDesc": "Il tuo browser non supporta l'accesso agli appunti. Gli utenti di Firefox potrebbero risolvere il problema seguendo ",
+        "unableToCopyDesc_theseSteps": "questi passaggi"
     },
     "accessibility": {
         "invokeProgressBar": "Barra di avanzamento generazione",
@@ -896,8 +926,8 @@
         "boolean": "Booleani",
         "node": "Nodo",
         "collection": "Raccolta",
-        "cannotConnectInputToInput": "Impossibile collegare Input a Input",
-        "cannotConnectOutputToOutput": "Impossibile collegare Output ad Output",
+        "cannotConnectInputToInput": "Impossibile collegare ingresso a ingresso",
+        "cannotConnectOutputToOutput": "Impossibile collegare uscita ad uscita",
         "cannotConnectToSelf": "Impossibile connettersi a se stesso",
         "mismatchedVersion": "Nodo non valido: il nodo {{node}} di tipo {{type}} ha una versione non corrispondente (provare ad aggiornare?)",
         "loadingNodes": "Caricamento nodi...",
@@ -972,7 +1002,39 @@
         "noWorkflows": "Nessun flusso di lavoro",
         "workflowHelpText": "Hai bisogno di aiuto? Consulta la nostra guida <LinkComponent>Introduzione ai flussi di lavoro</LinkComponent>.",
         "specialDesc": "Questa invocazione comporta una gestione speciale nell'applicazione. Ad esempio, i nodi Lotto vengono utilizzati per mettere in coda più grafici da un singolo flusso di lavoro.",
-        "internalDesc": "Questa invocazione è utilizzata internamente da Invoke. Potrebbe subire modifiche significative durante gli aggiornamenti dell'app e potrebbe essere rimossa in qualsiasi momento."
+        "internalDesc": "Questa invocazione è utilizzata internamente da Invoke. Potrebbe subire modifiche significative durante gli aggiornamenti dell'app e potrebbe essere rimossa in qualsiasi momento.",
+        "addItem": "Aggiungi elemento",
+        "generateValues": "Genera valori",
+        "generatorNoValues": "vuoto",
+        "linearDistribution": "Distribuzione lineare",
+        "parseString": "Analizza stringa",
+        "splitOn": "Diviso su",
+        "noBatchGroup": "nessun gruppo",
+        "generatorLoading": "caricamento",
+        "generatorLoadFromFile": "Carica da file",
+        "dynamicPromptsRandom": "Prompt dinamici (casuali)",
+        "dynamicPromptsCombinatorial": "Prompt dinamici (combinatori)",
+        "floatRangeGenerator": "Generatore di intervalli di numeri in virgola mobile",
+        "integerRangeGenerator": "Generatore di intervalli di numeri interi",
+        "uniformRandomDistribution": "Distribuzione casuale uniforme",
+        "generatorNRandomValues_one": "{{count}} valore casuale",
+        "generatorNRandomValues_many": "{{count}} valori casuali",
+        "generatorNRandomValues_other": "{{count}} valori casuali",
+        "arithmeticSequence": "Sequenza aritmetica",
+        "nodeName": "Nome del nodo",
+        "loadWorkflowDesc": "Caricare il flusso di lavoro?",
+        "loadWorkflowDesc2": "Il flusso di lavoro corrente presenta modifiche non salvate.",
+        "downloadWorkflowError": "Errore durante lo scaricamento del flusso di lavoro",
+        "deletedMissingNodeFieldFormElement": "Campo modulo mancante eliminato: nodo {{nodeId}} campo {{fieldName}}",
+        "loadingTemplates": "Caricamento {{name}}",
+        "unableToUpdateNode": "Aggiornamento del nodo non riuscito: nodo {{node}} di tipo {{type}} (potrebbe essere necessario eliminarlo e ricrearlo)",
+        "description": "Descrizione",
+        "generatorImagesCategory": "Categoria",
+        "generatorImages_one": "{{count}} immagine",
+        "generatorImages_many": "{{count}} immagini",
+        "generatorImages_other": "{{count}} immagini",
+        "generatorImagesFromBoard": "Immagini dalla Bacheca",
+        "missingSourceOrTargetNode": "Nodo sorgente o di destinazione mancante"
     },
     "boards": {
         "autoAddBoard": "Aggiungi automaticamente bacheca",
@@ -1094,7 +1156,14 @@
         "generation": "Generazione",
         "other": "Altro",
         "gallery": "Galleria",
-        "batchSize": "Dimensione del lotto"
+        "batchSize": "Dimensione del lotto",
+        "cancelAllExceptCurrentQueueItemAlertDialog2": "Vuoi davvero annullare tutti gli elementi in coda in sospeso?",
+        "confirm": "Conferma",
+        "cancelAllExceptCurrentQueueItemAlertDialog": "L'annullamento di tutti gli elementi della coda, eccetto quello corrente, interromperà gli elementi in sospeso ma consentirà il completamento di quello in corso.",
+        "cancelAllExceptCurrentTooltip": "Annulla tutto tranne l'elemento corrente",
+        "retrySucceeded": "Elemento rieseguito",
+        "retryItem": "Riesegui elemento",
+        "retryFailed": "Problema riesecuzione elemento"
     },
     "models": {
         "noMatchingModels": "Nessun modello corrispondente",
@@ -1138,7 +1207,8 @@
         "dynamicPrompts": "Prompt dinamici",
         "promptsPreview": "Anteprima dei prompt",
         "showDynamicPrompts": "Mostra prompt dinamici",
-        "loading": "Generazione prompt dinamici..."
+        "loading": "Generazione prompt dinamici...",
+        "promptsToGenerate": "Prompt da generare"
     },
     "popovers": {
         "paramScheduler": {
@@ -1671,7 +1741,38 @@
         "download": "Scarica",
         "copyShareLink": "Copia Condividi Link",
         "copyShareLinkForWorkflow": "Copia Condividi Link del Flusso di lavoro",
-        "delete": "Elimina"
+        "delete": "Elimina",
+        "openLibrary": "Apri la libreria",
+        "builder": {
+            "resetAllNodeFields": "Reimposta tutti i campi del nodo",
+            "row": "Riga",
+            "nodeField": "Campo del nodo",
+            "slider": "Cursore",
+            "emptyRootPlaceholderEditMode": "Per iniziare, trascina qui un elemento del modulo o un campo nodo.",
+            "containerPlaceholder": "Contenitore vuoto",
+            "headingPlaceholder": "Titolo vuoto",
+            "column": "Colonna",
+            "nodeFieldTooltip": "Per aggiungere un campo nodo, fare clic sul piccolo pulsante con il segno più sul campo nell'editor del flusso di lavoro, oppure trascinare il campo in base al suo nome nel modulo.",
+            "label": "Etichetta",
+            "deleteAllElements": "Elimina tutti gli elementi del modulo",
+            "addToForm": "Aggiungi al Modulo",
+            "layout": "Schema",
+            "builder": "Generatore Modulo",
+            "zoomToNode": "Zoom sul nodo",
+            "component": "Componente",
+            "showDescription": "Mostra Descrizione",
+            "singleLine": "Linea singola",
+            "multiLine": "Linea Multipla",
+            "both": "Entrambi",
+            "emptyRootPlaceholderViewMode": "Fare clic su Modifica per iniziare a creare un modulo per questo flusso di lavoro.",
+            "textPlaceholder": "Testo vuoto",
+            "workflowBuilderAlphaWarning": "Il generatore di flussi di lavoro è attualmente in versione alpha. Potrebbero esserci cambiamenti radicali prima della versione stabile.",
+            "heading": "Intestazione",
+            "divider": "Divisore",
+            "container": "Contenitore",
+            "text": "Testo",
+            "numberInput": "Ingresso numerico"
+        }
     },
     "accordions": {
         "compositing": {
@@ -1907,7 +2008,43 @@
             },
             "forMoreControl": "Per un maggiore controllo, fare clic su Avanzate qui sotto.",
             "advanced": "Avanzate",
-            "processingLayerWith": "Elaborazione del livello con il filtro {{type}}."
+            "processingLayerWith": "Elaborazione del livello con il filtro {{type}}.",
+            "img_blur": {
+                "label": "Sfoca immagine",
+                "description": "Sfoca il livello selezionato.",
+                "blur_type": "Tipo di sfocatura",
+                "blur_radius": "Raggio",
+                "gaussian_type": "Gaussiana"
+            },
+            "img_noise": {
+                "size": "Dimensione del rumore",
+                "salt_and_pepper_type": "Sale e pepe",
+                "gaussian_type": "Gaussiano",
+                "noise_color": "Rumore colorato",
+                "description": "Aggiunge rumore al livello selezionato.",
+                "noise_type": "Tipo di rumore",
+                "label": "Aggiungi rumore",
+                "noise_amount": "Quantità"
+            },
+            "adjust_image": {
+                "description": "Regola il canale selezionato di un'immagine.",
+                "alpha": "Alfa (RGBA)",
+                "label": "Regola l'immagine",
+                "blue": "Blu (RGBA)",
+                "luminosity": "Luminosità (LAB)",
+                "channel": "Canale",
+                "value_setting": "Valore",
+                "scale_values": "Scala i valori",
+                "red": "Rosso (RGBA)",
+                "green": "Verde (RGBA)",
+                "cyan": "Ciano (CMYK)",
+                "magenta": "Magenta (CMYK)",
+                "yellow": "Giallo (CMYK)",
+                "black": "Nero (CMYK)",
+                "hue": "Tonalità (HSV)",
+                "saturation": "Saturazione (HSV)",
+                "value": "Valore (HSV)"
+            }
         },
         "controlLayers_withCount_hidden": "Livelli di controllo ({{count}} nascosti)",
         "regionalGuidance_withCount_hidden": "Guida regionale ({{count}} nascosti)",
@@ -2014,7 +2151,10 @@
             "saveCanvasToGallery": "Salva la Tela nella Galleria",
             "saveToGalleryGroup": "Salva nella Galleria",
             "newInpaintMask": "Nuova maschera Inpaint",
-            "newRegionalGuidance": "Nuova Guida Regionale"
+            "newRegionalGuidance": "Nuova Guida Regionale",
+            "copyToClipboard": "Copia negli appunti",
+            "copyCanvasToClipboard": "Copia la tela negli appunti",
+            "copyBboxToClipboard": "Copia il riquadro di delimitazione negli appunti"
         },
         "newImg2ImgCanvasFromImage": "Nuova Immagine da immagine",
         "copyRasterLayerTo": "Copia $t(controlLayers.rasterLayer) in",
@@ -2054,7 +2194,7 @@
         "controlLayerEmptyState": "<UploadButton>Carica un'immagine</UploadButton>, trascina un'immagine dalla <GalleryButton>galleria</GalleryButton> su questo livello oppure disegna sulla tela per iniziare.",
         "useImage": "Usa immagine",
         "resetGenerationSettings": "Ripristina impostazioni di generazione",
-        "referenceImageEmptyState": "Per iniziare, <UploadButton>carica un'immagine</UploadButton> oppure trascina un'immagine dalla <GalleryButton>galleria</GalleryButton> su questo livello.",
+        "referenceImageEmptyState": "Per iniziare, <UploadButton>carica un'immagine</UploadButton>, trascina un'immagine dalla <GalleryButton>galleria</GalleryButton>, oppure <PullBboxButton>trascina il riquadro di delimitazione in questo livello</PullBboxButton> su questo livello.",
         "asRasterLayer": "Come $t(controlLayers.rasterLayer)",
         "asRasterLayerResize": "Come $t(controlLayers.rasterLayer) (Ridimensiona)",
         "asControlLayer": "Come $t(controlLayers.controlLayer)",
@@ -2077,6 +2217,20 @@
             "ipAdapterIncompatibleBaseModel": "modello base dell'immagine di riferimento incompatibile",
             "ipAdapterNoImageSelected": "nessuna immagine di riferimento selezionata",
             "rgAutoNegativeNotSupported": "Auto-Negativo non supportato per il modello base selezionato"
+        },
+        "pasteTo": "Incolla su",
+        "pasteToBboxDesc": "Nuovo livello (nel riquadro di delimitazione)",
+        "pasteToAssets": "Risorse",
+        "copyRegionError": "Errore durante la copia di {{region}}",
+        "pasteToAssetsDesc": "Incolla in Risorse",
+        "pasteToBbox": "Riquadro di delimitazione",
+        "pasteToCanvas": "Tela",
+        "pasteToCanvasDesc": "Nuovo livello (nella Tela)",
+        "pastedTo": "Incollato su {{destination}}",
+        "regionCopiedToClipboard": "{{region}} Copiato negli appunti",
+        "errors": {
+            "unableToFindImage": "Impossibile trovare l'immagine",
+            "unableToLoadImage": "Impossibile caricare l'immagine"
         }
     },
     "ui": {
@@ -2166,10 +2320,9 @@
     "newUserExperience": {
         "gettingStartedSeries": "Desideri maggiori informazioni? Consulta la nostra <LinkComponent>Getting Started Series</LinkComponent> per suggerimenti su come sfruttare appieno il potenziale di Invoke Studio.",
         "toGetStarted": "Per iniziare, inserisci un prompt nella casella e fai clic su <StrongComponent>Invoke</StrongComponent> per generare la tua prima immagine. Seleziona un modello di prompt per migliorare i risultati. Puoi scegliere di salvare le tue immagini direttamente nella <StrongComponent>Galleria</StrongComponent> o modificarle nella <StrongComponent>Tela</StrongComponent>.",
-        "importModels": "Importa modelli",
-        "downloadStarterModels": "Scarica i modelli per iniziare",
-        "noModelsInstalled": "Sembra che tu non abbia installato alcun modello",
-        "toGetStartedLocal": "Per iniziare, assicurati di scaricare o importare i modelli necessari per eseguire Invoke. Quindi, inserisci un prompt nella casella e fai clic su <StrongComponent>Invoke</StrongComponent> per generare la tua prima immagine. Seleziona un modello di prompt per migliorare i risultati. Puoi scegliere di salvare le tue immagini direttamente nella <StrongComponent>Galleria</StrongComponent> o modificarle nella <StrongComponent>Tela</StrongComponent>."
+        "noModelsInstalled": "Sembra che non hai installato alcun modello! Puoi <DownloadStarterModelsButton>scaricare un pacchetto di modelli di avvio</DownloadStarterModelsButton> o <ImportModelsButton>importare modelli</ImportModelsButton>.",
+        "toGetStartedLocal": "Per iniziare, assicurati di scaricare o importare i modelli necessari per eseguire Invoke. Quindi, inserisci un prompt nella casella e fai clic su <StrongComponent>Invoke</StrongComponent> per generare la tua prima immagine. Seleziona un modello di prompt per migliorare i risultati. Puoi scegliere di salvare le tue immagini direttamente nella <StrongComponent>Galleria</StrongComponent> o modificarle nella <StrongComponent>Tela</StrongComponent>.",
+        "lowVRAMMode": "Per prestazioni ottimali, segui la nostra <LinkComponent>guida per bassa VRAM</LinkComponent>."
     },
     "whatsNew": {
         "whatsNewInInvoke": "Novità in Invoke",
@@ -2177,7 +2330,8 @@
         "watchRecentReleaseVideos": "Guarda i video su questa versione",
         "watchUiUpdatesOverview": "Guarda le novità dell'interfaccia",
         "items": [
-            "<StrongComponent>Livelli di controllo Flux</StrongComponent>: nuovi modelli di controllo per il rilevamento dei bordi e la mappatura della profondità sono ora supportati per i modelli di Flux dev."
+            "Editor del flusso di lavoro: nuovo generatore di moduli trascina-e-rilascia per una creazione più facile del flusso di lavoro.",
+            "Altri miglioramenti: messa in coda dei lotti più rapida, migliore ampliamento, selettore colore migliorato e nodi metadati."
         ]
     },
     "system": {
@@ -2267,5 +2421,10 @@
         "watch": "Guarda",
         "studioSessionsDesc1": "Dai un'occhiata a <StudioSessionsPlaylistLink /> per approfondimenti su Invoke.",
         "studioSessionsDesc2": "Unisciti al nostro <DiscordLink /> per partecipare alle sessioni live e fare domande. Le sessioni vengono caricate sulla playlist la settimana successiva."
+    },
+    "modelCache": {
+        "clear": "Cancella la cache del modello",
+        "clearSucceeded": "Cache del modello cancellata",
+        "clearFailed": "Problema durante la cancellazione della cache del modello"
     }
 }

invokeai/frontend/web/public/locales/ja.json

@@ -32,29 +32,29 @@
         "learnMore": "もっと学ぶ",
         "random": "ランダム",
         "batch": "バッチマネージャー",
-        "advanced": "高度な設定",
+        "advanced": "高度",
         "created": "作成済",
         "green": "緑",
         "blue": "青",
         "alpha": "アルファ",
-        "outpaint": "アウトペイント",
+        "outpaint": "outpaint",
         "unknown": "不明",
         "updated": "更新済",
         "add": "追加",
-        "ai": "AI",
+        "ai": "ai",
         "copyError": "$t(gallery.copy) エラー",
         "data": "データ",
         "template": "テンプレート",
         "red": "赤",
         "or": "または",
-        "checkpoint": "チェックポイント",
+        "checkpoint": "Checkpoint",
         "direction": "方向",
         "simple": "シンプル",
         "save": "保存",
         "saveAs": "名前をつけて保存",
         "somethingWentWrong": "何かの問題が発生しました",
         "details": "詳細",
-        "inpaint": "インペイント",
+        "inpaint": "inpaint",
         "delete": "削除",
         "nextPage": "次のページ",
         "copy": "コピー",
@@ -70,12 +70,12 @@
         "unknownError": "未知のエラー",
         "orderBy": "並び順:",
         "enabled": "有効",
-        "notInstalled": "未インストール",
+        "notInstalled": "未 $t(common.installed)",
         "positivePrompt": "ポジティブプロンプト",
         "negativePrompt": "ネガティブプロンプト",
         "selected": "選択済み",
         "aboutDesc": "Invokeを業務で利用する場合はマークしてください:",
-        "beta": "ベータ",
+        "beta": "Beta",
         "disabled": "無効",
         "editor": "エディタ",
         "safetensors": "Safetensors",
@@ -93,7 +93,27 @@
         "reset": "リセット",
         "none": "なし",
         "new": "新規",
-        "close": "閉じる"
+        "close": "閉じる",
+        "warnings": "警告",
+        "dontShowMeThese": "次回から表示しない",
+        "goTo": "移動",
+        "generating": "生成中",
+        "loadingModel": "モデルをロード中",
+        "layout": "レイアウト",
+        "step": "ステップ",
+        "start": "開始",
+        "count": "回数",
+        "end": "終了",
+        "min": "最小",
+        "max": "最大",
+        "values": "値",
+        "resetToDefaults": "デフォルトに戻す",
+        "row": "行",
+        "column": "列",
+        "board": "ボード",
+        "seed": "シード",
+        "combinatorial": "組み合わせ",
+        "aboutHeading": "想像力をこの手に"
     },
     "gallery": {
         "galleryImageSize": "画像のサイズ",
@@ -106,11 +126,10 @@
         "featuresWillReset": "この画像を削除すると、これらの機能は即座にリセットされます。",
         "unstarImage": "スターを外す",
         "loading": "ロード中",
-        "assets": "アセット",
         "currentlyInUse": "この画像は現在下記の機能を使用しています:",
         "drop": "ドロップ",
         "dropOrUpload": "$t(gallery.drop) またはアップロード",
-        "deleteImage_other": "画像を削除",
+        "deleteImage_other": "画像 {{count}} 枚を削除",
         "deleteImagePermanent": "削除された画像は復元できません。",
         "download": "ダウンロード",
         "unableToLoad": "ギャラリーをロードできません",
@@ -156,7 +175,12 @@
         "displayBoardSearch": "ボード検索",
         "displaySearch": "画像を検索",
         "boardsSettings": "ボード設定",
-        "imagesSettings": "ギャラリー画像設定"
+        "imagesSettings": "ギャラリー画像設定",
+        "selectAllOnPage": "ページ上のすべてを選択",
+        "images": "画像",
+        "assetsTab": "プロジェクトで使用するためにアップロードされたファイル。",
+        "imagesTab": "Invoke内で作成および保存された画像。",
+        "assets": "アセット"
     },
     "hotkeys": {
         "searchHotkeys": "ホットキーを検索",
@@ -181,44 +205,121 @@
         },
         "canvas": {
             "redo": {
-                "title": "やり直し"
+                "title": "やり直し",
+                "desc": "最後のキャンバス操作をやり直します。"
             },
             "transformSelected": {
-                "title": "変形"
+                "title": "変形",
+                "desc": "選択したレイヤーを変形します。"
             },
             "undo": {
-                "title": "取り消し"
+                "title": "取り消し",
+                "desc": "最後のキャンバス操作を取り消します。"
             },
             "selectEraserTool": {
-                "title": "消しゴムツール"
+                "title": "消しゴムツール",
+                "desc": "消しゴムツールを選択します。"
             },
             "cancelTransform": {
-                "title": "変形をキャンセル"
+                "title": "変形をキャンセル",
+                "desc": "保留中の変形をキャンセルします。"
             },
             "resetSelected": {
-                "title": "レイヤーをリセット"
+                "title": "レイヤーをリセット",
+                "desc": "選択したレイヤーをリセットします。この操作はInpaint MaskおよびRegional Guidanceにのみ適用されます。"
             },
             "applyTransform": {
-                "title": "変形を適用"
+                "title": "変形を適用",
+                "desc": "保留中の変形を選択したレイヤーに適用します。"
             },
             "selectColorPickerTool": {
-                "title": "スポイトツール"
+                "title": "スポイトツール",
+                "desc": "スポイトツールを選択します。"
             },
             "fitBboxToCanvas": {
-                "title": "バウンディングボックスをキャンバスにフィット"
+                "title": "バウンディングボックスをキャンバスにフィット",
+                "desc": "バウンディングボックスがキャンバスに収まるように表示を拡大、位置調整します。"
             },
             "selectBrushTool": {
-                "title": "ブラシツール"
+                "title": "ブラシツール",
+                "desc": "ブラシツールを選択します。"
             },
             "selectMoveTool": {
-                "title": "移動ツール"
+                "title": "移動ツール",
+                "desc": "移動ツールを選択します。"
             },
             "selectBboxTool": {
-                "title": "バウンディングボックスツール"
+                "title": "バウンディングボックスツール",
+                "desc": "バウンディングボックスツールを選択します。"
             },
             "title": "キャンバス",
             "fitLayersToCanvas": {
-                "title": "レイヤーをキャンバスにフィット"
+                "title": "レイヤーをキャンバスにフィット",
+                "desc": "すべての表示レイヤーがキャンバスに収まるように表示を拡大、位置調整します。"
+            },
+            "setZoomTo400Percent": {
+                "desc": "キャンバスのズームを400%に設定します。",
+                "title": "400%にズーム"
+            },
+            "setZoomTo800Percent": {
+                "title": "800%にズーム",
+                "desc": "キャンバスのズームを800%に設定します。"
+            },
+            "quickSwitch": {
+                "title": "レイヤーのクイックスイッチ",
+                "desc": "最後に選択した2つのレイヤー間を切り替えます。レイヤーがブックマークされている場合、常にそのレイヤーと最後に選択したブックマークされていないレイヤーの間を切り替えます。"
+            },
+            "nextEntity": {
+                "title": "次のレイヤー",
+                "desc": "リスト内の次のレイヤーを選択します。"
+            },
+            "filterSelected": {
+                "title": "フィルター",
+                "desc": "選択したレイヤーをフィルターします。RasterおよびControlレイヤーにのみ適用されます。"
+            },
+            "prevEntity": {
+                "desc": "リスト内の前のレイヤーを選択します。",
+                "title": "前のレイヤー"
+            },
+            "setFillToWhite": {
+                "title": "ツール色を白に設定",
+                "desc": "現在のツールの色を白色に設定します。"
+            },
+            "selectViewTool": {
+                "title": "表示ツール",
+                "desc": "表示ツールを選択します。"
+            },
+            "setZoomTo100Percent": {
+                "title": "100%にズーム",
+                "desc": "キャンバスのズームを100%に設定します。"
+            },
+            "deleteSelected": {
+                "desc": "選択したレイヤーを削除します。",
+                "title": "レイヤーを削除"
+            },
+            "cancelFilter": {
+                "desc": "保留中のフィルターをキャンセルします。",
+                "title": "フィルターをキャンセル"
+            },
+            "applyFilter": {
+                "title": "フィルターを適用",
+                "desc": "保留中のフィルターを選択したレイヤーに適用します。"
+            },
+            "setZoomTo200Percent": {
+                "title": "200%にズーム",
+                "desc": "キャンバスのズームを200%に設定します。"
+            },
+            "decrementToolWidth": {
+                "title": "ツール幅を縮小する",
+                "desc": "選択中のブラシまたは消しゴムツールの幅を減少させます。"
+            },
+            "incrementToolWidth": {
+                "desc": "選択中のブラシまたは消しゴムツールの幅を増加させます。",
+                "title": "ツール幅を増加する"
+            },
+            "selectRectTool": {
+                "title": "矩形ツール",
+                "desc": "矩形ツールを選択します。"
             }
         },
         "workflows": {
@@ -227,6 +328,13 @@
             },
             "redo": {
                 "title": "やり直し"
+            },
+            "title": "ワークフロー",
+            "pasteSelection": {
+                "title": "ペースト"
+            },
+            "copySelection": {
+                "title": "コピー"
             }
         },
         "app": {
@@ -236,16 +344,62 @@
             },
             "title": "アプリケーション",
             "invoke": {
-                "title": "Invoke"
+                "title": "生成",
+                "desc": "生成をキューに追加し、キューの末尾に加えます。"
             },
             "cancelQueueItem": {
-                "title": "キャンセル"
+                "title": "キャンセル",
+                "desc": "現在処理中のキュー項目をキャンセルします。"
             },
             "clearQueue": {
-                "title": "キューをクリア"
+                "title": "キューをクリア",
+                "desc": "すべてのキュー項目をキャンセルして消去します。"
+            },
+            "selectCanvasTab": {
+                "desc": "キャンバスタブを選択します。",
+                "title": "キャンバスタブを選択"
+            },
+            "selectUpscalingTab": {
+                "desc": "アップスケーリングタブを選択します。",
+                "title": "アップスケーリングタブを選択"
+            },
+            "toggleRightPanel": {
+                "desc": "右パネルを表示または非表示。",
+                "title": "右パネルをトグル"
+            },
+            "selectModelsTab": {
+                "title": "モデルタブを選択",
+                "desc": "モデルタブを選択します。"
+            },
+            "invokeFront": {
+                "desc": "生成をキューに追加し、キューの先頭に加えます。",
+                "title": "生成（先頭）"
+            },
+            "resetPanelLayout": {
+                "title": "パネルレイアウトをリセット",
+                "desc": "左パネルと右パネルをデフォルトのサイズとレイアウトにリセットします。"
+            },
+            "togglePanels": {
+                "desc": "左パネルと右パネルを合わせて表示または非表示。",
+                "title": "パネルをトグル"
+            },
+            "selectWorkflowsTab": {
+                "desc": "ワークフロータブを選択します。",
+                "title": "ワークフロータブを選択"
+            },
+            "selectQueueTab": {
+                "title": "キュータブを選択",
+                "desc": "キュータブを選択します。"
+            },
+            "focusPrompt": {
+                "title": "プロンプトにフォーカス",
+                "desc": "カーソルをポジティブプロンプト欄に移動します。"
             }
         },
-        "hotkeys": "ホットキー"
+        "hotkeys": "ホットキー",
+        "gallery": {
+            "title": "ギャラリー"
+        }
     },
     "modelManager": {
         "modelManager": "モデルマネージャ",
@@ -256,13 +410,13 @@
         "name": "名前",
         "description": "概要",
         "config": "コンフィグ",
-        "repo_id": "Repo ID",
+        "repo_id": "リポジトリID",
         "width": "幅",
         "height": "高さ",
         "addModel": "モデルを追加",
         "availableModels": "モデルを有効化",
         "search": "検索",
-        "load": "Load",
+        "load": "ロード",
         "active": "active",
         "selected": "選択済",
         "delete": "削除",
@@ -282,7 +436,7 @@
         "modelConverted": "モデル変換が完了しました",
         "predictionType": "予測タイプ（SD 2.x モデルおよび一部のSD 1.x モデル用）",
         "selectModel": "モデルを選択",
-        "advanced": "高度な設定",
+        "advanced": "高度",
         "modelDeleted": "モデルが削除されました",
         "convertToDiffusersHelpText2": "このプロセスでは、モデルマネージャーのエントリーを同じモデルのディフューザーバージョンに置き換えます。",
         "modelUpdateFailed": "モデル更新が失敗しました",
@@ -295,7 +449,20 @@
         "convertToDiffusersHelpText4": "これは一回限りのプロセスです。コンピュータの仕様によっては、約30秒から60秒かかる可能性があります。",
         "cancel": "キャンセル",
         "uploadImage": "画像をアップロード",
-        "addModels": "モデルを追加"
+        "addModels": "モデルを追加",
+        "modelName": "モデル名",
+        "source": "ソース",
+        "path": "パス",
+        "modelSettings": "モデル設定",
+        "vae": "VAE",
+        "huggingFace": "HuggingFace",
+        "huggingFaceRepoID": "HuggingFace リポジトリID",
+        "metadata": "メタデータ",
+        "loraModels": "LoRA",
+        "edit": "編集",
+        "install": "インストール",
+        "huggingFacePlaceholder": "owner/model-name",
+        "variant": "Variant"
     },
     "parameters": {
         "images": "画像",
@@ -306,7 +473,7 @@
         "shuffle": "シャッフル",
         "strength": "強度",
         "upscaling": "アップスケーリング",
-        "scale": "Scale",
+        "scale": "スケール",
         "scaleBeforeProcessing": "処理前のスケール",
         "scaledWidth": "幅のスケール",
         "scaledHeight": "高さのスケール",
@@ -315,7 +482,7 @@
         "useSeed": "シード値を使用",
         "useAll": "すべてを使用",
         "info": "情報",
-        "showOptionsPanel": "オプションパネルを表示",
+        "showOptionsPanel": "サイドパネルを表示 (O or T)",
         "iterations": "生成回数",
         "general": "基本設定",
         "setToOptimalSize": "サイズをモデルに最適化",
@@ -329,16 +496,29 @@
         "useSize": "サイズを使用",
         "postProcessing": "ポストプロセス (Shift + U)",
         "denoisingStrength": "ノイズ除去強度",
-        "recallMetadata": "メタデータを再使用"
+        "recallMetadata": "メタデータを再使用",
+        "copyImage": "画像をコピー",
+        "positivePromptPlaceholder": "ポジティブプロンプト",
+        "negativePromptPlaceholder": "ネガティブプロンプト",
+        "type": "タイプ",
+        "cancel": {
+            "cancel": "キャンセル"
+        },
+        "cfgScale": "CFGスケール",
+        "tileSize": "タイルサイズ",
+        "coherenceMode": "モード"
     },
     "settings": {
         "models": "モデル",
-        "displayInProgress": "生成中の画像を表示する",
+        "displayInProgress": "生成中の画像を表示",
         "confirmOnDelete": "削除時に確認",
         "resetWebUI": "WebUIをリセット",
         "resetWebUIDesc1": "WebUIのリセットは、画像と保存された設定のキャッシュをリセットするだけです。画像を削除するわけではありません。",
         "resetWebUIDesc2": "もしギャラリーに画像が表示されないなど、何か問題が発生した場合はGitHubにissueを提出する前にリセットを試してください。",
-        "resetComplete": "WebUIはリセットされました。F5を押して再読み込みしてください。"
+        "resetComplete": "WebUIはリセットされました。",
+        "ui": "ユーザーインターフェイス",
+        "beta": "ベータ",
+        "developer": "開発者"
     },
     "toast": {
         "uploadFailed": "アップロード失敗",
@@ -346,7 +526,8 @@
         "imageUploadFailed": "画像のアップロードに失敗しました",
         "uploadFailedInvalidUploadDesc": "画像はPNGかJPGである必要があります。",
         "sentToUpscale": "アップスケーラーに転送しました",
-        "imageUploaded": "画像をアップロードしました"
+        "imageUploaded": "画像をアップロードしました",
+        "serverError": "サーバーエラー"
     },
     "accessibility": {
         "invokeProgressBar": "進捗バー",
@@ -357,7 +538,7 @@
         "menu": "メニュー",
         "createIssue": "問題を報告",
         "resetUI": "$t(accessibility.reset) UI",
-        "mode": "モード:",
+        "mode": "モード",
         "about": "Invoke について",
         "submitSupportTicket": "サポート依頼を送信する",
         "uploadImages": "画像をアップロード",
@@ -374,7 +555,20 @@
         "positivePrompt": "ポジティブプロンプト",
         "strength": "Image to Image 強度",
         "recallParameters": "パラメータを再使用",
-        "recallParameter": "{{label}} を再使用"
+        "recallParameter": "{{label}} を再使用",
+        "imageDimensions": "画像サイズ",
+        "imageDetails": "画像の詳細",
+        "model": "モデル",
+        "allPrompts": "すべてのプロンプト",
+        "cfgScale": "CFGスケール",
+        "createdBy": "作成:",
+        "metadata": "メタデータ",
+        "height": "高さ",
+        "negativePrompt": "ネガティブプロンプト",
+        "generationMode": "生成モード",
+        "vae": "VAE",
+        "cfgRescaleMultiplier": "$t(parameters.cfgRescaleMultiplier)",
+        "canvasV2Metadata": "キャンバス"
     },
     "queue": {
         "queueEmpty": "キューが空です",
@@ -406,7 +600,7 @@
         "batchQueuedDesc_other": "{{count}} セッションをキューの{{direction}}に追加しました",
         "graphQueued": "グラフをキューに追加しました",
         "batch": "バッチ",
-        "clearQueueAlertDialog": "キューをクリアすると、処理中のアイテムは直ちにキャンセルされ、キューは完全にクリアされます。",
+        "clearQueueAlertDialog": "キューをクリアすると、処理中の項目は直ちにキャンセルされ、キューは完全にクリアされます。保留中のフィルターもキャンセルされます。",
         "pending": "保留中",
         "resumeFailed": "処理の再開に問題があります",
         "clear": "クリア",
@@ -424,7 +618,7 @@
         "enqueueing": "バッチをキューに追加",
         "cancelBatchFailed": "バッチのキャンセルに問題があります",
         "clearQueueAlertDialog2": "キューをクリアしてもよろしいですか？",
-        "item": "アイテム",
+        "item": "項目",
         "graphFailedToQueue": "グラフをキューに追加できませんでした",
         "batchFieldValues": "バッチの詳細",
         "openQueue": "キューを開く",
@@ -440,7 +634,17 @@
         "upscaling": "アップスケール",
         "generation": "生成",
         "other": "その他",
-        "gallery": "ギャラリー"
+        "gallery": "ギャラリー",
+        "cancelAllExceptCurrentQueueItemAlertDialog2": "すべての保留中のキュー項目をキャンセルしてもよいですか？",
+        "cancelAllExceptCurrentTooltip": "現在の項目を除いてすべてキャンセル",
+        "origin": "先頭",
+        "destination": "宛先",
+        "confirm": "確認",
+        "retryItem": "項目をリトライ",
+        "batchSize": "バッチサイズ",
+        "retryFailed": "項目のリトライに問題があります",
+        "cancelAllExceptCurrentQueueItemAlertDialog": "現在の項目を除くすべてのキュー項目をキャンセルすると、保留中の項目は停止しますが、進行中の項目は完了します。",
+        "retrySucceeded": "項目がリトライされました"
     },
     "models": {
         "noMatchingModels": "一致するモデルがありません",
@@ -449,13 +653,14 @@
         "noModelsAvailable": "使用可能なモデルがありません",
         "selectModel": "モデルを選択してください",
         "concepts": "コンセプト",
-        "addLora": "LoRAを追加"
+        "addLora": "LoRAを追加",
+        "lora": "LoRA"
     },
     "nodes": {
         "addNode": "ノードを追加",
         "boolean": "ブーリアン",
         "addNodeToolTip": "ノードを追加 (Shift+A, Space)",
-        "missingTemplate": "テンプレートが見つかりません",
+        "missingTemplate": "Invalid node: タイプ {{type}} のノード {{node}} にテンプレートがありません（未インストール？）",
         "loadWorkflow": "ワークフローを読み込み",
         "hideLegendNodes": "フィールドタイプの凡例を非表示",
         "float": "浮動小数点",
@@ -466,7 +671,7 @@
         "currentImageDescription": "ノードエディタ内の現在の画像を表示",
         "downloadWorkflow": "ワークフローのJSONをダウンロード",
         "fieldTypesMustMatch": "フィールドタイプが一致している必要があります",
-        "edge": "輪郭",
+        "edge": "エッジ",
         "animatedEdgesHelp": "選択したエッジおよび選択したノードに接続されたエッジをアニメーション化します",
         "cannotDuplicateConnection": "重複した接続は作れません",
         "noWorkflow": "ワークフローがありません",
@@ -485,7 +690,20 @@
         "cannotConnectToSelf": "自身のノードには接続できません",
         "colorCodeEdges": "カラー-Code Edges",
         "loadingNodes": "ノードを読み込み中...",
-        "scheduler": "スケジューラー"
+        "scheduler": "スケジューラー",
+        "version": "バージョン",
+        "edit": "編集",
+        "nodeVersion": "ノードバージョン",
+        "workflowTags": "タグ",
+        "string": "文字列",
+        "workflowVersion": "バージョン",
+        "workflowAuthor": "作者",
+        "ipAdapter": "IP-Adapter",
+        "notes": "ノート",
+        "workflow": "ワークフロー",
+        "workflowName": "名前",
+        "workflowNotes": "ノート",
+        "enum": "Enum"
     },
     "boards": {
         "autoAddBoard": "自動追加するボード",
@@ -507,7 +725,7 @@
         "deleteBoard": "ボードの削除",
         "deleteBoardAndImages": "ボードと画像の削除",
         "deleteBoardOnly": "ボードのみ削除",
-        "deletedBoardsCannotbeRestored": "削除されたボードは復元できません",
+        "deletedBoardsCannotbeRestored": "削除されたボードは復元できません。\"ボードのみ削除\"を選択すると画像は未分類に移動されます。",
         "movingImagesToBoard_other": "{{count}} の画像をボードに移動:",
         "hideBoards": "ボードを隠す",
         "assetsWithCount_other": "{{count}} のアセット",
@@ -519,7 +737,12 @@
         "archiveBoard": "ボードをアーカイブ",
         "archived": "アーカイブ完了",
         "unarchiveBoard": "アーカイブされていないボード",
-        "imagesWithCount_other": "{{count}} の画像"
+        "imagesWithCount_other": "{{count}} の画像",
+        "updateBoardError": "ボード更新エラー",
+        "selectedForAutoAdd": "自動追加に選択済み",
+        "deletedPrivateBoardsCannotbeRestored": "削除されたボードは復元できません。\"ボードのみ削除\"を選択すると画像はその作成者のプライベートな未分類に移動されます。",
+        "noBoards": "{{boardType}} ボードがありません",
+        "viewBoards": "ボードを表示"
     },
     "invocationCache": {
         "invocationCache": "呼び出しキャッシュ",
@@ -571,6 +794,57 @@
         },
         "paramAspect": {
             "heading": "縦横比"
+        },
+        "refinerSteps": {
+            "heading": "ステップ"
+        },
+        "paramVAE": {
+            "heading": "VAE"
+        },
+        "scale": {
+            "heading": "スケール"
+        },
+        "refinerScheduler": {
+            "heading": "スケジューラー"
+        },
+        "compositingCoherenceMode": {
+            "heading": "モード"
+        },
+        "paramModel": {
+            "heading": "モデル"
+        },
+        "paramHeight": {
+            "heading": "高さ"
+        },
+        "paramSteps": {
+            "heading": "ステップ"
+        },
+        "ipAdapterMethod": {
+            "heading": "モード"
+        },
+        "paramSeed": {
+            "heading": "シード"
+        },
+        "paramIterations": {
+            "heading": "生成回数"
+        },
+        "controlNet": {
+            "heading": "ControlNet"
+        },
+        "paramWidth": {
+            "heading": "幅"
+        },
+        "lora": {
+            "heading": "LoRA"
+        },
+        "loraWeight": {
+            "heading": "重み"
+        },
+        "patchmatchDownScaleSize": {
+            "heading": "Downscale"
+        },
+        "controlNetWeight": {
+            "heading": "重み"
         }
     },
     "accordions": {
@@ -580,7 +854,8 @@
             "coherenceTab": "コヒーレンスパス"
         },
         "advanced": {
-            "title": "高度な設定"
+            "title": "高度",
+            "options": "$t(accordions.advanced.title) オプション"
         },
         "control": {
             "title": "コントロール"
@@ -609,7 +884,11 @@
     },
     "ui": {
         "tabs": {
-            "queue": "キュー"
+            "queue": "キュー",
+            "canvas": "キャンバス",
+            "workflows": "ワークフロー",
+            "models": "モデル",
+            "gallery": "ギャラリー"
         }
     },
     "controlLayers": {
@@ -624,7 +903,8 @@
             "bboxGroup": "バウンディングボックスから作成",
             "cropCanvasToBbox": "キャンバスをバウンディングボックスでクロップ",
             "newGlobalReferenceImage": "新規全域参照画像",
-            "newRegionalReferenceImage": "新規領域参照画像"
+            "newRegionalReferenceImage": "新規領域参照画像",
+            "canvasGroup": "キャンバス"
         },
         "regionalGuidance": "領域ガイダンス",
         "globalReferenceImage": "全域参照画像",
@@ -645,7 +925,8 @@
             "brush": "ブラシ",
             "rectangle": "矩形",
             "move": "移動",
-            "eraser": "消しゴム"
+            "eraser": "消しゴム",
+            "bbox": "Bbox"
         },
         "saveCanvasToGallery": "キャンバスをギャラリーに保存",
         "saveBboxToGallery": "バウンディングボックスをギャラリーへ保存",
@@ -663,7 +944,27 @@
         "canvas": "キャンバス",
         "fitBboxToLayers": "バウンディングボックスをレイヤーにフィット",
         "removeBookmark": "ブックマークを外す",
-        "savedToGalleryOk": "ギャラリーに保存しました"
+        "savedToGalleryOk": "ギャラリーに保存しました",
+        "controlMode": {
+            "prompt": "プロンプト"
+        },
+        "prompt": "プロンプト",
+        "settings": {
+            "snapToGrid": {
+                "off": "オフ",
+                "on": "オン"
+            }
+        },
+        "filter": {
+            "filter": "フィルター",
+            "spandrel_filter": {
+                "model": "モデル"
+            },
+            "apply": "適用",
+            "reset": "リセット",
+            "cancel": "キャンセル"
+        },
+        "weight": "重み"
     },
     "stylePresets": {
         "clearTemplateSelection": "選択したテンプレートをクリア",
@@ -675,15 +976,54 @@
         "createPromptTemplate": "プロンプトテンプレートを作成",
         "promptTemplateCleared": "プロンプトテンプレートをクリアしました",
         "searchByName": "名前で検索",
-        "toggleViewMode": "表示モードを切り替え"
+        "toggleViewMode": "表示モードを切り替え",
+        "negativePromptColumn": "'negative_prompt'",
+        "preview": "プレビュー",
+        "nameColumn": "'name'",
+        "type": "タイプ",
+        "private": "プライベート",
+        "name": "名称"
     },
     "upscaling": {
         "upscaleModel": "アップスケールモデル",
         "postProcessingModel": "ポストプロセスモデル",
-        "upscale": "アップスケール"
+        "upscale": "アップスケール",
+        "scale": "スケール"
     },
     "sdxl": {
         "denoisingStrength": "ノイズ除去強度",
-        "scheduler": "スケジューラー"
+        "scheduler": "スケジューラー",
+        "loading": "ロード中...",
+        "steps": "ステップ",
+        "refiner": "Refiner"
+    },
+    "modelCache": {
+        "clear": "モデルキャッシュを消去",
+        "clearSucceeded": "モデルキャッシュを消去しました",
+        "clearFailed": "モデルキャッシュの消去中に問題が発生"
+    },
+    "workflows": {
+        "workflows": "ワークフロー",
+        "ascending": "昇順",
+        "name": "名前",
+        "descending": "降順"
+    },
+    "system": {
+        "logNamespaces": {
+            "system": "システム",
+            "gallery": "ギャラリー",
+            "workflows": "ワークフロー",
+            "models": "モデル",
+            "canvas": "キャンバス",
+            "metadata": "メタデータ",
+            "queue": "キュー"
+        },
+        "logLevel": {
+            "debug": "Debug",
+            "info": "Info",
+            "error": "Error",
+            "fatal": "Fatal",
+            "warn": "Warn"
+        }
     }
 }