Enable LoRAPatcher.apply_smart_lora_patches(...) throughout the stack.

Currently translated at 99.3% (1633 of 1643 strings)

Co-authored-by: Riccardo Giovanetti <riccardo.giovanetti@gmail.com>
Translate-URL: https://hosted.weblate.org/projects/invokeai/web-ui/it/
Translation: InvokeAI/Web UI

docker/Dockerfile

@@ -2,29 +2,42 @@
 
 ## Builder stage
 
-FROM library/ubuntu:23.04 AS builder
+FROM library/ubuntu:24.04 AS builder
 
 ARG DEBIAN_FRONTEND=noninteractive
 RUN rm -f /etc/apt/apt.conf.d/docker-clean; echo 'Binary::apt::APT::Keep-Downloaded-Packages "true";' > /etc/apt/apt.conf.d/keep-cache
 RUN --mount=type=cache,target=/var/cache/apt,sharing=locked \
     --mount=type=cache,target=/var/lib/apt,sharing=locked \
     apt update && apt-get install -y \
-        git \
-        python3-venv \
-        python3-pip \
-        build-essential
+        build-essential \
+        git
 
-ENV INVOKEAI_SRC=/opt/invokeai
-ENV VIRTUAL_ENV=/opt/venv/invokeai
+# Install `uv` for package management
+COPY --from=ghcr.io/astral-sh/uv:0.5.5 /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_COMPILE_BYTECODE=1
+ENV UV_LINK_MODE=copy
+
 ARG GPU_DRIVER=cuda
 ARG TARGETPLATFORM="linux/amd64"
 # unused but available
 ARG BUILDPLATFORM
 
-WORKDIR ${INVOKEAI_SRC}
+# 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
+USER ubuntu
 
+# Install python and create the venv
+RUN uv python install ${PYTHON_VERSION} && \
+    uv venv --relocatable --prompt "invoke" --python ${PYTHON_VERSION} ${VIRTUAL_ENV}
+
+WORKDIR ${INVOKEAI_SRC}
 COPY invokeai ./invokeai
 COPY pyproject.toml ./
 
@@ -32,25 +45,18 @@ COPY pyproject.toml ./
 # the local working copy can be bind-mounted into the image
 # at path defined by ${INVOKEAI_SRC}
 # NOTE: there are no pytorch builds for arm64 + cuda, only cpu
-# x86_64/CUDA is default
-RUN --mount=type=cache,target=/root/.cache/pip \
-    python3 -m venv ${VIRTUAL_ENV} &&\
+# x86_64/CUDA is the default
+RUN --mount=type=cache,target=/home/ubuntu/.cache/uv,uid=1000,gid=1000 \
     if [ "$TARGETPLATFORM" = "linux/arm64" ] || [ "$GPU_DRIVER" = "cpu" ]; then \
         extra_index_url_arg="--extra-index-url https://download.pytorch.org/whl/cpu"; \
     elif [ "$GPU_DRIVER" = "rocm" ]; then \
         extra_index_url_arg="--extra-index-url https://download.pytorch.org/whl/rocm6.1"; \
     else \
         extra_index_url_arg="--extra-index-url https://download.pytorch.org/whl/cu124"; \
-    fi &&\
+    fi && \
+    uv pip install --python ${PYTHON_VERSION} $extra_index_url_arg -e "."
 
-    # xformers + triton fails to install on arm64
-    if [ "$GPU_DRIVER" = "cuda" ] && [ "$TARGETPLATFORM" = "linux/amd64" ]; then \
-        pip install $extra_index_url_arg -e ".[xformers]"; \
-    else \
-        pip install $extra_index_url_arg -e "."; \
-    fi
-
-# #### Build the Web UI ------------------------------------
+#### Build the Web UI ------------------------------------
 
 FROM node:20-slim AS web-builder
 ENV PNPM_HOME="/pnpm"
@@ -66,7 +72,7 @@ RUN npx vite build
 
 #### Runtime stage ---------------------------------------
 
-FROM library/ubuntu:23.04 AS runtime
+FROM library/ubuntu:24.04 AS runtime
 
 ARG DEBIAN_FRONTEND=noninteractive
 ENV PYTHONUNBUFFERED=1
@@ -83,17 +89,16 @@ RUN apt update && apt install -y --no-install-recommends \
         gosu \
         magic-wormhole \
         libglib2.0-0 \
-        libgl1-mesa-glx \
-        python3-venv \
-        python3-pip \
+        libgl1 \
+        libglx-mesa0 \
         build-essential \
         libopencv-dev \
         libstdc++-10-dev &&\
     apt-get clean && apt-get autoclean
 
-
 ENV INVOKEAI_SRC=/opt/invokeai
-ENV VIRTUAL_ENV=/opt/venv/invokeai
+ENV VIRTUAL_ENV=/opt/venv
+ENV PYTHON_VERSION=3.11
 ENV INVOKEAI_ROOT=/invokeai
 ENV INVOKEAI_HOST=0.0.0.0
 ENV INVOKEAI_PORT=9090
@@ -101,6 +106,14 @@ ENV PATH="$VIRTUAL_ENV/bin:$INVOKEAI_SRC:$PATH"
 ENV CONTAINER_UID=${CONTAINER_UID:-1000}
 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/
+USER ubuntu
+RUN uv python install ${PYTHON_VERSION}
+USER root
+
 # --link requires buldkit w/ dockerfile syntax 1.4
 COPY --link --from=builder ${INVOKEAI_SRC} ${INVOKEAI_SRC}
 COPY --link --from=builder ${VIRTUAL_ENV} ${VIRTUAL_ENV}
@@ -115,7 +128,7 @@ WORKDIR ${INVOKEAI_SRC}
 
 # build patchmatch
 RUN cd /usr/lib/$(uname -p)-linux-gnu/pkgconfig/ && ln -sf opencv4.pc opencv.pc
-RUN python3 -c "from patchmatch import patch_match"
+RUN python -c "from patchmatch import patch_match"
 
 RUN mkdir -p ${INVOKEAI_ROOT} && chown -R ${CONTAINER_UID}:${CONTAINER_GID} ${INVOKEAI_ROOT}
 

docker/docker-entrypoint.sh

@@ -16,6 +16,9 @@ set -e -o pipefail
 
 USER_ID=${CONTAINER_UID:-1000}
 USER=ubuntu
+# if the user does not exist, create it. It is expected to be present on ubuntu >=24.x
+_=$(id ${USER} 2>&1) || useradd -u ${USER_ID} ${USER}
+# ensure the UID is correct
 usermod -u ${USER_ID} ${USER} 1>/dev/null
 
 ### Set the $PUBLIC_KEY env var to enable SSH access.
@@ -36,6 +39,8 @@ fi
 mkdir -p "${INVOKEAI_ROOT}"
 chown --recursive ${USER} "${INVOKEAI_ROOT}" || true
 cd "${INVOKEAI_ROOT}"
+export HF_HOME=${HF_HOME:-$INVOKEAI_ROOT/.cache/huggingface}
+export MPLCONFIGDIR=${MPLCONFIGDIR:-$INVOKEAI_ROOT/.matplotlib}
 
 # Run the CMD as the Container User (not root).
 exec gosu ${USER} "$@"

docs/contributing/ARCHITECTURE.md

@@ -50,7 +50,7 @@ Applications are built on top of the invoke framework. They should construct `in
 
 ### Web UI
 
-The Web UI is built on top of an HTTP API built with [FastAPI](https://fastapi.tiangolo.com/) and [Socket.IO](https://socket.io/). The frontend code is found in `/frontend` and the backend code is found in `/ldm/invoke/app/api_app.py` and `/ldm/invoke/app/api/`. The code is further organized as such:
+The Web UI is built on top of an HTTP API built with [FastAPI](https://fastapi.tiangolo.com/) and [Socket.IO](https://socket.io/). The frontend code is found in `/invokeai/frontend` and the backend code is found in `/invokeai/app/api_app.py` and `/invokeai/app/api/`. The code is further organized as such:
 
 | Component | Description |
 | --- | --- |
@@ -62,7 +62,7 @@ The Web UI is built on top of an HTTP API built with [FastAPI](https://fastapi.t
 
 ### CLI
 
-The CLI is built automatically from invocation metadata, and also supports invocation piping and auto-linking. Code is available in `/ldm/invoke/app/cli_app.py`.
+The CLI is built automatically from invocation metadata, and also supports invocation piping and auto-linking. Code is available in `/invokeai/frontend/cli`.
 
 ## Invoke
 
@@ -70,7 +70,7 @@ The Invoke framework provides the interface to the underlying AI systems and is
 
 ### Invoker
 
-The invoker (`/ldm/invoke/app/services/invoker.py`) is the primary interface through which applications interact with the framework. Its primary purpose is to create, manage, and invoke sessions. It also maintains two sets of services:
+The invoker (`/invokeai/app/services/invoker.py`) is the primary interface through which applications interact with the framework. Its primary purpose is to create, manage, and invoke sessions. It also maintains two sets of services:
 - **invocation services**, which are used by invocations to interact with core functionality.
 - **invoker services**, which are used by the invoker to manage sessions and manage the invocation queue.
 
@@ -82,12 +82,12 @@ The session graph does not support looping. This is left as an application probl
 
 ### Invocations
 
-Invocations represent individual units of execution, with inputs and outputs. All invocations are located in `/ldm/invoke/app/invocations`, and are all automatically discovered and made available in the applications. These are the primary way to expose new functionality in Invoke.AI, and the [implementation guide](INVOCATIONS.md) explains how to add new invocations.
+Invocations represent individual units of execution, with inputs and outputs. All invocations are located in `/invokeai/app/invocations`, and are all automatically discovered and made available in the applications. These are the primary way to expose new functionality in Invoke.AI, and the [implementation guide](INVOCATIONS.md) explains how to add new invocations.
 
 ### Services
 
-Services provide invocations access AI Core functionality and other necessary functionality (e.g. image storage). These are available in `/ldm/invoke/app/services`. As a general rule, new services should provide an interface as an abstract base class, and may provide a lightweight local implementation by default in their module. The goal for all services should be to enable the usage of different implementations (e.g. using cloud storage for image storage), but should not load any module dependencies unless that implementation has been used (i.e. don't import anything that won't be used, especially if it's expensive to import).
+Services provide invocations access AI Core functionality and other necessary functionality (e.g. image storage). These are available in `/invokeai/app/services`. As a general rule, new services should provide an interface as an abstract base class, and may provide a lightweight local implementation by default in their module. The goal for all services should be to enable the usage of different implementations (e.g. using cloud storage for image storage), but should not load any module dependencies unless that implementation has been used (i.e. don't import anything that won't be used, especially if it's expensive to import).
 
 ## AI Core
 
-The AI Core is represented by the rest of the code base (i.e. the code outside of `/ldm/invoke/app/`).
+The AI Core is represented by the rest of the code base (i.e. the code outside of `/invokeai/app/`).

docs/contributing/INVOCATIONS.md

@@ -287,8 +287,8 @@ new Invocation ready to be used.
 
 Once you've created a Node, the next step is to share it with the community! The
 best way to do this is to submit a Pull Request to add the Node to the
-[Community Nodes](nodes/communityNodes) list. If you're not sure how to do that,
-take a look a at our [contributing nodes overview](contributingNodes).
+[Community Nodes](../nodes/communityNodes.md) list. If you're not sure how to do that,
+take a look a at our [contributing nodes overview](../nodes/contributingNodes.md).
 
 ## Advanced
 

docs/contributing/MODEL_MANAGER.md

@@ -9,20 +9,20 @@ model. These are the:
   configuration information. Among other things, the record service
   tracks the type of the model, its provenance, and where it can be
   found on disk.
-  
+
 * _ModelInstallServiceBase_ A service for installing models to
   disk. It uses `DownloadQueueServiceBase` to download models and
   their metadata, and `ModelRecordServiceBase` to store that
   information. It is also responsible for managing the InvokeAI
   `models` directory and its contents.
-  
+
 * _DownloadQueueServiceBase_
   A multithreaded downloader responsible
   for downloading models from a remote source to disk. The download
   queue has special methods for downloading repo_id folders from
   Hugging Face, as well as discriminating among model versions in
   Civitai, but can be used for arbitrary content.
-  
+
   * _ModelLoadServiceBase_
   Responsible for loading a model from disk
   into RAM and VRAM and getting it ready for inference.
@@ -207,9 +207,9 @@ for use in the InvokeAI web server. Its signature is:
 
 ```
 def open(
-       cls, 
-    config: InvokeAIAppConfig, 
-    conn: Optional[sqlite3.Connection] = None, 
+       cls,
+    config: InvokeAIAppConfig,
+    conn: Optional[sqlite3.Connection] = None,
     lock: Optional[threading.Lock] = None
     ) -> Union[ModelRecordServiceSQL, ModelRecordServiceFile]:
 ```
@@ -363,7 +363,7 @@ functionality:
 
 * Registering a model config record for a model already located on the
   local filesystem, without moving it or changing its path.
-  
+
 * Installing a model alreadiy located on the local filesystem, by
   moving it into the InvokeAI root directory under the
   `models` folder (or wherever config parameter `models_dir`
@@ -371,21 +371,21 @@ functionality:
 
 * Probing of models to determine their type, base type and other key
   information.
-  
+
 * Interface with the InvokeAI event bus to provide status updates on
   the download, installation and registration process.
-  
+
 * Downloading a model from an arbitrary URL and installing it in
   `models_dir`.
 
 * Special handling for HuggingFace repo_ids to recursively download
   the contents of the repository, paying attention to alternative
   variants such as fp16.
-  
+
 * Saving tags and other metadata about the model into the invokeai database
   when fetching from a repo that provides that type of information,
   (currently only HuggingFace).
-  
+
 ### Initializing the installer
 
 A default installer is created at InvokeAI api startup time and stored
@@ -461,7 +461,7 @@ revision.
 `config` is an optional dict of values that will override the
 autoprobed values for model type, base, scheduler prediction type, and
 so forth. See [Model configuration and
-probing](#Model-configuration-and-probing) for details.
+probing](#model-configuration-and-probing) for details.
 
 `access_token` is an optional access token for accessing resources
 that need authentication.
@@ -494,7 +494,7 @@ source8 = URLModelSource(url='https://civitai.com/api/download/models/63006', ac
 
 for source in [source1, source2, source3, source4, source5, source6, source7]:
    install_job = installer.install_model(source)
-   
+
 source2job = installer.wait_for_installs(timeout=120)
 for source in sources:
     job = source2job[source]
@@ -504,7 +504,7 @@ for source in sources:
   print(f"{source} installed as {model_key}")
  elif job.errored:
      print(f"{source}: {job.error_type}.\nStack trace:\n{job.error}")
- 
+
 ```
 
 As shown here, the `import_model()` method accepts a variety of

docs/contributing/TESTS.md

@@ -1,6 +1,6 @@
 # InvokeAI Backend Tests
 
-We use `pytest` to run the backend python tests. (See [pyproject.toml](/pyproject.toml) for the default `pytest` options.)
+We use `pytest` to run the backend python tests. (See [pyproject.toml](https://github.com/invoke-ai/InvokeAI/blob/main/pyproject.toml) for the default `pytest` options.)
 
 ## Fast vs. Slow
 All tests are categorized as either 'fast' (no test annotation) or 'slow' (annotated with the `@pytest.mark.slow` decorator).
@@ -33,7 +33,7 @@ pytest tests -m ""
 
 ## Test Organization
 
-All backend tests are in the [`tests/`](/tests/) directory. This directory mirrors the organization of the `invokeai/` directory. For example, tests for `invokeai/model_management/model_manager.py` would be found in `tests/model_management/test_model_manager.py`.
+All backend tests are in the [`tests/`](https://github.com/invoke-ai/InvokeAI/tree/main/tests) directory. This directory mirrors the organization of the `invokeai/` directory. For example, tests for `invokeai/model_management/model_manager.py` would be found in `tests/model_management/test_model_manager.py`.
 
 TODO: The above statement is aspirational. A re-organization of legacy tests is required to make it true.
 

docs/contributing/contribution_guides/development.md

@@ -2,7 +2,7 @@
 
 ## **What do I need to know to help?**
 
-If you are looking to help with a code contribution, InvokeAI uses several different technologies under the hood: Python (Pydantic, FastAPI, diffusers) and Typescript (React, Redux Toolkit, ChakraUI, Mantine, Konva). Familiarity with StableDiffusion and image generation concepts is helpful, but not essential. 
+If you are looking to help with a code contribution, InvokeAI uses several different technologies under the hood: Python (Pydantic, FastAPI, diffusers) and Typescript (React, Redux Toolkit, ChakraUI, Mantine, Konva). Familiarity with StableDiffusion and image generation concepts is helpful, but not essential.
 
 
 ## **Get Started**
@@ -12,7 +12,7 @@ To get started, take a look at our [new contributors checklist](newContributorCh
 Once you're setup, for more information, you can review the documentation specific to your area of interest:
 
 * #### [InvokeAI Architecure](../ARCHITECTURE.md)
-* #### [Frontend Documentation](https://github.com/invoke-ai/InvokeAI/tree/main/invokeai/frontend/web)
+* #### [Frontend Documentation](../frontend/index.md)
 * #### [Node Documentation](../INVOCATIONS.md)
 * #### [Local Development](../LOCAL_DEVELOPMENT.md)
 
@@ -20,15 +20,15 @@ Once you're setup, for more information, you can review the documentation specif
 
 If you don't feel ready to make a code contribution yet, no problem! You can also help out in other ways, such as [documentation](documentation.md), [translation](translation.md) or helping support other users and triage issues as they're reported in GitHub.
 
-There are two paths to making a development contribution: 
+There are two paths to making a development contribution:
 
 1. Choosing an open issue to address. Open issues can be found in the [Issues](https://github.com/invoke-ai/InvokeAI/issues?q=is%3Aissue+is%3Aopen) section of the InvokeAI repository. These are tagged by the issue type (bug, enhancement, etc.) along with the “good first issues” tag denoting if they are suitable for first time contributors.
-    1. Additional items can be found on our [roadmap](https://github.com/orgs/invoke-ai/projects/7). The roadmap is organized in terms of priority, and contains features of varying size and complexity. If there is an inflight item you’d like to help with, reach out to the contributor assigned to the item to see how you can help. 
+    1. Additional items can be found on our [roadmap](https://github.com/orgs/invoke-ai/projects/7). The roadmap is organized in terms of priority, and contains features of varying size and complexity. If there is an inflight item you’d like to help with, reach out to the contributor assigned to the item to see how you can help.
 2. Opening a new issue or feature to add. **Please make sure you have searched through existing issues before creating new ones.**
 
 *Regardless of what you choose, please post in the  [#dev-chat](https://discord.com/channels/1020123559063990373/1049495067846524939) channel of the Discord before you start development in order to confirm that the issue or feature is aligned with the current direction of the project. We value our contributors time and effort and want to ensure that no one’s time is being misspent.*
 
-## Best Practices: 
+## Best Practices:
 * Keep your pull requests small. Smaller pull requests are more likely to be accepted and merged
 * Comments! Commenting your code helps reviewers easily understand your contribution
 * Use Python and Typescript’s typing systems, and consider using an editor with [LSP](https://microsoft.github.io/language-server-protocol/) support to streamline development
@@ -38,7 +38,7 @@ There are two paths to making a development contribution:
 
 If you need help, you can ask questions in the [#dev-chat](https://discord.com/channels/1020123559063990373/1049495067846524939) channel of the Discord.
 
-For frontend related work, **@psychedelicious** is the best person to reach out to. 
+For frontend related work, **@psychedelicious** is the best person to reach out to.
 
 For backend related work, please reach out to **@blessedcoolant**, **@lstein**, **@StAlKeR7779** or **@psychedelicious**.
 

docs/contributing/contribution_guides/newContributorChecklist.md

@@ -22,15 +22,15 @@ Before starting these steps, ensure you have your local environment [configured
 2. Fork the [InvokeAI](https://github.com/invoke-ai/InvokeAI) repository to your GitHub profile. This means that you will have a copy of the repository under **your-GitHub-username/InvokeAI**.
 3. Clone the repository to your local machine using:
 
-   ```bash
-   git clone https://github.com/your-GitHub-username/InvokeAI.git
-   ```
+    ```bash
+    git clone https://github.com/your-GitHub-username/InvokeAI.git
+    ```
 
 If you're unfamiliar with using Git through the commandline, [GitHub Desktop](https://desktop.github.com) is a easy-to-use alternative with a UI. You can do all the same steps listed here, but through the interface. 4. Create a new branch for your fix using:
 
-    ```bash
-    git checkout -b branch-name-here
-    ```
+  ```bash
+  git checkout -b branch-name-here
+  ```
 
 5. Make the appropriate changes for the issue you are trying to address or the feature that you want to add.
 6. Add the file contents of the changed files to the "snapshot" git uses to manage the state of the project, also known as the index:

docs/contributing/dev-environment.md

@@ -27,9 +27,9 @@ If you just want to use Invoke, you should use the [installer][installer link].
 
 5. Activate the venv (you'll need to do this every time you want to run the app):
 
-        ```sh
-        source .venv/bin/activate
-        ```
+      ```sh
+      source .venv/bin/activate
+      ```
 
 6. Install the repo as an [editable install][editable install link]:
 
@@ -37,7 +37,7 @@ If you just want to use Invoke, you should use the [installer][installer link].
       pip install -e ".[dev,test,xformers]" --use-pep517 --extra-index-url https://download.pytorch.org/whl/cu121
       ```
 
-      Refer to the [manual installation][manual install link]] instructions for more determining the correct install options. `xformers` is optional, but `dev` and `test` are not.
+      Refer to the [manual installation][manual install link] instructions for more determining the correct install options. `xformers` is optional, but `dev` and `test` are not.
 
 7. Install the frontend dev toolchain:
 

docs/contributing/index.md

@@ -34,11 +34,11 @@ Please reach out to @hipsterusername on [Discord](https://discord.gg/ZmtBAhwWhy)
 
 ## Contributors
 
-This project is a combined effort of dedicated people from across the world. [Check out the list of all these amazing people](https://invoke-ai.github.io/InvokeAI/other/CONTRIBUTORS/). We thank them for their time, hard work and effort.
+This project is a combined effort of dedicated people from across the world. [Check out the list of all these amazing people](contributors.md). We thank them for their time, hard work and effort.
 
 ## Code of Conduct
 
-The InvokeAI community is a welcoming place, and we want your help in maintaining that. Please review our [Code of Conduct](https://github.com/invoke-ai/InvokeAI/blob/main/CODE_OF_CONDUCT.md) to learn more - it's essential to maintaining a respectful and inclusive environment.
+The InvokeAI community is a welcoming place, and we want your help in maintaining that. Please review our [Code of Conduct](../CODE_OF_CONDUCT.md) to learn more - it's essential to maintaining a respectful and inclusive environment.
 
 By making a contribution to this project, you certify that:
 

invokeai/app/api/routers/session_queue.py

@@ -110,7 +110,7 @@ async def cancel_by_batch_ids(
 @session_queue_router.put(
     "/{queue_id}/cancel_by_destination",
     operation_id="cancel_by_destination",
-    responses={200: {"model": CancelByBatchIDsResult}},
+    responses={200: {"model": CancelByDestinationResult}},
 )
 async def cancel_by_destination(
     queue_id: str = Path(description="The queue id to perform this operation on"),

invokeai/app/invocations/__init__.py

@@ -15,6 +15,11 @@ custom_nodes_readme_path = str(custom_nodes_path / "README.md")
 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:

invokeai/app/invocations/compel.py

@@ -82,10 +82,11 @@ class CompelInvocation(BaseInvocation):
             # apply all patches while the model is on the target device
             text_encoder_info.model_on_device() as (cached_weights, text_encoder),
             tokenizer_info as tokenizer,
-            LoRAPatcher.apply_lora_patches(
+            LoRAPatcher.apply_smart_lora_patches(
                 model=text_encoder,
                 patches=_lora_loader(),
                 prefix="lora_te_",
+                dtype=TorchDevice.choose_torch_dtype(),
                 cached_weights=cached_weights,
             ),
             # Apply CLIP Skip after LoRA to prevent LoRA application from failing on skipped layers.
@@ -179,10 +180,11 @@ class SDXLPromptInvocationBase:
             # apply all patches while the model is on the target device
             text_encoder_info.model_on_device() as (cached_weights, text_encoder),
             tokenizer_info as tokenizer,
-            LoRAPatcher.apply_lora_patches(
+            LoRAPatcher.apply_smart_lora_patches(
                 text_encoder,
                 patches=_lora_loader(),
                 prefix=lora_prefix,
+                dtype=TorchDevice.choose_torch_dtype(),
                 cached_weights=cached_weights,
             ),
             # Apply CLIP Skip after LoRA to prevent LoRA application from failing on skipped layers.

invokeai/app/invocations/denoise_latents.py

@@ -1003,10 +1003,11 @@ class DenoiseLatentsInvocation(BaseInvocation):
             ModelPatcher.apply_freeu(unet, self.unet.freeu_config),
             SeamlessExt.static_patch_model(unet, self.unet.seamless_axes),  # FIXME
             # Apply the LoRA after unet has been moved to its target device for faster patching.
-            LoRAPatcher.apply_lora_patches(
+            LoRAPatcher.apply_smart_lora_patches(
                 model=unet,
                 patches=_lora_loader(),
                 prefix="lora_unet_",
+                dtype=unet.dtype,
                 cached_weights=cached_weights,
             ),
         ):

invokeai/app/invocations/fields.py

@@ -56,6 +56,7 @@ class UIType(str, Enum, metaclass=MetaEnum):
     CLIPLEmbedModel = "CLIPLEmbedModelField"
     CLIPGEmbedModel = "CLIPGEmbedModelField"
     SpandrelImageToImageModel = "SpandrelImageToImageModelField"
+    StructuralLoRAModel = "StructuralLoRAModelField"
     # endregion
 
     # region Misc Field Types
@@ -143,6 +144,7 @@ class FieldDescriptions:
     controlnet_model = "ControlNet model to load"
     vae_model = "VAE model to load"
     lora_model = "LoRA model to load"
+    structural_lora_model = "Structural LoRA model to load"
     main_model = "Main model (UNet, VAE, CLIP) to load"
     flux_model = "Flux model (Transformer) to load"
     sd3_model = "SD3 model (MMDiTX) to load"
@@ -250,6 +252,11 @@ class FluxConditioningField(BaseModel):
     """A conditioning tensor primitive value"""
 
     conditioning_name: str = Field(description="The name of conditioning tensor")
+    mask: Optional[TensorField] = Field(
+        default=None,
+        description="The mask associated with this conditioning tensor. Excluded regions should be set to False, "
+        "included regions should be set to True.",
+    )
 
 
 class SD3ConditioningField(BaseModel):

invokeai/app/invocations/flux_denoise.py

@@ -1,5 +1,5 @@
 from contextlib import ExitStack
-from typing import Callable, Iterator, Optional, Tuple
+from typing import Callable, Iterator, Optional, Tuple, Union
 
 import numpy as np
 import numpy.typing as npt
@@ -8,6 +8,8 @@ import torchvision.transforms as tv_transforms
 from torchvision.transforms.functional import resize as tv_resize
 from transformers import CLIPImageProcessor, CLIPVisionModelWithProjection
 
+from invokeai.backend.flux.modules.autoencoder import AutoEncoder
+
 from invokeai.app.invocations.baseinvocation import BaseInvocation, Classification, invocation
 from invokeai.app.invocations.fields import (
     DenoiseMaskField,
@@ -22,7 +24,7 @@ from invokeai.app.invocations.fields import (
 )
 from invokeai.app.invocations.flux_controlnet import FluxControlNetField
 from invokeai.app.invocations.ip_adapter import IPAdapterField
-from invokeai.app.invocations.model import TransformerField, VAEField
+from invokeai.app.invocations.model import TransformerField, VAEField, StructuralLoRAField, LoRAField
 from invokeai.app.invocations.primitives import LatentsOutput
 from invokeai.app.services.shared.invocation_context import InvocationContext
 from invokeai.backend.flux.controlnet.instantx_controlnet_flux import InstantXControlNetFlux
@@ -30,6 +32,7 @@ from invokeai.backend.flux.controlnet.xlabs_controlnet_flux import XLabsControlN
 from invokeai.backend.flux.denoise import denoise
 from invokeai.backend.flux.extensions.inpaint_extension import InpaintExtension
 from invokeai.backend.flux.extensions.instantx_controlnet_extension import InstantXControlNetExtension
+from invokeai.backend.flux.extensions.regional_prompting_extension import RegionalPromptingExtension
 from invokeai.backend.flux.extensions.xlabs_controlnet_extension import XLabsControlNetExtension
 from invokeai.backend.flux.extensions.xlabs_ip_adapter_extension import XLabsIPAdapterExtension
 from invokeai.backend.flux.ip_adapter.xlabs_ip_adapter_flux import XlabsIpAdapterFlux
@@ -42,6 +45,9 @@ from invokeai.backend.flux.sampling_utils import (
     pack,
     unpack,
 )
+from invokeai.backend.flux.flux_tools_sampling_utils import prepare_control
+from invokeai.backend.flux.modules.conditioner import HFEncoder
+from invokeai.backend.flux.text_conditioning import FluxTextConditioning
 from invokeai.backend.lora.conversions.flux_lora_constants import FLUX_LORA_TRANSFORMER_PREFIX
 from invokeai.backend.lora.lora_model_raw import LoRAModelRaw
 from invokeai.backend.lora.lora_patcher import LoRAPatcher
@@ -56,7 +62,7 @@ from invokeai.backend.util.devices import TorchDevice
     title="FLUX Denoise",
     tags=["image", "flux"],
     category="image",
-    version="3.2.1",
+    version="3.2.2",
     classification=Classification.Prototype,
 )
 class FluxDenoiseInvocation(BaseInvocation, WithMetadata, WithBoard):
@@ -87,10 +93,10 @@ class FluxDenoiseInvocation(BaseInvocation, WithMetadata, WithBoard):
         input=Input.Connection,
         title="Transformer",
     )
-    positive_text_conditioning: FluxConditioningField = InputField(
+    positive_text_conditioning: FluxConditioningField | list[FluxConditioningField] = InputField(
         description=FieldDescriptions.positive_cond, input=Input.Connection
     )
-    negative_text_conditioning: FluxConditioningField | None = InputField(
+    negative_text_conditioning: FluxConditioningField | list[FluxConditioningField] | None = InputField(
         default=None,
         description="Negative conditioning tensor. Can be None if cfg_scale is 1.0.",
         input=Input.Connection,
@@ -139,36 +145,12 @@ class FluxDenoiseInvocation(BaseInvocation, WithMetadata, WithBoard):
         name = context.tensors.save(tensor=latents)
         return LatentsOutput.build(latents_name=name, latents=latents, seed=None)
 
-    def _load_text_conditioning(
-        self, context: InvocationContext, conditioning_name: str, dtype: torch.dtype
-    ) -> Tuple[torch.Tensor, torch.Tensor]:
-        # Load the conditioning data.
-        cond_data = context.conditioning.load(conditioning_name)
-        assert len(cond_data.conditionings) == 1
-        flux_conditioning = cond_data.conditionings[0]
-        assert isinstance(flux_conditioning, FLUXConditioningInfo)
-        flux_conditioning = flux_conditioning.to(dtype=dtype)
-        t5_embeddings = flux_conditioning.t5_embeds
-        clip_embeddings = flux_conditioning.clip_embeds
-        return t5_embeddings, clip_embeddings
-
     def _run_diffusion(
         self,
         context: InvocationContext,
     ):
         inference_dtype = torch.bfloat16
 
-        # Load the conditioning data.
-        pos_t5_embeddings, pos_clip_embeddings = self._load_text_conditioning(
-            context, self.positive_text_conditioning.conditioning_name, inference_dtype
-        )
-        neg_t5_embeddings: torch.Tensor | None = None
-        neg_clip_embeddings: torch.Tensor | None = None
-        if self.negative_text_conditioning is not None:
-            neg_t5_embeddings, neg_clip_embeddings = self._load_text_conditioning(
-                context, self.negative_text_conditioning.conditioning_name, inference_dtype
-            )
-
         # Load the input latents, if provided.
         init_latents = context.tensors.load(self.latents.latents_name) if self.latents else None
         if init_latents is not None:
@@ -183,15 +165,45 @@ class FluxDenoiseInvocation(BaseInvocation, WithMetadata, WithBoard):
             dtype=inference_dtype,
             seed=self.seed,
         )
+        b, _c, latent_h, latent_w = noise.shape
+        packed_h = latent_h // 2
+        packed_w = latent_w // 2
+
+        # Load the conditioning data.
+        pos_text_conditionings = self._load_text_conditioning(
+            context=context,
+            cond_field=self.positive_text_conditioning,
+            packed_height=packed_h,
+            packed_width=packed_w,
+            dtype=inference_dtype,
+            device=TorchDevice.choose_torch_device(),
+        )
+        neg_text_conditionings: list[FluxTextConditioning] | None = None
+        if self.negative_text_conditioning is not None:
+            neg_text_conditionings = self._load_text_conditioning(
+                context=context,
+                cond_field=self.negative_text_conditioning,
+                packed_height=packed_h,
+                packed_width=packed_w,
+                dtype=inference_dtype,
+                device=TorchDevice.choose_torch_device(),
+            )
+        pos_regional_prompting_extension = RegionalPromptingExtension.from_text_conditioning(
+            pos_text_conditionings, img_seq_len=packed_h * packed_w
+        )
+        neg_regional_prompting_extension = (
+            RegionalPromptingExtension.from_text_conditioning(neg_text_conditionings, img_seq_len=packed_h * packed_w)
+            if neg_text_conditionings
+            else None
+        )
 
         transformer_info = context.models.load(self.transformer.transformer)
         is_schnell = "schnell" in transformer_info.config.config_path
 
         # Calculate the timestep schedule.
-        image_seq_len = noise.shape[-1] * noise.shape[-2] // 4
         timesteps = get_schedule(
             num_steps=self.num_steps,
-            image_seq_len=image_seq_len,
+            image_seq_len=packed_h * packed_w,
             shift=not is_schnell,
         )
 
@@ -228,28 +240,17 @@ class FluxDenoiseInvocation(BaseInvocation, WithMetadata, WithBoard):
 
         inpaint_mask = self._prep_inpaint_mask(context, x)
 
-        b, _c, latent_h, latent_w = x.shape
         img_ids = generate_img_ids(h=latent_h, w=latent_w, batch_size=b, device=x.device, dtype=x.dtype)
 
-        pos_bs, pos_t5_seq_len, _ = pos_t5_embeddings.shape
-        pos_txt_ids = torch.zeros(
-            pos_bs, pos_t5_seq_len, 3, dtype=inference_dtype, device=TorchDevice.choose_torch_device()
-        )
-        neg_txt_ids: torch.Tensor | None = None
-        if neg_t5_embeddings is not None:
-            neg_bs, neg_t5_seq_len, _ = neg_t5_embeddings.shape
-            neg_txt_ids = torch.zeros(
-                neg_bs, neg_t5_seq_len, 3, dtype=inference_dtype, device=TorchDevice.choose_torch_device()
-            )
-
         # Pack all latent tensors.
         init_latents = pack(init_latents) if init_latents is not None else None
         inpaint_mask = pack(inpaint_mask) if inpaint_mask is not None else None
         noise = pack(noise)
         x = pack(x)
 
-        # Now that we have 'packed' the latent tensors, verify that we calculated the image_seq_len correctly.
-        assert image_seq_len == x.shape[1]
+        # Now that we have 'packed' the latent tensors, verify that we calculated the image_seq_len, packed_h, and
+        # packed_w correctly.
+        assert packed_h * packed_w == x.shape[1]
 
         # Prepare inpaint extension.
         inpaint_extension: InpaintExtension | None = None
@@ -287,6 +288,16 @@ class FluxDenoiseInvocation(BaseInvocation, WithMetadata, WithBoard):
                 dtype=inference_dtype,
                 device=x.device,
             )
+            img_cond = None
+            if struct_lora := self.transformer.structural_lora:
+                # What should we do when we have multiple of these?
+                if not self.controlnet_vae:
+                    raise ValueError("controlnet_vae must be set when using a strutural lora")
+                ae_info = context.models.load(self.controlnet_vae.vae)
+                img = context.images.get_pil(struct_lora.img.image_name)
+                with ae_info as ae:
+                    assert isinstance(ae, AutoEncoder)
+                    img_cond = prepare_control(self.height, self.width, self.seed, ae, img)
 
             # Load the transformer model.
             (cached_weights, transformer) = exit_stack.enter_context(transformer_info.model_on_device())
@@ -299,10 +310,11 @@ class FluxDenoiseInvocation(BaseInvocation, WithMetadata, WithBoard):
             if config.format in [ModelFormat.Checkpoint]:
                 # The model is non-quantized, so we can apply the LoRA weights directly into the model.
                 exit_stack.enter_context(
-                    LoRAPatcher.apply_lora_patches(
+                    LoRAPatcher.apply_smart_lora_patches(
                         model=transformer,
                         patches=self._lora_iterator(context),
                         prefix=FLUX_LORA_TRANSFORMER_PREFIX,
+                        dtype=inference_dtype,
                         cached_weights=cached_weights,
                     )
                 )
@@ -314,7 +326,7 @@ class FluxDenoiseInvocation(BaseInvocation, WithMetadata, WithBoard):
                 # The model is quantized, so apply the LoRA weights as sidecar layers. This results in slower inference,
                 # than directly patching the weights, but is agnostic to the quantization format.
                 exit_stack.enter_context(
-                    LoRAPatcher.apply_lora_sidecar_patches(
+                    LoRAPatcher.apply_lora_wrapper_patches(
                         model=transformer,
                         patches=self._lora_iterator(context),
                         prefix=FLUX_LORA_TRANSFORMER_PREFIX,
@@ -338,12 +350,8 @@ class FluxDenoiseInvocation(BaseInvocation, WithMetadata, WithBoard):
                 model=transformer,
                 img=x,
                 img_ids=img_ids,
-                txt=pos_t5_embeddings,
-                txt_ids=pos_txt_ids,
-                vec=pos_clip_embeddings,
-                neg_txt=neg_t5_embeddings,
-                neg_txt_ids=neg_txt_ids,
-                neg_vec=neg_clip_embeddings,
+                pos_regional_prompting_extension=pos_regional_prompting_extension,
+                neg_regional_prompting_extension=neg_regional_prompting_extension,
                 timesteps=timesteps,
                 step_callback=self._build_step_callback(context),
                 guidance=self.guidance,
@@ -352,11 +360,49 @@ class FluxDenoiseInvocation(BaseInvocation, WithMetadata, WithBoard):
                 controlnet_extensions=controlnet_extensions,
                 pos_ip_adapter_extensions=pos_ip_adapter_extensions,
                 neg_ip_adapter_extensions=neg_ip_adapter_extensions,
+                img_cond=img_cond
             )
 
         x = unpack(x.float(), self.height, self.width)
         return x
 
+    def _load_text_conditioning(
+        self,
+        context: InvocationContext,
+        cond_field: FluxConditioningField | list[FluxConditioningField],
+        packed_height: int,
+        packed_width: int,
+        dtype: torch.dtype,
+        device: torch.device,
+    ) -> list[FluxTextConditioning]:
+        """Load text conditioning data from a FluxConditioningField or a list of FluxConditioningFields."""
+        # Normalize to a list of FluxConditioningFields.
+        cond_list = [cond_field] if isinstance(cond_field, FluxConditioningField) else cond_field
+
+        text_conditionings: list[FluxTextConditioning] = []
+        for cond_field in cond_list:
+            # Load the text embeddings.
+            cond_data = context.conditioning.load(cond_field.conditioning_name)
+            assert len(cond_data.conditionings) == 1
+            flux_conditioning = cond_data.conditionings[0]
+            assert isinstance(flux_conditioning, FLUXConditioningInfo)
+            flux_conditioning = flux_conditioning.to(dtype=dtype, device=device)
+            t5_embeddings = flux_conditioning.t5_embeds
+            clip_embeddings = flux_conditioning.clip_embeds
+
+            # Load the mask, if provided.
+            mask: Optional[torch.Tensor] = None
+            if cond_field.mask is not None:
+                mask = context.tensors.load(cond_field.mask.tensor_name)
+                mask = mask.to(device=device)
+                mask = RegionalPromptingExtension.preprocess_regional_prompt_mask(
+                    mask, packed_height, packed_width, dtype, device
+                )
+
+            text_conditionings.append(FluxTextConditioning(t5_embeddings, clip_embeddings, mask))
+
+        return text_conditionings
+
     @classmethod
     def prep_cfg_scale(
         cls, cfg_scale: float | list[float], timesteps: list[float], cfg_scale_start_step: int, cfg_scale_end_step: int
@@ -652,7 +698,10 @@ class FluxDenoiseInvocation(BaseInvocation, WithMetadata, WithBoard):
         return pos_ip_adapter_extensions, neg_ip_adapter_extensions
 
     def _lora_iterator(self, context: InvocationContext) -> Iterator[Tuple[LoRAModelRaw, float]]:
-        for lora in self.transformer.loras:
+        loras: list[Union[LoRAField, StructuralLoRAField]] = [*self.transformer.loras]
+        if self.transformer.structural_lora:
+            loras.append(self.transformer.structural_lora)
+        for lora in loras:
             lora_info = context.models.load(lora.lora)
             assert isinstance(lora_info.model, LoRAModelRaw)
             yield (lora_info.model, lora.weight)

invokeai/app/invocations/flux_model_loader.py

@@ -81,8 +81,8 @@ class FluxModelLoaderInvocation(BaseInvocation):
         assert isinstance(transformer_config, CheckpointConfigBase)
 
         return FluxModelLoaderOutput(
-            transformer=TransformerField(transformer=transformer, loras=[]),
-            clip=CLIPField(tokenizer=tokenizer, text_encoder=clip_encoder, loras=[], skipped_layers=0),
+            transformer=TransformerField(transformer=transformer, loras=[], structural_loras=[]),
+            clip=CLIPField(tokenizer=tokenizer, text_encoder=clip_encoder, loras=[], structural_loras=[], skipped_layers=0),
             t5_encoder=T5EncoderField(tokenizer=tokenizer2, text_encoder=t5_encoder),
             vae=VAEField(vae=vae),
             max_seq_len=max_seq_lengths[transformer_config.config_path],

invokeai/app/invocations/flux_structural_lora_loader.py

@@ -0,0 +1,70 @@
+from typing import Optional, Literal
+
+from invokeai.app.invocations.baseinvocation import (
+    BaseInvocation,
+    BaseInvocationOutput,
+    Classification,
+    invocation,
+    invocation_output,
+)
+from invokeai.app.invocations.fields import FieldDescriptions, Input, InputField, OutputField, UIType, ImageField
+from invokeai.app.invocations.model import VAEField, StructuralLoRAField, ModelIdentifierField, TransformerField
+from invokeai.app.services.shared.invocation_context import InvocationContext
+
+
+@invocation_output("flux_structural_lora_loader_output")
+class FluxStructuralLoRALoaderOutput(BaseInvocationOutput):
+    """Flux Structural LoRA Loader Output"""
+
+    transformer: Optional[TransformerField] = OutputField(
+        default=None, description=FieldDescriptions.transformer, title="FLUX Transformer"
+    )
+
+
+@invocation(
+    "flux_structural_lora_loader",
+    title="Flux Structural LoRA",
+    tags=["lora", "model", "flux"],
+    category="model",
+    version="1.1.0",
+    classification=Classification.Prototype,
+)
+class FluxStructuralLoRALoaderInvocation(BaseInvocation):
+    """Apply a LoRA model to a FLUX transformer and/or text encoder."""
+
+    lora: ModelIdentifierField = InputField(
+        description=FieldDescriptions.structural_lora_model, title="Structural LoRA", ui_type=UIType.StructuralLoRAModel
+    )
+    transformer: TransformerField | None = InputField(
+        default=None,
+        description=FieldDescriptions.transformer,
+        input=Input.Connection,
+        title="FLUX Transformer",
+    )
+    image: ImageField = InputField(
+        description="The image to encode.",
+    )
+    weight: float = InputField(default=0.75, description=FieldDescriptions.lora_weight)
+
+    def invoke(self, context: InvocationContext) -> FluxStructuralLoRALoaderOutput:
+        lora_key = self.lora.key
+
+        if not context.models.exists(lora_key):
+            raise ValueError(f"Unknown lora: {lora_key}!")
+
+        # Check for existing LoRAs with the same key.
+        if self.transformer and self.transformer.structural_lora and self.transformer.structural_lora.lora.key == lora_key:
+            raise ValueError(f'Structural LoRA "{lora_key}" already applied to transformer.')
+
+        output = FluxStructuralLoRALoaderOutput()
+
+        # Attach LoRA layers to the models.
+        if self.transformer is not None:
+            output.transformer = self.transformer.model_copy(deep=True)
+            output.transformer.structural_lora = StructuralLoRAField(
+                lora=self.lora,
+                img=self.image,
+                weight=self.weight,
+            )
+
+        return output

invokeai/app/invocations/flux_text_encoder.py

@@ -1,11 +1,18 @@
 from contextlib import ExitStack
-from typing import Iterator, Literal, Tuple
+from typing import Iterator, Literal, Optional, Tuple
 
 import torch
 from transformers import CLIPTextModel, CLIPTokenizer, T5EncoderModel, T5Tokenizer
 
 from invokeai.app.invocations.baseinvocation import BaseInvocation, Classification, invocation
-from invokeai.app.invocations.fields import FieldDescriptions, Input, InputField, UIComponent
+from invokeai.app.invocations.fields import (
+    FieldDescriptions,
+    FluxConditioningField,
+    Input,
+    InputField,
+    TensorField,
+    UIComponent,
+)
 from invokeai.app.invocations.model import CLIPField, T5EncoderField
 from invokeai.app.invocations.primitives import FluxConditioningOutput
 from invokeai.app.services.shared.invocation_context import InvocationContext
@@ -15,6 +22,7 @@ from invokeai.backend.lora.lora_model_raw import LoRAModelRaw
 from invokeai.backend.lora.lora_patcher import LoRAPatcher
 from invokeai.backend.model_manager.config import ModelFormat
 from invokeai.backend.stable_diffusion.diffusion.conditioning_data import ConditioningFieldData, FLUXConditioningInfo
+from invokeai.backend.util.devices import TorchDevice
 
 
 @invocation(
@@ -22,7 +30,7 @@ from invokeai.backend.stable_diffusion.diffusion.conditioning_data import Condit
     title="FLUX Text Encoding",
     tags=["prompt", "conditioning", "flux"],
     category="conditioning",
-    version="1.1.0",
+    version="1.1.1",
     classification=Classification.Prototype,
 )
 class FluxTextEncoderInvocation(BaseInvocation):
@@ -41,9 +49,9 @@ class FluxTextEncoderInvocation(BaseInvocation):
     t5_max_seq_len: Literal[256, 512] = InputField(
         description="Max sequence length for the T5 encoder. Expected to be 256 for FLUX schnell models and 512 for FLUX dev models."
     )
-    prompt: str = InputField(
-        description="Text prompt to encode.",
-        ui_component=UIComponent.Textarea,
+    prompt: str = InputField(description="Text prompt to encode.", ui_component=UIComponent.Textarea)
+    mask: Optional[TensorField] = InputField(
+        default=None, description="A mask defining the region that this conditioning prompt applies to."
     )
 
     @torch.no_grad()
@@ -57,7 +65,9 @@ class FluxTextEncoderInvocation(BaseInvocation):
         )
 
         conditioning_name = context.conditioning.save(conditioning_data)
-        return FluxConditioningOutput.build(conditioning_name)
+        return FluxConditioningOutput(
+            conditioning=FluxConditioningField(conditioning_name=conditioning_name, mask=self.mask)
+        )
 
     def _t5_encode(self, context: InvocationContext) -> torch.Tensor:
         t5_tokenizer_info = context.models.load(self.t5_encoder.tokenizer)
@@ -102,10 +112,11 @@ class FluxTextEncoderInvocation(BaseInvocation):
             if clip_text_encoder_config.format in [ModelFormat.Diffusers]:
                 # The model is non-quantized, so we can apply the LoRA weights directly into the model.
                 exit_stack.enter_context(
-                    LoRAPatcher.apply_lora_patches(
+                    LoRAPatcher.apply_smart_lora_patches(
                         model=clip_text_encoder,
                         patches=self._clip_lora_iterator(context),
                         prefix=FLUX_LORA_CLIP_PREFIX,
+                        dtype=TorchDevice.choose_torch_dtype(),
                         cached_weights=cached_weights,
                     )
                 )

invokeai/app/invocations/model.py

@@ -1,5 +1,5 @@
 import copy
-from typing import List, Optional
+from typing import List, Optional, Literal
 
 from pydantic import BaseModel, Field
 
@@ -10,7 +10,7 @@ from invokeai.app.invocations.baseinvocation import (
     invocation,
     invocation_output,
 )
-from invokeai.app.invocations.fields import FieldDescriptions, Input, InputField, OutputField, UIType
+from invokeai.app.invocations.fields import FieldDescriptions, Input, InputField, OutputField, UIType, ImageField
 from invokeai.app.services.shared.invocation_context import InvocationContext
 from invokeai.app.shared.models import FreeUConfig
 from invokeai.backend.model_manager.config import (
@@ -65,11 +65,6 @@ class CLIPField(BaseModel):
     loras: List[LoRAField] = Field(description="LoRAs to apply on model loading")
 
 
-class TransformerField(BaseModel):
-    transformer: ModelIdentifierField = Field(description="Info to load Transformer submodel")
-    loras: List[LoRAField] = Field(description="LoRAs to apply on model loading")
-
-
 class T5EncoderField(BaseModel):
     tokenizer: ModelIdentifierField = Field(description="Info to load tokenizer submodel")
     text_encoder: ModelIdentifierField = Field(description="Info to load text_encoder submodel")
@@ -79,6 +74,13 @@ class VAEField(BaseModel):
     vae: ModelIdentifierField = Field(description="Info to load vae submodel")
     seamless_axes: List[str] = Field(default_factory=list, description='Axes("x" and "y") to which apply seamless')
 
+class StructuralLoRAField(LoRAField):
+    img: ImageField = Field(description="Image to use in structural conditioning")
+
+class TransformerField(BaseModel):
+    transformer: ModelIdentifierField = Field(description="Info to load Transformer submodel")
+    loras: List[LoRAField] = Field(description="LoRAs to apply on model loading")
+    structural_lora: Optional[StructuralLoRAField] = Field(description="Structural LoRAs to apply on model loading", default=None)
 
 @invocation_output("unet_output")
 class UNetOutput(BaseInvocationOutput):

invokeai/app/invocations/sd3_text_encoder.py

@@ -21,6 +21,7 @@ from invokeai.backend.lora.lora_model_raw import LoRAModelRaw
 from invokeai.backend.lora.lora_patcher import LoRAPatcher
 from invokeai.backend.model_manager.config import ModelFormat
 from invokeai.backend.stable_diffusion.diffusion.conditioning_data import ConditioningFieldData, SD3ConditioningInfo
+from invokeai.backend.util.devices import TorchDevice
 
 # The SD3 T5 Max Sequence Length set based on the default in diffusers.
 SD3_T5_MAX_SEQ_LEN = 256
@@ -150,10 +151,11 @@ class Sd3TextEncoderInvocation(BaseInvocation):
             if clip_text_encoder_config.format in [ModelFormat.Diffusers]:
                 # The model is non-quantized, so we can apply the LoRA weights directly into the model.
                 exit_stack.enter_context(
-                    LoRAPatcher.apply_lora_patches(
+                    LoRAPatcher.apply_smart_lora_patches(
                         model=clip_text_encoder,
                         patches=self._clip_lora_iterator(context, clip_model),
                         prefix=FLUX_LORA_CLIP_PREFIX,
+                        dtype=TorchDevice.choose_torch_dtype(),
                         cached_weights=cached_weights,
                     )
                 )

invokeai/app/invocations/tiled_multi_diffusion_denoise_latents.py

@@ -207,7 +207,9 @@ class TiledMultiDiffusionDenoiseLatents(BaseInvocation):
         with (
             ExitStack() as exit_stack,
             unet_info as unet,
-            LoRAPatcher.apply_lora_patches(model=unet, patches=_lora_loader(), prefix="lora_unet_"),
+            LoRAPatcher.apply_smart_lora_patches(
+                model=unet, patches=_lora_loader(), prefix="lora_unet_", dtype=unet.dtype
+            ),
         ):
             assert isinstance(unet, UNet2DConditionModel)
             latents = latents.to(device=unet.device, dtype=unet.dtype)

invokeai/app/services/config/config_default.py

@@ -4,6 +4,7 @@
 from __future__ import annotations
 
 import copy
+import filecmp
 import locale
 import os
 import re
@@ -525,9 +526,35 @@ def get_config() -> InvokeAIAppConfig:
     ]
     example_config.write_file(config.config_file_path.with_suffix(".example.yaml"), as_example=True)
 
-    # Copy all legacy configs - We know `__path__[0]` is correct here
+    # Copy all legacy configs only if needed
+    # We know `__path__[0]` is correct here
     configs_src = Path(model_configs.__path__[0])  # pyright: ignore [reportUnknownMemberType, reportUnknownArgumentType, reportAttributeAccessIssue]
-    shutil.copytree(configs_src, config.legacy_conf_path, dirs_exist_ok=True)
+    dest_path = config.legacy_conf_path
+
+    # Create destination (we don't need to check for existence)
+    dest_path.mkdir(parents=True, exist_ok=True)
+
+    # Compare directories recursively
+    comparison = filecmp.dircmp(configs_src, dest_path)
+    need_copy = any(
+        [
+            comparison.left_only,  # Files exist only in source
+            comparison.diff_files,  # Files that differ
+            comparison.common_funny,  # Files that couldn't be compared
+        ]
+    )
+
+    if need_copy:
+        # Get permissions from destination directory
+        dest_mode = dest_path.stat().st_mode
+
+        # Copy directory tree
+        shutil.copytree(configs_src, dest_path, dirs_exist_ok=True)
+
+        # Set permissions on copied files to match destination directory
+        dest_path.chmod(dest_mode)
+        for p in dest_path.glob("**/*"):
+            p.chmod(dest_mode)
 
     if config.config_file_path.exists():
         config_from_file = load_and_migrate_config(config.config_file_path)

invokeai/app/services/session_processor/session_processor_default.py

@@ -378,6 +378,9 @@ class DefaultSessionProcessor(SessionProcessorBase):
         self._poll_now()
 
     async def _on_queue_item_status_changed(self, event: FastAPIEvent[QueueItemStatusChangedEvent]) -> None:
+        # Make sure the cancel event is for the currently processing queue item
+        if self._queue_item and self._queue_item.item_id != event[1].item_id:
+            return
         if self._queue_item and event[1].status in ["completed", "failed", "canceled"]:
             # When the queue item is canceled via HTTP, the queue item status is set to `"canceled"` and this event is
             # emitted. We need to respond to this event and stop graph execution. This is done by setting the cancel

invokeai/app/services/shared/sqlite_migrator/migrations/migration_11.py

@@ -35,7 +35,7 @@ class Migration11Callback:
 
     def _remove_convert_cache(self) -> None:
         """Rename models/.cache to models/.convert_cache."""
-        self._logger.info("Removing .cache directory. Converted models will now be cached in .convert_cache.")
+        self._logger.info("Removing models/.cache directory. Converted models will now be cached in .convert_cache.")
         legacy_convert_path = self._app_config.root_path / "models" / ".cache"
         shutil.rmtree(legacy_convert_path, ignore_errors=True)
 

invokeai/backend/flux/custom_block_processor.py

@@ -1,9 +1,10 @@
 import einops
 import torch
 
+from invokeai.backend.flux.extensions.regional_prompting_extension import RegionalPromptingExtension
 from invokeai.backend.flux.extensions.xlabs_ip_adapter_extension import XLabsIPAdapterExtension
 from invokeai.backend.flux.math import attention
-from invokeai.backend.flux.modules.layers import DoubleStreamBlock
+from invokeai.backend.flux.modules.layers import DoubleStreamBlock, SingleStreamBlock
 
 
 class CustomDoubleStreamBlockProcessor:
@@ -13,7 +14,12 @@ class CustomDoubleStreamBlockProcessor:
 
     @staticmethod
     def _double_stream_block_forward(
-        block: DoubleStreamBlock, img: torch.Tensor, txt: torch.Tensor, vec: torch.Tensor, pe: torch.Tensor
+        block: DoubleStreamBlock,
+        img: torch.Tensor,
+        txt: torch.Tensor,
+        vec: torch.Tensor,
+        pe: torch.Tensor,
+        attn_mask: torch.Tensor | None = None,
     ) -> tuple[torch.Tensor, torch.Tensor, torch.Tensor]:
         """This function is a direct copy of DoubleStreamBlock.forward(), but it returns some of the intermediate
         values.
@@ -40,7 +46,7 @@ class CustomDoubleStreamBlockProcessor:
         k = torch.cat((txt_k, img_k), dim=2)
         v = torch.cat((txt_v, img_v), dim=2)
 
-        attn = attention(q, k, v, pe=pe)
+        attn = attention(q, k, v, pe=pe, attn_mask=attn_mask)
         txt_attn, img_attn = attn[:, : txt.shape[1]], attn[:, txt.shape[1] :]
 
         # calculate the img bloks
@@ -63,11 +69,15 @@ class CustomDoubleStreamBlockProcessor:
         vec: torch.Tensor,
         pe: torch.Tensor,
         ip_adapter_extensions: list[XLabsIPAdapterExtension],
+        regional_prompting_extension: RegionalPromptingExtension,
     ) -> tuple[torch.Tensor, torch.Tensor]:
         """A custom implementation of DoubleStreamBlock.forward() with additional features:
         - IP-Adapter support
         """
-        img, txt, img_q = CustomDoubleStreamBlockProcessor._double_stream_block_forward(block, img, txt, vec, pe)
+        attn_mask = regional_prompting_extension.get_double_stream_attn_mask(block_index)
+        img, txt, img_q = CustomDoubleStreamBlockProcessor._double_stream_block_forward(
+            block, img, txt, vec, pe, attn_mask=attn_mask
+        )
 
         # Apply IP-Adapter conditioning.
         for ip_adapter_extension in ip_adapter_extensions:
@@ -81,3 +91,48 @@ class CustomDoubleStreamBlockProcessor:
             )
 
         return img, txt
+
+
+class CustomSingleStreamBlockProcessor:
+    """A class containing a custom implementation of SingleStreamBlock.forward() with additional features (masking,
+    etc.)
+    """
+
+    @staticmethod
+    def _single_stream_block_forward(
+        block: SingleStreamBlock,
+        x: torch.Tensor,
+        vec: torch.Tensor,
+        pe: torch.Tensor,
+        attn_mask: torch.Tensor | None = None,
+    ) -> torch.Tensor:
+        """This function is a direct copy of SingleStreamBlock.forward()."""
+        mod, _ = block.modulation(vec)
+        x_mod = (1 + mod.scale) * block.pre_norm(x) + mod.shift
+        qkv, mlp = torch.split(block.linear1(x_mod), [3 * block.hidden_size, block.mlp_hidden_dim], dim=-1)
+
+        q, k, v = einops.rearrange(qkv, "B L (K H D) -> K B H L D", K=3, H=block.num_heads)
+        q, k = block.norm(q, k, v)
+
+        # compute attention
+        attn = attention(q, k, v, pe=pe, attn_mask=attn_mask)
+        # compute activation in mlp stream, cat again and run second linear layer
+        output = block.linear2(torch.cat((attn, block.mlp_act(mlp)), 2))
+        return x + mod.gate * output
+
+    @staticmethod
+    def custom_single_block_forward(
+        timestep_index: int,
+        total_num_timesteps: int,
+        block_index: int,
+        block: SingleStreamBlock,
+        img: torch.Tensor,
+        vec: torch.Tensor,
+        pe: torch.Tensor,
+        regional_prompting_extension: RegionalPromptingExtension,
+    ) -> torch.Tensor:
+        """A custom implementation of SingleStreamBlock.forward() with additional features:
+        - Masking
+        """
+        attn_mask = regional_prompting_extension.get_single_stream_attn_mask(block_index)
+        return CustomSingleStreamBlockProcessor._single_stream_block_forward(block, img, vec, pe, attn_mask=attn_mask)

invokeai/backend/flux/denoise.py

@@ -7,6 +7,7 @@ from tqdm import tqdm
 from invokeai.backend.flux.controlnet.controlnet_flux_output import ControlNetFluxOutput, sum_controlnet_flux_outputs
 from invokeai.backend.flux.extensions.inpaint_extension import InpaintExtension
 from invokeai.backend.flux.extensions.instantx_controlnet_extension import InstantXControlNetExtension
+from invokeai.backend.flux.extensions.regional_prompting_extension import RegionalPromptingExtension
 from invokeai.backend.flux.extensions.xlabs_controlnet_extension import XLabsControlNetExtension
 from invokeai.backend.flux.extensions.xlabs_ip_adapter_extension import XLabsIPAdapterExtension
 from invokeai.backend.flux.model import Flux
@@ -18,14 +19,8 @@ def denoise(
     # model input
     img: torch.Tensor,
     img_ids: torch.Tensor,
-    # positive text conditioning
-    txt: torch.Tensor,
-    txt_ids: torch.Tensor,
-    vec: torch.Tensor,
-    # negative text conditioning
-    neg_txt: torch.Tensor | None,
-    neg_txt_ids: torch.Tensor | None,
-    neg_vec: torch.Tensor | None,
+    pos_regional_prompting_extension: RegionalPromptingExtension,
+    neg_regional_prompting_extension: RegionalPromptingExtension | None,
     # sampling parameters
     timesteps: list[float],
     step_callback: Callable[[PipelineIntermediateState], None],
@@ -35,6 +30,8 @@ def denoise(
     controlnet_extensions: list[XLabsControlNetExtension | InstantXControlNetExtension],
     pos_ip_adapter_extensions: list[XLabsIPAdapterExtension],
     neg_ip_adapter_extensions: list[XLabsIPAdapterExtension],
+    # extra img tokens
+    img_cond: torch.Tensor | None = None,
 ):
     # step 0 is the initial state
     total_steps = len(timesteps) - 1
@@ -61,9 +58,9 @@ def denoise(
                     total_num_timesteps=total_steps,
                     img=img,
                     img_ids=img_ids,
-                    txt=txt,
-                    txt_ids=txt_ids,
-                    y=vec,
+                    txt=pos_regional_prompting_extension.regional_text_conditioning.t5_embeddings,
+                    txt_ids=pos_regional_prompting_extension.regional_text_conditioning.t5_txt_ids,
+                    y=pos_regional_prompting_extension.regional_text_conditioning.clip_embeddings,
                     timesteps=t_vec,
                     guidance=guidance_vec,
                 )
@@ -74,13 +71,13 @@ def denoise(
         # controlnet_residuals datastructure is efficient in that it likely contains multiple references to the same
         # tensors. Calculating the sum materializes each tensor into its own instance.
         merged_controlnet_residuals = sum_controlnet_flux_outputs(controlnet_residuals)
-
+        pred_img = torch.cat((img, img_cond), dim=-1) if img_cond is not None else img
         pred = model(
-            img=img,
+            img=pred_img,
             img_ids=img_ids,
-            txt=txt,
-            txt_ids=txt_ids,
-            y=vec,
+            txt=pos_regional_prompting_extension.regional_text_conditioning.t5_embeddings,
+            txt_ids=pos_regional_prompting_extension.regional_text_conditioning.t5_txt_ids,
+            y=pos_regional_prompting_extension.regional_text_conditioning.clip_embeddings,
             timesteps=t_vec,
             guidance=guidance_vec,
             timestep_index=step_index,
@@ -88,6 +85,7 @@ def denoise(
             controlnet_double_block_residuals=merged_controlnet_residuals.double_block_residuals,
             controlnet_single_block_residuals=merged_controlnet_residuals.single_block_residuals,
             ip_adapter_extensions=pos_ip_adapter_extensions,
+            regional_prompting_extension=pos_regional_prompting_extension,
         )
 
         step_cfg_scale = cfg_scale[step_index]
@@ -97,15 +95,15 @@ def denoise(
             # TODO(ryand): Add option to run positive and negative predictions in a single batch for better performance
             # on systems with sufficient VRAM.
 
-            if neg_txt is None or neg_txt_ids is None or neg_vec is None:
+            if neg_regional_prompting_extension is None:
                 raise ValueError("Negative text conditioning is required when cfg_scale is not 1.0.")
 
             neg_pred = model(
                 img=img,
                 img_ids=img_ids,
-                txt=neg_txt,
-                txt_ids=neg_txt_ids,
-                y=neg_vec,
+                txt=neg_regional_prompting_extension.regional_text_conditioning.t5_embeddings,
+                txt_ids=neg_regional_prompting_extension.regional_text_conditioning.t5_txt_ids,
+                y=neg_regional_prompting_extension.regional_text_conditioning.clip_embeddings,
                 timesteps=t_vec,
                 guidance=guidance_vec,
                 timestep_index=step_index,
@@ -113,6 +111,7 @@ def denoise(
                 controlnet_double_block_residuals=None,
                 controlnet_single_block_residuals=None,
                 ip_adapter_extensions=neg_ip_adapter_extensions,
+                regional_prompting_extension=neg_regional_prompting_extension,
             )
             pred = neg_pred + step_cfg_scale * (pred - neg_pred)
 

invokeai/backend/flux/extensions/regional_prompting_extension.py

@@ -0,0 +1,276 @@
+from typing import Optional
+
+import torch
+import torchvision
+
+from invokeai.backend.flux.text_conditioning import FluxRegionalTextConditioning, FluxTextConditioning
+from invokeai.backend.stable_diffusion.diffusion.conditioning_data import Range
+from invokeai.backend.util.devices import TorchDevice
+from invokeai.backend.util.mask import to_standard_float_mask
+
+
+class RegionalPromptingExtension:
+    """A class for managing regional prompting with FLUX.
+
+    This implementation is inspired by https://arxiv.org/pdf/2411.02395 (though there are significant differences).
+    """
+
+    def __init__(
+        self,
+        regional_text_conditioning: FluxRegionalTextConditioning,
+        restricted_attn_mask: torch.Tensor | None = None,
+    ):
+        self.regional_text_conditioning = regional_text_conditioning
+        self.restricted_attn_mask = restricted_attn_mask
+
+    def get_double_stream_attn_mask(self, block_index: int) -> torch.Tensor | None:
+        order = [self.restricted_attn_mask, None]
+        return order[block_index % len(order)]
+
+    def get_single_stream_attn_mask(self, block_index: int) -> torch.Tensor | None:
+        order = [self.restricted_attn_mask, None]
+        return order[block_index % len(order)]
+
+    @classmethod
+    def from_text_conditioning(cls, text_conditioning: list[FluxTextConditioning], img_seq_len: int):
+        """Create a RegionalPromptingExtension from a list of text conditionings.
+
+        Args:
+            text_conditioning (list[FluxTextConditioning]): The text conditionings to use for regional prompting.
+            img_seq_len (int): The image sequence length (i.e. packed_height * packed_width).
+        """
+        regional_text_conditioning = cls._concat_regional_text_conditioning(text_conditioning)
+        attn_mask_with_restricted_img_self_attn = cls._prepare_restricted_attn_mask(
+            regional_text_conditioning, img_seq_len
+        )
+        return cls(
+            regional_text_conditioning=regional_text_conditioning,
+            restricted_attn_mask=attn_mask_with_restricted_img_self_attn,
+        )
+
+    # Keeping _prepare_unrestricted_attn_mask for reference as an alternative masking strategy:
+    #
+    # @classmethod
+    # def _prepare_unrestricted_attn_mask(
+    #     cls,
+    #     regional_text_conditioning: FluxRegionalTextConditioning,
+    #     img_seq_len: int,
+    # ) -> torch.Tensor:
+    #     """Prepare an 'unrestricted' attention mask. In this context, 'unrestricted' means that:
+    #     - img self-attention is not masked.
+    #     - img regions attend to both txt within their own region and to global prompts.
+    #     """
+    #     device = TorchDevice.choose_torch_device()
+
+    #     # Infer txt_seq_len from the t5_embeddings tensor.
+    #     txt_seq_len = regional_text_conditioning.t5_embeddings.shape[1]
+
+    #     # In the attention blocks, the txt seq and img seq are concatenated and then attention is applied.
+    #     # Concatenation happens in the following order: [txt_seq, img_seq].
+    #     # There are 4 portions of the attention mask to consider as we prepare it:
+    #     # 1. txt attends to itself
+    #     # 2. txt attends to corresponding regional img
+    #     # 3. regional img attends to corresponding txt
+    #     # 4. regional img attends to itself
+
+    #     # Initialize empty attention mask.
+    #     regional_attention_mask = torch.zeros(
+    #         (txt_seq_len + img_seq_len, txt_seq_len + img_seq_len), device=device, dtype=torch.float16
+    #     )
+
+    #     for image_mask, t5_embedding_range in zip(
+    #         regional_text_conditioning.image_masks, regional_text_conditioning.t5_embedding_ranges, strict=True
+    #     ):
+    #         # 1. txt attends to itself
+    #         regional_attention_mask[
+    #             t5_embedding_range.start : t5_embedding_range.end, t5_embedding_range.start : t5_embedding_range.end
+    #         ] = 1.0
+
+    #         # 2. txt attends to corresponding regional img
+    #         # Note that we reshape to (1, img_seq_len) to ensure broadcasting works as desired.
+    #         fill_value = image_mask.view(1, img_seq_len) if image_mask is not None else 1.0
+    #         regional_attention_mask[t5_embedding_range.start : t5_embedding_range.end, txt_seq_len:] = fill_value
+
+    #         # 3. regional img attends to corresponding txt
+    #         # Note that we reshape to (img_seq_len, 1) to ensure broadcasting works as desired.
+    #         fill_value = image_mask.view(img_seq_len, 1) if image_mask is not None else 1.0
+    #         regional_attention_mask[txt_seq_len:, t5_embedding_range.start : t5_embedding_range.end] = fill_value
+
+    #     # 4. regional img attends to itself
+    #     # Allow unrestricted img self attention.
+    #     regional_attention_mask[txt_seq_len:, txt_seq_len:] = 1.0
+
+    #     # Convert attention mask to boolean.
+    #     regional_attention_mask = regional_attention_mask > 0.5
+
+    #     return regional_attention_mask
+
+    @classmethod
+    def _prepare_restricted_attn_mask(
+        cls,
+        regional_text_conditioning: FluxRegionalTextConditioning,
+        img_seq_len: int,
+    ) -> torch.Tensor | None:
+        """Prepare a 'restricted' attention mask. In this context, 'restricted' means that:
+        - img self-attention is only allowed within regions.
+        - img regions only attend to txt within their own region, not to global prompts.
+        """
+        # Identify background region. I.e. the region that is not covered by any region masks.
+        background_region_mask: None | torch.Tensor = None
+        for image_mask in regional_text_conditioning.image_masks:
+            if image_mask is not None:
+                if background_region_mask is None:
+                    background_region_mask = torch.ones_like(image_mask)
+                background_region_mask *= 1 - image_mask
+
+        if background_region_mask is None:
+            # There are no region masks, short-circuit and return None.
+            # TODO(ryand): We could restrict txt-txt attention across multiple global prompts, but this would
+            # is a rare use case and would make the logic here significantly more complicated.
+            return None
+
+        device = TorchDevice.choose_torch_device()
+
+        # Infer txt_seq_len from the t5_embeddings tensor.
+        txt_seq_len = regional_text_conditioning.t5_embeddings.shape[1]
+
+        # In the attention blocks, the txt seq and img seq are concatenated and then attention is applied.
+        # Concatenation happens in the following order: [txt_seq, img_seq].
+        # There are 4 portions of the attention mask to consider as we prepare it:
+        # 1. txt attends to itself
+        # 2. txt attends to corresponding regional img
+        # 3. regional img attends to corresponding txt
+        # 4. regional img attends to itself
+
+        # Initialize empty attention mask.
+        regional_attention_mask = torch.zeros(
+            (txt_seq_len + img_seq_len, txt_seq_len + img_seq_len), device=device, dtype=torch.float16
+        )
+
+        for image_mask, t5_embedding_range in zip(
+            regional_text_conditioning.image_masks, regional_text_conditioning.t5_embedding_ranges, strict=True
+        ):
+            # 1. txt attends to itself
+            regional_attention_mask[
+                t5_embedding_range.start : t5_embedding_range.end, t5_embedding_range.start : t5_embedding_range.end
+            ] = 1.0
+
+            if image_mask is not None:
+                # 2. txt attends to corresponding regional img
+                # Note that we reshape to (1, img_seq_len) to ensure broadcasting works as desired.
+                regional_attention_mask[t5_embedding_range.start : t5_embedding_range.end, txt_seq_len:] = (
+                    image_mask.view(1, img_seq_len)
+                )
+
+                # 3. regional img attends to corresponding txt
+                # Note that we reshape to (img_seq_len, 1) to ensure broadcasting works as desired.
+                regional_attention_mask[txt_seq_len:, t5_embedding_range.start : t5_embedding_range.end] = (
+                    image_mask.view(img_seq_len, 1)
+                )
+
+                # 4. regional img attends to itself
+                image_mask = image_mask.view(img_seq_len, 1)
+                regional_attention_mask[txt_seq_len:, txt_seq_len:] += image_mask @ image_mask.T
+            else:
+                # We don't allow attention between non-background image regions and global prompts. This helps to ensure
+                # that regions focus on their local prompts. We do, however, allow attention between background regions
+                # and global prompts. If we didn't do this, then the background regions would not attend to any txt
+                # embeddings, which we found experimentally to cause artifacts.
+
+                # 2. global txt attends to background region
+                # Note that we reshape to (1, img_seq_len) to ensure broadcasting works as desired.
+                regional_attention_mask[t5_embedding_range.start : t5_embedding_range.end, txt_seq_len:] = (
+                    background_region_mask.view(1, img_seq_len)
+                )
+
+                # 3. background region attends to global txt
+                # Note that we reshape to (img_seq_len, 1) to ensure broadcasting works as desired.
+                regional_attention_mask[txt_seq_len:, t5_embedding_range.start : t5_embedding_range.end] = (
+                    background_region_mask.view(img_seq_len, 1)
+                )
+
+        # Allow background regions to attend to themselves.
+        regional_attention_mask[txt_seq_len:, txt_seq_len:] += background_region_mask.view(img_seq_len, 1)
+        regional_attention_mask[txt_seq_len:, txt_seq_len:] += background_region_mask.view(1, img_seq_len)
+
+        # Convert attention mask to boolean.
+        regional_attention_mask = regional_attention_mask > 0.5
+
+        return regional_attention_mask
+
+    @classmethod
+    def _concat_regional_text_conditioning(
+        cls,
+        text_conditionings: list[FluxTextConditioning],
+    ) -> FluxRegionalTextConditioning:
+        """Concatenate regional text conditioning data into a single conditioning tensor (with associated masks)."""
+        concat_t5_embeddings: list[torch.Tensor] = []
+        concat_t5_embedding_ranges: list[Range] = []
+        image_masks: list[torch.Tensor | None] = []
+
+        # Choose global CLIP embedding.
+        # Use the first global prompt's CLIP embedding as the global CLIP embedding. If there is no global prompt, use
+        # the first prompt's CLIP embedding.
+        global_clip_embedding: torch.Tensor = text_conditionings[0].clip_embeddings
+        for text_conditioning in text_conditionings:
+            if text_conditioning.mask is None:
+                global_clip_embedding = text_conditioning.clip_embeddings
+                break
+
+        cur_t5_embedding_len = 0
+        for text_conditioning in text_conditionings:
+            concat_t5_embeddings.append(text_conditioning.t5_embeddings)
+
+            concat_t5_embedding_ranges.append(
+                Range(start=cur_t5_embedding_len, end=cur_t5_embedding_len + text_conditioning.t5_embeddings.shape[1])
+            )
+
+            image_masks.append(text_conditioning.mask)
+
+            cur_t5_embedding_len += text_conditioning.t5_embeddings.shape[1]
+
+        t5_embeddings = torch.cat(concat_t5_embeddings, dim=1)
+
+        # Initialize the txt_ids tensor.
+        pos_bs, pos_t5_seq_len, _ = t5_embeddings.shape
+        t5_txt_ids = torch.zeros(
+            pos_bs, pos_t5_seq_len, 3, dtype=t5_embeddings.dtype, device=TorchDevice.choose_torch_device()
+        )
+
+        return FluxRegionalTextConditioning(
+            t5_embeddings=t5_embeddings,
+            clip_embeddings=global_clip_embedding,
+            t5_txt_ids=t5_txt_ids,
+            image_masks=image_masks,
+            t5_embedding_ranges=concat_t5_embedding_ranges,
+        )
+
+    @staticmethod
+    def preprocess_regional_prompt_mask(
+        mask: Optional[torch.Tensor], packed_height: int, packed_width: int, dtype: torch.dtype, device: torch.device
+    ) -> torch.Tensor:
+        """Preprocess a regional prompt mask to match the target height and width.
+        If mask is None, returns a mask of all ones with the target height and width.
+        If mask is not None, resizes the mask to the target height and width using 'nearest' interpolation.
+
+        packed_height and packed_width are the target height and width of the mask in the 'packed' latent space.
+
+        Returns:
+            torch.Tensor: The processed mask. shape: (1, 1, packed_height * packed_width).
+        """
+
+        if mask is None:
+            return torch.ones((1, 1, packed_height * packed_width), dtype=dtype, device=device)
+
+        mask = to_standard_float_mask(mask, out_dtype=dtype)
+
+        tf = torchvision.transforms.Resize(
+            (packed_height, packed_width), interpolation=torchvision.transforms.InterpolationMode.NEAREST
+        )
+
+        # Add a batch dimension to the mask, because torchvision expects shape (batch, channels, h, w).
+        mask = mask.unsqueeze(0)  # Shape: (1, h, w) -> (1, 1, h, w)
+        resized_mask = tf(mask)
+
+        # Flatten the height and width dimensions into a single image_seq_len dimension.
+        return resized_mask.flatten(start_dim=2)

invokeai/backend/flux/flux_tools_sampling_utils.py

@@ -0,0 +1,27 @@
+import torch
+import numpy as np
+from PIL import Image
+from einops import rearrange
+
+from invokeai.backend.flux.modules.autoencoder import AutoEncoder
+
+def prepare_control(
+    height: int,
+    width: int,
+    seed: int,
+    ae: AutoEncoder,
+    cond_image: Image.Image,
+) -> torch.Tensor:
+    # load and encode the conditioning image
+    img_cond = cond_image.convert("RGB")
+    img_cond = img_cond.resize((width, height), Image.Resampling.LANCZOS)
+    img_cond = np.array(img_cond)
+    img_cond = torch.from_numpy(img_cond).float()
+    img_cond = rearrange(img_cond, "h w c -> 1 c h w")
+    ae_dtype = next(iter(ae.parameters())).dtype
+    ae_device = next(iter(ae.parameters())).device
+    img_cond = img_cond.to(device=ae_device, dtype=ae_dtype)
+    generator = torch.Generator(device=ae_device).manual_seed(seed)
+    img_cond = ae.encode(img_cond, sample=True, generator=generator)
+    img_cond = rearrange(img_cond, "b c (h ph) (w pw) -> b (h w) (c ph pw)", ph=2, pw=2)
+    return img_cond

invokeai/backend/flux/math.py

@@ -5,10 +5,10 @@ from einops import rearrange
 from torch import Tensor
 
 
-def attention(q: Tensor, k: Tensor, v: Tensor, pe: Tensor) -> Tensor:
+def attention(q: Tensor, k: Tensor, v: Tensor, pe: Tensor, attn_mask: Tensor | None = None) -> Tensor:
     q, k = apply_rope(q, k, pe)
 
-    x = torch.nn.functional.scaled_dot_product_attention(q, k, v)
+    x = torch.nn.functional.scaled_dot_product_attention(q, k, v, attn_mask=attn_mask)
     x = rearrange(x, "B H L D -> B L (H D)")
 
     return x
@@ -24,12 +24,12 @@ def rope(pos: Tensor, dim: int, theta: int) -> Tensor:
     out = torch.einsum("...n,d->...nd", pos, omega)
     out = torch.stack([torch.cos(out), -torch.sin(out), torch.sin(out), torch.cos(out)], dim=-1)
     out = rearrange(out, "b n d (i j) -> b n d i j", i=2, j=2)
-    return out.float()
+    return out.to(dtype=pos.dtype, device=pos.device)
 
 
 def apply_rope(xq: Tensor, xk: Tensor, freqs_cis: Tensor) -> tuple[Tensor, Tensor]:
-    xq_ = xq.float().reshape(*xq.shape[:-1], -1, 1, 2)
-    xk_ = xk.float().reshape(*xk.shape[:-1], -1, 1, 2)
+    xq_ = xq.view(*xq.shape[:-1], -1, 1, 2)
+    xk_ = xk.view(*xk.shape[:-1], -1, 1, 2)
     xq_out = freqs_cis[..., 0] * xq_[..., 0] + freqs_cis[..., 1] * xq_[..., 1]
     xk_out = freqs_cis[..., 0] * xk_[..., 0] + freqs_cis[..., 1] * xk_[..., 1]
-    return xq_out.reshape(*xq.shape).type_as(xq), xk_out.reshape(*xk.shape).type_as(xk)
+    return xq_out.view(*xq.shape).type_as(xq), xk_out.view(*xk.shape).type_as(xk)

invokeai/backend/flux/model.py

@@ -4,8 +4,13 @@ from dataclasses import dataclass
 
 import torch
 from torch import Tensor, nn
+from typing import Optional
 
-from invokeai.backend.flux.custom_block_processor import CustomDoubleStreamBlockProcessor
+from invokeai.backend.flux.custom_block_processor import (
+    CustomDoubleStreamBlockProcessor,
+    CustomSingleStreamBlockProcessor,
+)
+from invokeai.backend.flux.extensions.regional_prompting_extension import RegionalPromptingExtension
 from invokeai.backend.flux.extensions.xlabs_ip_adapter_extension import XLabsIPAdapterExtension
 from invokeai.backend.flux.modules.layers import (
     DoubleStreamBlock,
@@ -31,6 +36,7 @@ class FluxParams:
     theta: int
     qkv_bias: bool
     guidance_embed: bool
+    out_channels: Optional[int] = None
 
 
 class Flux(nn.Module):
@@ -43,7 +49,7 @@ class Flux(nn.Module):
 
         self.params = params
         self.in_channels = params.in_channels
-        self.out_channels = self.in_channels
+        self.out_channels = params.out_channels or self.in_channels
         if params.hidden_size % params.num_heads != 0:
             raise ValueError(f"Hidden size {params.hidden_size} must be divisible by num_heads {params.num_heads}")
         pe_dim = params.hidden_size // params.num_heads
@@ -95,6 +101,7 @@ class Flux(nn.Module):
         controlnet_double_block_residuals: list[Tensor] | None,
         controlnet_single_block_residuals: list[Tensor] | None,
         ip_adapter_extensions: list[XLabsIPAdapterExtension],
+        regional_prompting_extension: RegionalPromptingExtension,
     ) -> Tensor:
         if img.ndim != 3 or txt.ndim != 3:
             raise ValueError("Input img and txt tensors must have 3 dimensions.")
@@ -117,7 +124,6 @@ class Flux(nn.Module):
             assert len(controlnet_double_block_residuals) == len(self.double_blocks)
         for block_index, block in enumerate(self.double_blocks):
             assert isinstance(block, DoubleStreamBlock)
-
             img, txt = CustomDoubleStreamBlockProcessor.custom_double_block_forward(
                 timestep_index=timestep_index,
                 total_num_timesteps=total_num_timesteps,
@@ -128,6 +134,7 @@ class Flux(nn.Module):
                 vec=vec,
                 pe=pe,
                 ip_adapter_extensions=ip_adapter_extensions,
+                regional_prompting_extension=regional_prompting_extension,
             )
 
             if controlnet_double_block_residuals is not None:
@@ -140,7 +147,17 @@ class Flux(nn.Module):
             assert len(controlnet_single_block_residuals) == len(self.single_blocks)
 
         for block_index, block in enumerate(self.single_blocks):
-            img = block(img, vec=vec, pe=pe)
+            assert isinstance(block, SingleStreamBlock)
+            img = CustomSingleStreamBlockProcessor.custom_single_block_forward(
+                timestep_index=timestep_index,
+                total_num_timesteps=total_num_timesteps,
+                block_index=block_index,
+                block=block,
+                img=img,
+                vec=vec,
+                pe=pe,
+                regional_prompting_extension=regional_prompting_extension,
+            )
 
             if controlnet_single_block_residuals is not None:
                 img[:, txt.shape[1] :, ...] += controlnet_single_block_residuals[block_index]

invokeai/backend/flux/modules/image_embedders.py

@@ -0,0 +1,50 @@
+import os
+import cv2
+import numpy as np
+import torch
+
+from einops import rearrange, repeat
+from PIL import Image
+from safetensors.torch import load_file as load_sft
+from torch import nn
+from transformers import AutoModelForDepthEstimation, AutoProcessor, SiglipImageProcessor, SiglipVisionModel
+
+class DepthImageEncoder:
+    depth_model_name = "LiheYoung/depth-anything-large-hf"
+    def __init__(self, device):
+        self.device = device
+        self.depth_model = AutoModelForDepthEstimation.from_pretrained(self.depth_model_name).to(device)
+        self.processor = AutoProcessor.from_pretrained(self.depth_model_name)
+    def __call__(self, img: torch.Tensor) -> torch.Tensor:
+        hw = img.shape[-2:]
+        img = torch.clamp(img, -1.0, 1.0)
+        img_byte = ((img + 1.0) * 127.5).byte()
+        img = self.processor(img_byte, return_tensors="pt")["pixel_values"]
+        depth = self.depth_model(img.to(self.device)).predicted_depth
+        depth = repeat(depth, "b h w -> b 3 h w")
+        depth = torch.nn.functional.interpolate(depth, hw, mode="bicubic", antialias=True)
+        depth = depth / 127.5 - 1.0
+        return depth
+
+class CannyImageEncoder:
+    def __init__(
+        self,
+        device,
+        min_t: int = 50,
+        max_t: int = 200,
+    ):
+        self.device = device
+        self.min_t = min_t
+        self.max_t = max_t
+    def __call__(self, img: torch.Tensor) -> torch.Tensor:
+        assert img.shape[0] == 1, "Only batch size 1 is supported"
+        img = rearrange(img[0], "c h w -> h w c")
+        img = torch.clamp(img, -1.0, 1.0)
+        img_np = ((img + 1.0) * 127.5).numpy().astype(np.uint8)
+        # Apply Canny edge detection
+        canny = cv2.Canny(img_np, self.min_t, self.max_t)
+        # Convert back to torch tensor and reshape
+        canny = torch.from_numpy(canny).float() / 127.5 - 1.0
+        canny = rearrange(canny, "h w -> 1 1 h w")
+        canny = repeat(canny, "b 1 ... -> b 3 ...")
+        return canny.to(self.device)

invokeai/backend/flux/modules/layers.py

@@ -66,10 +66,7 @@ class RMSNorm(torch.nn.Module):
         self.scale = nn.Parameter(torch.ones(dim))
 
     def forward(self, x: Tensor):
-        x_dtype = x.dtype
-        x = x.float()
-        rrms = torch.rsqrt(torch.mean(x**2, dim=-1, keepdim=True) + 1e-6)
-        return (x * rrms).to(dtype=x_dtype) * self.scale
+        return torch.nn.functional.rms_norm(x, self.scale.shape, self.scale, eps=1e-6)
 
 
 class QKNorm(torch.nn.Module):

invokeai/backend/flux/text_conditioning.py

@@ -0,0 +1,36 @@
+from dataclasses import dataclass
+
+import torch
+
+from invokeai.backend.stable_diffusion.diffusion.conditioning_data import Range
+
+
+@dataclass
+class FluxTextConditioning:
+    t5_embeddings: torch.Tensor
+    clip_embeddings: torch.Tensor
+    # If mask is None, the prompt is a global prompt.
+    mask: torch.Tensor | None
+
+
+@dataclass
+class FluxRegionalTextConditioning:
+    # Concatenated text embeddings.
+    # Shape: (1, concatenated_txt_seq_len, 4096)
+    t5_embeddings: torch.Tensor
+    # Shape: (1, concatenated_txt_seq_len, 3)
+    t5_txt_ids: torch.Tensor
+
+    # Global CLIP embeddings.
+    # Shape: (1, 768)
+    clip_embeddings: torch.Tensor
+
+    # A binary mask indicating the regions of the image that the prompt should be applied to. If None, the prompt is a
+    # global prompt.
+    # image_masks[i] is the mask for the ith prompt.
+    # image_masks[i] has shape (1, image_seq_len) and dtype torch.bool.
+    image_masks: list[torch.Tensor | None]
+
+    # List of ranges that represent the embedding ranges for each mask.
+    # t5_embedding_ranges[i] contains the range of the t5 embeddings that correspond to image_masks[i].
+    t5_embedding_ranges: list[Range]

invokeai/backend/lora/conversions/flux_control_lora_utils.py

@@ -0,0 +1,65 @@
+import re
+import torch
+
+from typing import Any, Dict
+from invokeai.backend.lora.layers.any_lora_layer import AnyLoRALayer
+from invokeai.backend.lora.layers.utils import any_lora_layer_from_state_dict
+from invokeai.backend.lora.lora_model_raw import LoRAModelRaw
+from invokeai.backend.lora.conversions.flux_lora_constants import FLUX_LORA_TRANSFORMER_PREFIX
+from invokeai.backend.lora.layers.lora_layer import LoRALayer
+from invokeai.backend.lora.layers.set_parameter_layer import SetParameterLayer
+
+
+# A regex pattern that matches all of the keys in the Flux Dev/Canny LoRA format.
+# Example keys:
+#   guidance_in.in_layer.lora_B.bias
+#   single_blocks.0.linear1.lora_A.weight
+#   double_blocks.0.img_attn.norm.key_norm.scale
+FLUX_STRUCTURAL_TRANSFORMER_KEY_REGEX = r"(final_layer|vector_in|txt_in|time_in|img_in|guidance_in|\w+_blocks)(\.(\d+))?\.(lora_(A|B)|(in|out)_layer|adaLN_modulation|img_attn|img_mlp|img_mod|txt_attn|txt_mlp|txt_mod|linear|linear1|linear2|modulation|norm)\.?(.*)"
+
+def is_state_dict_likely_flux_control(state_dict: Dict[str, Any]) -> bool:
+    """Checks if the provided state dict is likely in the FLUX Control 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.)
+    """
+    return all(
+        re.match(FLUX_STRUCTURAL_TRANSFORMER_KEY_REGEX, k) or re.match(FLUX_STRUCTURAL_TRANSFORMER_KEY_REGEX, k)
+        for k in state_dict.keys()
+    )
+
+def lora_model_from_flux_control_state_dict(state_dict: Dict[str, torch.Tensor]) -> LoRAModelRaw:
+    # converted_state_dict = _convert_lora_bfl_control(state_dict=state_dict)
+    # Group keys by layer.
+    grouped_state_dict: dict[str, dict[str, torch.Tensor]] = {}
+    for key, value in state_dict.items():
+        key_props = key.split(".")
+        # Got it loading using lora_down and lora_up but it didn't seem to match this lora's structure
+        # Leaving this in since it doesn't hurt anything and may be better
+        layer_prop_size = -2 if any(prop in key for prop in ["lora_B", "lora_A"]) else -1
+        layer_name = ".".join(key_props[:layer_prop_size])
+        param_name = ".".join(key_props[layer_prop_size:])
+        if layer_name not in grouped_state_dict:
+            grouped_state_dict[layer_name] = {}
+        grouped_state_dict[layer_name][param_name] = value
+
+    # Create LoRA layers.
+    layers: dict[str, AnyLoRALayer] = {}
+    for layer_key, layer_state_dict in grouped_state_dict.items():
+        # Convert to a full layer diff
+        prefixed_key = f"{FLUX_LORA_TRANSFORMER_PREFIX}{layer_key}"
+        if all(k in layer_state_dict for k in ["lora_A.weight", "lora_B.bias", "lora_B.weight"]):
+            layers[prefixed_key] = LoRALayer(
+                layer_state_dict["lora_B.weight"],
+                None,
+                layer_state_dict["lora_A.weight"],
+                None,
+                layer_state_dict["lora_B.bias"]
+            )
+        elif "scale" in layer_state_dict:
+            layers[prefixed_key] = SetParameterLayer("scale", layer_state_dict["scale"])
+        else:
+            raise AssertionError(f"{layer_key} not expected")
+    # Create and return the LoRAModelRaw.
+    return LoRAModelRaw(layers=layers)
+

invokeai/backend/lora/layers/any_lora_layer.py

@@ -7,5 +7,6 @@ from invokeai.backend.lora.layers.loha_layer import LoHALayer
 from invokeai.backend.lora.layers.lokr_layer import LoKRLayer
 from invokeai.backend.lora.layers.lora_layer import LoRALayer
 from invokeai.backend.lora.layers.norm_layer import NormLayer
+from invokeai.backend.lora.layers.set_parameter_layer import SetParameterLayer
 
-AnyLoRALayer = Union[LoRALayer, LoHALayer, LoKRLayer, FullLayer, IA3Layer, NormLayer, ConcatenatedLoRALayer]
+AnyLoRALayer = Union[LoRALayer, LoHALayer, LoKRLayer, FullLayer, IA3Layer, NormLayer, ConcatenatedLoRALayer, SetParameterLayer]

invokeai/backend/lora/layers/reshape_weight_layer.py

@@ -0,0 +1,34 @@
+from typing import Dict, Optional
+
+import torch
+
+from invokeai.backend.lora.layers.lora_layer_base import LoRALayerBase
+from invokeai.backend.util.calc_tensor_size import calc_tensor_size
+
+
+class ReshapeWeightLayer(LoRALayerBase):
+    # TODO: Just everything in this class 
+    def __init__(self, weight: Optional[torch.Tensor], bias: Optional[torch.Tensor], scale: Optional[torch.Tensor]):
+        super().__init__(alpha=None, bias=bias)
+        self.weight = torch.nn.Parameter(weight) if weight is not None else None
+        self.bias = torch.nn.Parameter(bias) if bias is not None else None
+        self.manual_scale = scale
+
+    def scale(self):
+        return self.manual_scale.float() if self.manual_scale is not None else super().scale()
+
+    def rank(self) -> int | None:
+        return None
+
+    def get_weight(self, orig_weight: torch.Tensor) -> torch.Tensor:
+        return orig_weight
+
+    def to(self, device: torch.device | None = None, dtype: torch.dtype | None = None):
+        super().to(device=device, dtype=dtype)
+        if self.weight is not None:
+            self.weight = self.weight.to(device=device, dtype=dtype)
+        if self.manual_scale is not None:
+            self.manual_scale = self.manual_scale.to(device=device, dtype=dtype)
+
+    def calc_size(self) -> int:
+        return super().calc_size() + calc_tensor_size(self.manual_scale)

invokeai/backend/lora/layers/set_parameter_layer.py

@@ -0,0 +1,29 @@
+from typing import Dict, Optional
+
+import torch
+
+from invokeai.backend.lora.layers.lora_layer_base import LoRALayerBase
+from invokeai.backend.util.calc_tensor_size import calc_tensor_size
+
+
+class SetParameterLayer(LoRALayerBase):
+    def __init__(self, param_name: str, weight: torch.Tensor):
+        super().__init__(None, None)
+        self.weight = weight
+        self.param_name = param_name
+
+    def rank(self) -> int | None:
+        return None
+
+    def get_weight(self, orig_weight: torch.Tensor) -> torch.Tensor:
+        return self.weight - orig_weight
+    
+    def get_parameters(self, orig_module: torch.nn.Module) -> Dict[str, torch.Tensor]:
+        return {self.param_name: self.get_weight(orig_module.get_parameter(self.param_name))}
+
+    def to(self, device: torch.device | None = None, dtype: torch.dtype | None = None):
+        super().to(device=device, dtype=dtype)
+        self.weight = self.weight.to(device=device, dtype=dtype)
+
+    def calc_size(self) -> int:
+        return super().calc_size() + calc_tensor_size(self.weight)

invokeai/backend/lora/layers/utils.py

@@ -9,6 +9,7 @@ from invokeai.backend.lora.layers.loha_layer import LoHALayer
 from invokeai.backend.lora.layers.lokr_layer import LoKRLayer
 from invokeai.backend.lora.layers.lora_layer import LoRALayer
 from invokeai.backend.lora.layers.norm_layer import NormLayer
+from invokeai.backend.lora.layers.set_parameter_layer import SetParameterLayer
 
 
 def any_lora_layer_from_state_dict(state_dict: Dict[str, torch.Tensor]) -> AnyLoRALayer:

invokeai/backend/lora/lora_layer_wrappers.py

@@ -0,0 +1,133 @@
+import torch
+
+from invokeai.backend.lora.layers.any_lora_layer import AnyLoRALayer
+from invokeai.backend.lora.layers.concatenated_lora_layer import ConcatenatedLoRALayer
+from invokeai.backend.lora.layers.lora_layer import LoRALayer
+
+
+class LoRASidecarWrapper(torch.nn.Module):
+    def __init__(self, orig_module: torch.nn.Module, lora_layers: list[AnyLoRALayer], lora_weights: list[float]):
+        super().__init__()
+        self._orig_module = orig_module
+        self._lora_layers = lora_layers
+        self._lora_weights = lora_weights
+
+    @property
+    def orig_module(self) -> torch.nn.Module:
+        return self._orig_module
+
+    def add_lora_layer(self, lora_layer: AnyLoRALayer, lora_weight: float):
+        self._lora_layers.append(lora_layer)
+        self._lora_weights.append(lora_weight)
+
+    @torch.no_grad()
+    def _get_lora_patched_parameters(
+        self, orig_params: dict[str, torch.Tensor], lora_layers: list[AnyLoRALayer], lora_weights: list[float]
+    ) -> dict[str, torch.Tensor]:
+        params: dict[str, torch.Tensor] = {}
+        for lora_layer, lora_weight in zip(lora_layers, lora_weights, strict=True):
+            layer_params = lora_layer.get_parameters(self._orig_module)
+            for param_name, param_weight in layer_params.items():
+                if orig_params[param_name].shape != param_weight.shape:
+                    param_weight = param_weight.reshape(orig_params[param_name].shape)
+
+                if param_name not in params:
+                    params[param_name] = param_weight * (lora_layer.scale() * lora_weight)
+                else:
+                    params[param_name] += param_weight * (lora_layer.scale() * lora_weight)
+
+        return params
+
+
+class LoRALinearWrapper(LoRASidecarWrapper):
+    def _lora_linear_forward(self, input: torch.Tensor, lora_layer: LoRALayer, lora_weight: float) -> torch.Tensor:
+        """An optimized implementation of the residual calculation for a Linear LoRALayer."""
+        x = torch.nn.functional.linear(input, lora_layer.down)
+        if lora_layer.mid is not None:
+            x = torch.nn.functional.linear(x, lora_layer.mid)
+        x = torch.nn.functional.linear(x, lora_layer.up, bias=lora_layer.bias)
+        x *= lora_weight * lora_layer.scale()
+        return x
+
+    def _concatenated_lora_forward(
+        self, input: torch.Tensor, concatenated_lora_layer: ConcatenatedLoRALayer, lora_weight: float
+    ) -> torch.Tensor:
+        """An optimized implementation of the residual calculation for a Linear 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 forward(self, input: torch.Tensor) -> torch.Tensor:
+        # Split the LoRA layers into those that have optimized implementations and those that don't.
+        optimized_layer_types = (LoRALayer, ConcatenatedLoRALayer)
+        optimized_layers = [
+            (layer, weight)
+            for layer, weight in zip(self._lora_layers, self._lora_weights, strict=True)
+            if isinstance(layer, optimized_layer_types)
+        ]
+        non_optimized_layers = [
+            (layer, weight)
+            for layer, weight in zip(self._lora_layers, self._lora_weights, strict=True)
+            if not isinstance(layer, optimized_layer_types)
+        ]
+
+        # First, calculate the residual for LoRA layers for which there is an optimized implementation.
+        residual = None
+        for lora_layer, lora_weight in optimized_layers:
+            if isinstance(lora_layer, LoRALayer):
+                added_residual = self._lora_linear_forward(input, lora_layer, lora_weight)
+            elif isinstance(lora_layer, ConcatenatedLoRALayer):
+                added_residual = self._concatenated_lora_forward(input, lora_layer, lora_weight)
+            else:
+                raise ValueError(f"Unsupported LoRA layer type: {type(lora_layer)}")
+
+            if residual is None:
+                residual = added_residual
+            else:
+                residual += added_residual
+
+        # Next, calculate the residuals for the LoRA layers for which there is no optimized implementation.
+        if non_optimized_layers:
+            unoptimized_layers, unoptimized_weights = zip(*non_optimized_layers, strict=True)
+            params = self._get_lora_patched_parameters(
+                orig_params={"weight": self._orig_module.weight, "bias": self._orig_module.bias},
+                lora_layers=unoptimized_layers,
+                lora_weights=unoptimized_weights,
+            )
+            added_residual = torch.nn.functional.linear(input, params["weight"], params.get("bias", None))
+            if residual is None:
+                residual = added_residual
+            else:
+                residual += added_residual
+
+        return self.orig_module(input) + residual
+
+
+class LoRAConv1dWrapper(LoRASidecarWrapper):
+    def forward(self, input: torch.Tensor) -> torch.Tensor:
+        params = self._get_lora_patched_parameters(
+            orig_params={"weight": self._orig_module.weight, "bias": self._orig_module.bias},
+            lora_layers=self._lora_layers,
+            lora_weights=self._lora_weights,
+        )
+        return self.orig_module(input) + torch.nn.functional.conv1d(input, params["weight"], params.get("bias", None))
+
+
+class LoRAConv2dWrapper(LoRASidecarWrapper):
+    def forward(self, input: torch.Tensor) -> torch.Tensor:
+        params = self._get_lora_patched_parameters(
+            orig_params={"weight": self._orig_module.weight, "bias": self._orig_module.bias},
+            lora_layers=self._lora_layers,
+            lora_weights=self._lora_weights,
+        )
+        return self.orig_module(input) + torch.nn.functional.conv2d(input, params["weight"], params.get("bias", None))

invokeai/backend/lora/lora_patcher.py

@@ -4,19 +4,126 @@ from typing import Dict, Iterable, Optional, Tuple
 import torch
 
 from invokeai.backend.lora.layers.any_lora_layer import AnyLoRALayer
-from invokeai.backend.lora.layers.concatenated_lora_layer import ConcatenatedLoRALayer
-from invokeai.backend.lora.layers.lora_layer import LoRALayer
-from invokeai.backend.lora.lora_model_raw import LoRAModelRaw
-from invokeai.backend.lora.sidecar_layers.concatenated_lora.concatenated_lora_linear_sidecar_layer import (
-    ConcatenatedLoRALinearSidecarLayer,
+from invokeai.backend.lora.lora_layer_wrappers import (
+    LoRAConv1dWrapper,
+    LoRAConv2dWrapper,
+    LoRALinearWrapper,
+    LoRASidecarWrapper,
 )
-from invokeai.backend.lora.sidecar_layers.lora.lora_linear_sidecar_layer import LoRALinearSidecarLayer
-from invokeai.backend.lora.sidecar_layers.lora_sidecar_module import LoRASidecarModule
+from invokeai.backend.lora.lora_model_raw import LoRAModelRaw
 from invokeai.backend.util.devices import TorchDevice
 from invokeai.backend.util.original_weights_storage import OriginalWeightsStorage
 
 
 class LoRAPatcher:
+    @staticmethod
+    @torch.no_grad()
+    @contextmanager
+    def apply_smart_lora_patches(
+        model: torch.nn.Module,
+        patches: Iterable[Tuple[LoRAModelRaw, float]],
+        prefix: str,
+        dtype: torch.dtype,
+        cached_weights: Optional[Dict[str, torch.Tensor]] = None,
+    ):
+        """Apply 'smart' LoRA patching that chooses whether to use direct patching or a sidecar wrapper for each module."""
+
+        # original_weights are stored for unpatching layers that are directly patched.
+        original_weights = OriginalWeightsStorage(cached_weights)
+        # original_modules are stored for unpatching layers that are wrapped in a LoRASidecarWrapper.
+        original_modules: dict[str, torch.nn.Module] = {}
+        try:
+            for patch, patch_weight in patches:
+                LoRAPatcher._apply_smart_lora_patch(
+                    model=model,
+                    prefix=prefix,
+                    patch=patch,
+                    patch_weight=patch_weight,
+                    original_weights=original_weights,
+                    original_modules=original_modules,
+                    dtype=dtype,
+                )
+
+            yield
+        finally:
+            # Restore directly patched layers.
+            for param_key, weight in original_weights.get_changed_weights():
+                model.get_parameter(param_key).copy_(weight)
+
+            # Restore LoRASidecarWrapper modules.
+            # Note: This logic assumes no nested modules in original_modules.
+            for module_key, orig_module in original_modules.items():
+                module_parent_key, module_name = LoRAPatcher._split_parent_key(module_key)
+                parent_module = model.get_submodule(module_parent_key)
+                LoRAPatcher._set_submodule(parent_module, module_name, orig_module)
+
+    @staticmethod
+    @torch.no_grad()
+    def _apply_smart_lora_patch(
+        model: torch.nn.Module,
+        prefix: str,
+        patch: LoRAModelRaw,
+        patch_weight: float,
+        original_weights: OriginalWeightsStorage,
+        original_modules: dict[str, torch.nn.Module],
+        dtype: torch.dtype,
+    ):
+        """Apply a single LoRA patch to a model using the 'smart' patching strategy that chooses whether to use direct
+        patching or a sidecar wrapper for each module.
+        """
+        if patch_weight == 0:
+            return
+
+        # If the layer keys contain a dot, then they are not flattened, and can be directly used to access model
+        # submodules. If the layer keys do not contain a dot, then they are flattened, meaning that all '.' have been
+        # replaced with '_'. Non-flattened keys are preferred, because they allow submodules to be accessed directly
+        # without searching, but some legacy code still uses flattened keys.
+        layer_keys_are_flattened = "." not in next(iter(patch.layers.keys()))
+
+        prefix_len = len(prefix)
+
+        for layer_key, layer in patch.layers.items():
+            if not layer_key.startswith(prefix):
+                continue
+
+            module_key, module = LoRAPatcher._get_submodule(
+                model, layer_key[prefix_len:], layer_key_is_flattened=layer_keys_are_flattened
+            )
+
+            # Decide whether to use direct patching or a sidecar wrapper.
+            # Direct patching is preferred, because it results in better runtime speed.
+            # Reasons to use sidecar patching:
+            # - The module is already wrapped in a LoRASidecarWrapper.
+            # - The module is quantized.
+            # - The module is on the CPU (and we don't want to store a second full copy of the original weights on the
+            #   CPU, since this would double the RAM usage)
+            # NOTE: For now, we don't check if the layer is quantized here. We assume that this is checked in the caller
+            # and that the caller will use the 'apply_lora_wrapper_patches' method if the layer is quantized.
+            # TODO(ryand): Handle the case where we are running without a GPU. Should we set a config flag that allows
+            # forcing full patching even on the CPU?
+            if isinstance(module, LoRASidecarWrapper) or LoRAPatcher._is_any_part_of_layer_on_cpu(module):
+                LoRAPatcher._apply_lora_layer_wrapper_patch(
+                    model=model,
+                    module_to_patch=module,
+                    module_to_patch_key=module_key,
+                    patch=layer,
+                    patch_weight=patch_weight,
+                    original_modules=original_modules,
+                    dtype=dtype,
+                )
+            else:
+                LoRAPatcher._apply_lora_layer_patch(
+                    module_to_patch=module,
+                    module_to_patch_key=module_key,
+                    patch=layer,
+                    patch_weight=patch_weight,
+                    original_weights=original_weights,
+                )
+
+    @staticmethod
+    def _is_any_part_of_layer_on_cpu(layer: torch.nn.Module) -> bool:
+        return any(p.device.type == "cpu" for p in layer.parameters())
+
     @staticmethod
     @torch.no_grad()
     @contextmanager
@@ -40,7 +147,7 @@ class LoRAPatcher:
         original_weights = OriginalWeightsStorage(cached_weights)
         try:
             for patch, patch_weight in patches:
-                LoRAPatcher.apply_lora_patch(
+                LoRAPatcher._apply_lora_patch(
                     model=model,
                     prefix=prefix,
                     patch=patch,
@@ -52,11 +159,12 @@ class LoRAPatcher:
             yield
         finally:
             for param_key, weight in original_weights.get_changed_weights():
-                model.get_parameter(param_key).copy_(weight)
+                cur_param = model.get_parameter(param_key)
+                cur_param.data = weight.to(dtype=cur_param.dtype, device=cur_param.device, copy=True)
 
     @staticmethod
     @torch.no_grad()
-    def apply_lora_patch(
+    def _apply_lora_patch(
         model: torch.nn.Module,
         prefix: str,
         patch: LoRAModelRaw,
@@ -91,48 +199,84 @@ class LoRAPatcher:
                 model, layer_key[prefix_len:], layer_key_is_flattened=layer_keys_are_flattened
             )
 
-            # All of the LoRA weight calculations will be done on the same device as the module weight.
-            # (Performance will be best if this is a CUDA device.)
-            device = module.weight.device
-            dtype = module.weight.dtype
+            LoRAPatcher._apply_lora_layer_patch(
+                module_to_patch=module,
+                module_to_patch_key=module_key,
+                patch=layer,
+                patch_weight=patch_weight,
+                original_weights=original_weights,
+            )
 
-            layer_scale = layer.scale()
+    @staticmethod
+    @torch.no_grad()
+    def _apply_lora_layer_patch(
+        module_to_patch: torch.nn.Module,
+        module_to_patch_key: str,
+        patch: AnyLoRALayer,
+        patch_weight: float,
+        original_weights: OriginalWeightsStorage,
+    ):
+        # All of the LoRA weight calculations will be done on the same device as the module weight.
+        # (Performance will be best if this is a CUDA device.)
+        first_param = next(module_to_patch.parameters())
+        device = first_param.device
+        dtype = first_param.dtype
 
-            # We intentionally move to the target device first, then cast. Experimentally, this was found to
-            # be significantly faster for 16-bit CPU tensors being moved to a CUDA device than doing the
-            # same thing in a single call to '.to(...)'.
-            layer.to(device=device)
-            layer.to(dtype=torch.float32)
+        layer_scale = patch.scale()
 
-            # TODO(ryand): Using torch.autocast(...) over explicit casting may offer a speed benefit on CUDA
-            # devices here. Experimentally, it was found to be very slow on CPU. More investigation needed.
-            for param_name, lora_param_weight in layer.get_parameters(module).items():
-                param_key = module_key + "." + param_name
-                module_param = module.get_parameter(param_name)
+        # We intentionally move to the target device first, then cast. Experimentally, this was found to
+        # be significantly faster for 16-bit CPU tensors being moved to a CUDA device than doing the
+        # same thing in a single call to '.to(...)'.
+        patch.to(device=device)
+        patch.to(dtype=torch.float32)
 
-                # Save original weight
-                original_weights.save(param_key, module_param)
+        # TODO(ryand): Using torch.autocast(...) over explicit casting may offer a speed benefit on CUDA
+        # devices here. Experimentally, it was found to be very slow on CPU. More investigation needed.
+        for param_name, lora_param_weight in patch.get_parameters(module_to_patch).items():
+            param_key = module_to_patch_key + "." + param_name
+            module_param = module_to_patch.get_parameter(param_name)
 
-                if module_param.shape != lora_param_weight.shape:
+            # Save original weight
+            original_weights.save(param_key, module_param)
+
+            if module_param.shape != lora_param_weight.shape:
+                if module_param.nelement() == lora_param_weight.nelement():
                     lora_param_weight = lora_param_weight.reshape(module_param.shape)
+                else:
+                    # This condition was added to handle layers in FLUX control LoRAs.
+                    # TODO(ryand): Move the weight update into the LoRA layer so that the LoRAPatcher doesn't need
+                    # to worry about this?
+                    expanded_weight = torch.zeros_like(
+                        lora_param_weight, dtype=module_param.dtype, device=module_param.device
+                    )
+                    slices = tuple(slice(0, dim) for dim in module_param.shape)
+                    expanded_weight[slices] = module_param
+                    setattr(
+                        module,
+                        param_name,
+                        torch.nn.Parameter(expanded_weight, requires_grad=module_param.requires_grad),
+                    )
+                    module_param = expanded_weight
 
-                lora_param_weight *= patch_weight * layer_scale
-                module_param += lora_param_weight.to(dtype=dtype)
+            lora_param_weight *= patch_weight * layer_scale
+            module_param += lora_param_weight.to(dtype=dtype)
 
-            layer.to(device=TorchDevice.CPU_DEVICE)
+        patch.to(device=TorchDevice.CPU_DEVICE)
 
     @staticmethod
     @torch.no_grad()
     @contextmanager
-    def apply_lora_sidecar_patches(
+    def apply_lora_wrapper_patches(
         model: torch.nn.Module,
         patches: Iterable[Tuple[LoRAModelRaw, float]],
         prefix: str,
         dtype: torch.dtype,
     ):
-        """Apply one or more LoRA sidecar patches to a model within a context manager. Sidecar patches incur some
-        overhead compared to normal LoRA patching, but they allow for LoRA layers to applied to base layers in any
-        quantization format.
+        """Apply one or more LoRA wrapper patches to a model within a context manager. Wrapper patches incur some
+        runtime overhead compared to normal LoRA patching, but they enable:
+        - LoRA layers to be applied to quantized models
+        - LoRA layers to be applied to CPU layers without needing to store a full copy of the original weights (i.e.
+          avoid doubling the memory requirements).
 
         Args:
             model (torch.nn.Module): The model to patch.
@@ -140,14 +284,11 @@ class LoRAPatcher:
                 associated weights. An iterator is used so that the LoRA patches do not need to be loaded into memory
                 all at once.
             prefix (str): The keys in the patches will be filtered to only include weights with this prefix.
-            dtype (torch.dtype): The compute dtype of the sidecar layers. This cannot easily be inferred from the model,
-                since the sidecar layers are typically applied on top of quantized layers whose weight dtype is
-                different from their compute dtype.
         """
         original_modules: dict[str, torch.nn.Module] = {}
         try:
             for patch, patch_weight in patches:
-                LoRAPatcher._apply_lora_sidecar_patch(
+                LoRAPatcher._apply_lora_wrapper_patch(
                     model=model,
                     prefix=prefix,
                     patch=patch,
@@ -165,7 +306,7 @@ class LoRAPatcher:
                 LoRAPatcher._set_submodule(parent_module, module_name, orig_module)
 
     @staticmethod
-    def _apply_lora_sidecar_patch(
+    def _apply_lora_wrapper_patch(
         model: torch.nn.Module,
         patch: LoRAModelRaw,
         patch_weight: float,
@@ -173,7 +314,7 @@ class LoRAPatcher:
         original_modules: dict[str, torch.nn.Module],
         dtype: torch.dtype,
     ):
-        """Apply a single LoRA sidecar patch to a model."""
+        """Apply a single LoRA wrapper patch to a model."""
 
         if patch_weight == 0:
             return
@@ -194,28 +335,47 @@ class LoRAPatcher:
                 model, layer_key[prefix_len:], layer_key_is_flattened=layer_keys_are_flattened
             )
 
-            # Initialize the LoRA sidecar layer.
-            lora_sidecar_layer = LoRAPatcher._initialize_lora_sidecar_layer(module, layer, patch_weight)
+            LoRAPatcher._apply_lora_layer_wrapper_patch(
+                model=model,
+                module_to_patch=module,
+                module_to_patch_key=module_key,
+                patch=layer,
+                patch_weight=patch_weight,
+                original_modules=original_modules,
+                dtype=dtype,
+            )
 
-            # Replace the original module with a LoRASidecarModule if it has not already been done.
-            if module_key in original_modules:
-                # The module has already been patched with a LoRASidecarModule. Append to it.
-                assert isinstance(module, LoRASidecarModule)
-                lora_sidecar_module = module
-            else:
-                # The module has not yet been patched with a LoRASidecarModule. Create one.
-                lora_sidecar_module = LoRASidecarModule(module, [])
-                original_modules[module_key] = module
-                module_parent_key, module_name = LoRAPatcher._split_parent_key(module_key)
-                module_parent = model.get_submodule(module_parent_key)
-                LoRAPatcher._set_submodule(module_parent, module_name, lora_sidecar_module)
+    @staticmethod
+    @torch.no_grad()
+    def _apply_lora_layer_wrapper_patch(
+        model: torch.nn.Module,
+        module_to_patch: torch.nn.Module,
+        module_to_patch_key: str,
+        patch: AnyLoRALayer,
+        patch_weight: float,
+        original_modules: dict[str, torch.nn.Module],
+        dtype: torch.dtype,
+    ):
+        """Apply a single LoRA wrapper patch to a model."""
 
-            # Move the LoRA sidecar layer to the same device/dtype as the orig module.
-            # TODO(ryand): Experiment with moving to the device first, then casting. This could be faster.
-            lora_sidecar_layer.to(device=lora_sidecar_module.orig_module.weight.device, dtype=dtype)
+        # Replace the original module with a LoRASidecarWrapper if it has not already been done.
+        if not isinstance(module_to_patch, LoRASidecarWrapper):
+            lora_wrapper_layer = LoRAPatcher._initialize_lora_wrapper_layer(module_to_patch)
+            original_modules[module_to_patch_key] = module_to_patch
+            module_parent_key, module_name = LoRAPatcher._split_parent_key(module_to_patch_key)
+            module_parent = model.get_submodule(module_parent_key)
+            LoRAPatcher._set_submodule(module_parent, module_name, lora_wrapper_layer)
+            orig_module = module_to_patch
+        else:
+            assert module_to_patch_key in original_modules
+            lora_wrapper_layer = module_to_patch
+            orig_module = module_to_patch.orig_module
 
-            # Add the LoRA sidecar layer to the LoRASidecarModule.
-            lora_sidecar_module.add_lora_layer(lora_sidecar_layer)
+        # Move the LoRA layer to the same device/dtype as the orig module.
+        patch.to(device=orig_module.weight.device, dtype=dtype)
+
+        # Add the LoRA wrapper layer to the LoRASidecarWrapper.
+        lora_wrapper_layer.add_lora_layer(patch, patch_weight)
 
     @staticmethod
     def _split_parent_key(module_key: str) -> tuple[str, str]:
@@ -236,17 +396,13 @@ class LoRAPatcher:
             raise ValueError(f"Invalid module key: {module_key}")
 
     @staticmethod
-    def _initialize_lora_sidecar_layer(orig_layer: torch.nn.Module, lora_layer: AnyLoRALayer, patch_weight: float):
-        # TODO(ryand): Add support for more original layer types and LoRA layer types.
-        if isinstance(orig_layer, torch.nn.Linear) or (
-            isinstance(orig_layer, LoRASidecarModule) and isinstance(orig_layer.orig_module, torch.nn.Linear)
-        ):
-            if isinstance(lora_layer, LoRALayer):
-                return LoRALinearSidecarLayer(lora_layer=lora_layer, weight=patch_weight)
-            elif isinstance(lora_layer, ConcatenatedLoRALayer):
-                return ConcatenatedLoRALinearSidecarLayer(concatenated_lora_layer=lora_layer, weight=patch_weight)
-            else:
-                raise ValueError(f"Unsupported Linear LoRA layer type: {type(lora_layer)}")
+    def _initialize_lora_wrapper_layer(orig_layer: torch.nn.Module):
+        if isinstance(orig_layer, torch.nn.Linear):
+            return LoRALinearWrapper(orig_layer, [], [])
+        elif isinstance(orig_layer, torch.nn.Conv1d):
+            return LoRAConv1dWrapper(orig_layer, [], [])
+        elif isinstance(orig_layer, torch.nn.Conv2d):
+            return LoRAConv2dWrapper(orig_layer, [], [])
         else:
             raise ValueError(f"Unsupported layer type: {type(orig_layer)}")
 

invokeai/backend/lora/sidecar_layers/concatenated_lora/concatenated_lora_linear_sidecar_layer.py

@@ -1,34 +0,0 @@
-import torch
-
-from invokeai.backend.lora.layers.concatenated_lora_layer import ConcatenatedLoRALayer
-
-
-class ConcatenatedLoRALinearSidecarLayer(torch.nn.Module):
-    def __init__(
-        self,
-        concatenated_lora_layer: ConcatenatedLoRALayer,
-        weight: float,
-    ):
-        super().__init__()
-
-        self._concatenated_lora_layer = concatenated_lora_layer
-        self._weight = weight
-
-    def forward(self, input: torch.Tensor) -> torch.Tensor:
-        x_chunks: list[torch.Tensor] = []
-        for lora_layer in self._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 *= self._weight * lora_layer.scale()
-            x_chunks.append(x_chunk)
-
-        # TODO(ryand): Generalize to support concat_axis != 0.
-        assert self._concatenated_lora_layer.concat_axis == 0
-        x = torch.cat(x_chunks, dim=-1)
-        return x
-
-    def to(self, device: torch.device | None = None, dtype: torch.dtype | None = None):
-        self._concatenated_lora_layer.to(device=device, dtype=dtype)
-        return self

invokeai/backend/lora/sidecar_layers/lora/lora_linear_sidecar_layer.py

@@ -1,27 +0,0 @@
-import torch
-
-from invokeai.backend.lora.layers.lora_layer import LoRALayer
-
-
-class LoRALinearSidecarLayer(torch.nn.Module):
-    def __init__(
-        self,
-        lora_layer: LoRALayer,
-        weight: float,
-    ):
-        super().__init__()
-
-        self._lora_layer = lora_layer
-        self._weight = weight
-
-    def forward(self, x: torch.Tensor) -> torch.Tensor:
-        x = torch.nn.functional.linear(x, self._lora_layer.down)
-        if self._lora_layer.mid is not None:
-            x = torch.nn.functional.linear(x, self._lora_layer.mid)
-        x = torch.nn.functional.linear(x, self._lora_layer.up, bias=self._lora_layer.bias)
-        x *= self._weight * self._lora_layer.scale()
-        return x
-
-    def to(self, device: torch.device | None = None, dtype: torch.dtype | None = None):
-        self._lora_layer.to(device=device, dtype=dtype)
-        return self

invokeai/backend/lora/sidecar_layers/lora_sidecar_module.py

@@ -1,24 +0,0 @@
-import torch
-
-
-class LoRASidecarModule(torch.nn.Module):
-    """A LoRA sidecar module that wraps an original module and adds LoRA layers to it."""
-
-    def __init__(self, orig_module: torch.nn.Module, lora_layers: list[torch.nn.Module]):
-        super().__init__()
-        self.orig_module = orig_module
-        self._lora_layers = lora_layers
-
-    def add_lora_layer(self, lora_layer: torch.nn.Module):
-        self._lora_layers.append(lora_layer)
-
-    def forward(self, input: torch.Tensor) -> torch.Tensor:
-        x = self.orig_module(input)
-        for lora_layer in self._lora_layers:
-            x += lora_layer(input)
-        return x
-
-    def to(self, device: torch.device | None = None, dtype: torch.dtype | None = None):
-        self._orig_module.to(device=device, dtype=dtype)
-        for lora_layer in self._lora_layers:
-            lora_layer.to(device=device, dtype=dtype)

invokeai/backend/model_manager/config.py

@@ -67,6 +67,7 @@ class ModelType(str, Enum):
     Main = "main"
     VAE = "vae"
     LoRA = "lora"
+    StructuralLoRa = "structural_lora"
     ControlNet = "controlnet"  # used by model_probe
     TextualInversion = "embedding"
     IPAdapter = "ip_adapter"
@@ -273,6 +274,18 @@ class LoRALyCORISConfig(LoRAConfigBase):
         return Tag(f"{ModelType.LoRA.value}.{ModelFormat.LyCORIS.value}")
 
 
+class StructuralLoRALyCORISConfig(ModelConfigBase):
+    """Model config for Structural LoRA/Lycoris models."""
+
+    type: Literal[ModelType.StructuralLoRa] = ModelType.StructuralLoRa
+    trigger_phrases: Optional[set[str]] = Field(description="Set of trigger phrases for this model", default=None)
+    format: Literal[ModelFormat.LyCORIS] = ModelFormat.LyCORIS
+
+    @staticmethod
+    def get_tag() -> Tag:
+        return Tag(f"{ModelType.StructuralLoRa.value}.{ModelFormat.LyCORIS.value}")
+
+
 class LoRADiffusersConfig(LoRAConfigBase):
     """Model config for LoRA/Diffusers models."""
 
@@ -535,6 +548,7 @@ AnyModelConfig = Annotated[
         Annotated[ControlNetDiffusersConfig, ControlNetDiffusersConfig.get_tag()],
         Annotated[ControlNetCheckpointConfig, ControlNetCheckpointConfig.get_tag()],
         Annotated[LoRALyCORISConfig, LoRALyCORISConfig.get_tag()],
+        Annotated[StructuralLoRALyCORISConfig, StructuralLoRALyCORISConfig.get_tag()],
         Annotated[LoRADiffusersConfig, LoRADiffusersConfig.get_tag()],
         Annotated[T5EncoderConfig, T5EncoderConfig.get_tag()],
         Annotated[T5EncoderBnbQuantizedLlmInt8bConfig, T5EncoderBnbQuantizedLlmInt8bConfig.get_tag()],

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

@@ -13,8 +13,9 @@ from invokeai.backend.lora.conversions.flux_diffusers_lora_conversion_utils impo
     lora_model_from_flux_diffusers_state_dict,
 )
 from invokeai.backend.lora.conversions.flux_kohya_lora_conversion_utils import (
-    lora_model_from_flux_kohya_state_dict,
+    is_state_dict_likely_in_flux_kohya_format, lora_model_from_flux_kohya_state_dict,
 )
+from invokeai.backend.lora.conversions.flux_control_lora_utils import is_state_dict_likely_flux_control, lora_model_from_flux_control_state_dict
 from invokeai.backend.lora.conversions.sd_lora_conversion_utils import lora_model_from_sd_state_dict
 from invokeai.backend.lora.conversions.sdxl_lora_conversion_utils import convert_sdxl_keys_to_diffusers_format
 from invokeai.backend.model_manager import (
@@ -32,6 +33,7 @@ from invokeai.backend.model_manager.load.model_loader_registry import ModelLoade
 
 @ModelLoaderRegistry.register(base=BaseModelType.Any, type=ModelType.LoRA, format=ModelFormat.Diffusers)
 @ModelLoaderRegistry.register(base=BaseModelType.Any, type=ModelType.LoRA, format=ModelFormat.LyCORIS)
+@ModelLoaderRegistry.register(base=BaseModelType.Flux, type=ModelType.StructuralLoRa, format=ModelFormat.LyCORIS)
 class LoRALoader(ModelLoader):
     """Class to load LoRA models."""
 
@@ -75,7 +77,10 @@ class LoRALoader(ModelLoader):
                 # https://github.com/huggingface/diffusers/blob/main/examples/dreambooth/train_dreambooth_lora_flux.py#L1194
                 model = lora_model_from_flux_diffusers_state_dict(state_dict=state_dict, alpha=None)
             elif config.format == ModelFormat.LyCORIS:
-                model = lora_model_from_flux_kohya_state_dict(state_dict=state_dict)
+                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_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}")
         elif self._model_base in [BaseModelType.StableDiffusion1, BaseModelType.StableDiffusion2]:

invokeai/backend/model_manager/probe.py

@@ -18,6 +18,7 @@ from invokeai.backend.flux.ip_adapter.state_dict_utils import is_state_dict_xlab
 from invokeai.backend.lora.conversions.flux_diffusers_lora_conversion_utils import (
     is_state_dict_likely_in_flux_diffusers_format,
 )
+from invokeai.backend.lora.conversions.flux_control_lora_utils import is_state_dict_likely_flux_control
 from invokeai.backend.lora.conversions.flux_kohya_lora_conversion_utils import is_state_dict_likely_in_flux_kohya_format
 from invokeai.backend.model_hash.model_hash import HASHING_ALGORITHMS, ModelHash
 from invokeai.backend.model_manager.config import (
@@ -258,6 +259,18 @@ class ModelProbe(object):
         ckpt = checkpoint if checkpoint else read_checkpoint_meta(model_path, scan=True)
         ckpt = ckpt.get("state_dict", ckpt)
 
+        if isinstance(ckpt, dict) and "img_in.lora_A.weight" in ckpt and "img_in.lora_B.weight" in ckpt:
+            tensor_a, tensor_b = ckpt["img_in.lora_A.weight"], ckpt["img_in.lora_B.weight"]
+            if (
+                tensor_a is not None
+                and isinstance(tensor_a, torch.Tensor)
+                and tensor_a.shape[1] == 128
+                and tensor_b is not None
+                and isinstance(tensor_b, torch.Tensor)
+                and tensor_b.shape[0] == 3072
+            ):
+                return ModelType.StructuralLoRa
+
         for key in [str(k) for k in ckpt.keys()]:
             if key.startswith(
                 (
@@ -624,8 +637,10 @@ class LoRACheckpointProbe(CheckpointProbeBase):
         return ModelFormat.LyCORIS
 
     def get_base_type(self) -> BaseModelType:
-        if is_state_dict_likely_in_flux_kohya_format(self.checkpoint) or is_state_dict_likely_in_flux_diffusers_format(
-            self.checkpoint
+        if (
+            is_state_dict_likely_in_flux_kohya_format(self.checkpoint)
+            or is_state_dict_likely_in_flux_diffusers_format(self.checkpoint)
+            or is_state_dict_likely_flux_control(self.checkpoint)
         ):
             return BaseModelType.Flux
 
@@ -1046,6 +1061,7 @@ ModelProbe.register_probe("diffusers", ModelType.SpandrelImageToImage, SpandrelI
 ModelProbe.register_probe("checkpoint", ModelType.Main, PipelineCheckpointProbe)
 ModelProbe.register_probe("checkpoint", ModelType.VAE, VaeCheckpointProbe)
 ModelProbe.register_probe("checkpoint", ModelType.LoRA, LoRACheckpointProbe)
+ModelProbe.register_probe("checkpoint", ModelType.StructuralLoRa, LoRACheckpointProbe)
 ModelProbe.register_probe("checkpoint", ModelType.TextualInversion, TextualInversionCheckpointProbe)
 ModelProbe.register_probe("checkpoint", ModelType.ControlNet, ControlNetCheckpointProbe)
 ModelProbe.register_probe("checkpoint", ModelType.IPAdapter, IPAdapterCheckpointProbe)

invokeai/backend/model_manager/util/model_util.py

@@ -52,16 +52,15 @@ def read_checkpoint_meta(path: Union[str, Path], scan: bool = True) -> Dict[str,
         except Exception:
             # TODO: create issue for support "meta"?
             checkpoint = safetensors.torch.load_file(path, device="cpu")
+    elif str(path).endswith(".gguf"):
+        # The GGUF reader used here uses numpy memmap, so these tensors are not loaded into memory during this function
+        checkpoint = gguf_sd_loader(Path(path), compute_dtype=torch.float32)
     else:
         if scan:
             scan_result = scan_file_path(path)
             if scan_result.infected_files != 0 or scan_result.scan_err:
                 raise Exception(f'The model file "{path}" is potentially infected by malware. Aborting import.')
-        if str(path).endswith(".gguf"):
-            # The GGUF reader used here uses numpy memmap, so these tensors are not loaded into memory during this function
-            checkpoint = gguf_sd_loader(Path(path), compute_dtype=torch.float32)
-        else:
-            checkpoint = torch.load(path, map_location=torch.device("meta"))
+        checkpoint = torch.load(path, map_location=torch.device("meta"))
     return checkpoint
 
 

invokeai/frontend/web/README.md

@@ -1,3 +1,3 @@
 # Invoke UI
 
-<https://invoke-ai.github.io/InvokeAI/contributing/frontend/OVERVIEW/>
+<https://invoke-ai.github.io/InvokeAI/contributing/frontend/>

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

@@ -96,7 +96,9 @@
         "new": "Neu",
         "ok": "OK",
         "close": "Schließen",
-        "clipboard": "Zwischenablage"
+        "clipboard": "Zwischenablage",
+        "generating": "Generieren",
+        "loadingModel": "Lade Modell"
     },
     "gallery": {
         "galleryImageSize": "Bildgröße",
@@ -591,7 +593,15 @@
         "loraTriggerPhrases": "LoRA-Auslösephrasen",
         "installingBundle": "Bündel wird installiert",
         "triggerPhrases": "Auslösephrasen",
-        "mainModelTriggerPhrases": "Hauptmodell-Auslösephrasen"
+        "mainModelTriggerPhrases": "Hauptmodell-Auslösephrasen",
+        "noDefaultSettings": "Für dieses Modell sind keine Standardeinstellungen konfiguriert. Besuchen Sie den Modell-Manager, um Standardeinstellungen hinzuzufügen.",
+        "defaultSettingsOutOfSync": "Einige Einstellungen stimmen nicht mit den Standardeinstellungen des Modells überein:",
+        "clipLEmbed": "CLIP-L einbetten",
+        "clipGEmbed": "CLIP-G einbetten",
+        "hfTokenLabel": "HuggingFace-Token (für einige Modelle erforderlich)",
+        "hfTokenHelperText": "Für die Nutzung einiger Modelle ist ein HF-Token erforderlich. Klicken Sie hier, um Ihr Token zu erstellen oder zu erhalten.",
+        "hfForbidden": "Sie haben keinen Zugriff auf dieses HF-Modell",
+        "hfTokenInvalid": "Ungültiges oder fehlendes HF-Token"
     },
     "parameters": {
         "images": "Bilder",
@@ -632,12 +642,6 @@
         "remixImage": "Remix des Bilds erstellen",
         "imageActions": "Weitere Bildaktionen",
         "invoke": {
-            "layer": {
-                "t2iAdapterIncompatibleBboxWidth": "$t(parameters.invoke.layer.t2iAdapterRequiresDimensionsToBeMultipleOf) {{multiple}}, Bbox-Breite ist {{width}}",
-                "t2iAdapterIncompatibleScaledBboxWidth": "$t(parameters.invoke.layer.t2iAdapterRequiresDimensionsToBeMultipleOf) {{multiple}}, Skalierte Bbox-Breite ist {{width}}",
-                "t2iAdapterIncompatibleScaledBboxHeight": "$t(parameters.invoke.layer.t2iAdapterRequiresDimensionsToBeMultipleOf) {{multiple}}, Skalierte Bbox-Höhe ist {{height}}",
-                "t2iAdapterIncompatibleBboxHeight": "$t(parameters.invoke.layer.t2iAdapterRequiresDimensionsToBeMultipleOf) {{multiple}}, Bbox-Höhe ist {{height}}"
-            },
             "fluxModelIncompatibleScaledBboxWidth": "$t(parameters.invoke.fluxRequiresDimensionsToBeMultipleOf16), Skalierte Bbox-Breite ist {{width}}",
             "fluxModelIncompatibleScaledBboxHeight": "$t(parameters.invoke.fluxRequiresDimensionsToBeMultipleOf16), Skalierte Bbox-Höhe ist {{height}}",
             "fluxModelIncompatibleBboxWidth": "$t(parameters.invoke.fluxRequiresDimensionsToBeMultipleOf16), Bbox-Breite ist {{width}}",
@@ -841,7 +845,8 @@
         "upscaling": "Hochskalierung",
         "canvas": "Leinwand",
         "prompts_one": "Prompt",
-        "prompts_other": "Prompts"
+        "prompts_other": "Prompts",
+        "batchSize": "Stapelgröße"
     },
     "metadata": {
         "negativePrompt": "Negativ Beschreibung",
@@ -1081,6 +1086,21 @@
         },
         "patchmatchDownScaleSize": {
             "heading": "Herunterskalieren"
+        },
+        "paramHeight": {
+            "heading": "Höhe",
+            "paragraphs": [
+                "Höhe des generierten Bildes. Muss ein Vielfaches von 8 sein."
+            ]
+        },
+        "paramUpscaleMethod": {
+            "heading": "Vergrößerungsmethode",
+            "paragraphs": [
+                "Methode zum Hochskalieren des Bildes für High Resolution Fix."
+            ]
+        },
+        "paramHrf": {
+            "heading": "High Resolution Fix aktivieren"
         }
     },
     "invocationCache": {

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

@@ -176,7 +176,8 @@
         "reset": "Reset",
         "none": "None",
         "new": "New",
-        "generating": "Generating"
+        "generating": "Generating",
+        "warnings": "Warnings"
     },
     "hrf": {
         "hrf": "High Resolution Fix",
@@ -808,6 +809,7 @@
         "starterBundleHelpText": "Easily install all models needed to get started with a base model, including a main model, controlnets, IP adapters, and more. Selecting a bundle will skip any models that you already have installed.",
         "starterModels": "Starter Models",
         "starterModelsInModelManager": "Starter Models can be found in Model Manager",
+        "structuralLora": "Structural LoRA",
         "syncModels": "Sync Models",
         "textualInversions": "Textual Inversions",
         "triggerPhrases": "Trigger Phrases",
@@ -1038,20 +1040,7 @@
             "canvasIsSelectingObject": "Canvas is busy (selecting object)",
             "noPrompts": "No prompts generated",
             "noNodesInGraph": "No nodes in graph",
-            "systemDisconnected": "System disconnected",
-            "layer": {
-                "controlAdapterNoModelSelected": "no Control Adapter model selected",
-                "controlAdapterIncompatibleBaseModel": "incompatible Control Adapter base model",
-                "t2iAdapterIncompatibleBboxWidth": "$t(parameters.invoke.layer.t2iAdapterRequiresDimensionsToBeMultipleOf) {{multiple}}, bbox width is {{width}}",
-                "t2iAdapterIncompatibleBboxHeight": "$t(parameters.invoke.layer.t2iAdapterRequiresDimensionsToBeMultipleOf) {{multiple}}, bbox height is {{height}}",
-                "t2iAdapterIncompatibleScaledBboxWidth": "$t(parameters.invoke.layer.t2iAdapterRequiresDimensionsToBeMultipleOf) {{multiple}}, scaled bbox width is {{width}}",
-                "t2iAdapterIncompatibleScaledBboxHeight": "$t(parameters.invoke.layer.t2iAdapterRequiresDimensionsToBeMultipleOf) {{multiple}}, scaled bbox height is {{height}}",
-                "ipAdapterNoModelSelected": "no IP adapter selected",
-                "ipAdapterIncompatibleBaseModel": "incompatible IP Adapter base model",
-                "ipAdapterNoImageSelected": "no IP Adapter image selected",
-                "rgNoPromptsOrIPAdapters": "no text prompts or IP Adapters",
-                "rgNoRegion": "no region selected"
-            }
+            "systemDisconnected": "System disconnected"
         },
         "maskBlur": "Mask Blur",
         "negativePromptPlaceholder": "Negative Prompt",
@@ -1713,6 +1702,8 @@
         "controlLayer": "Control Layer",
         "inpaintMask": "Inpaint Mask",
         "regionalGuidance": "Regional Guidance",
+        "referenceImageRegional": "Reference Image (Regional)",
+        "referenceImageGlobal": "Reference Image (Global)",
         "asRasterLayer": "As $t(controlLayers.rasterLayer)",
         "asRasterLayerResize": "As $t(controlLayers.rasterLayer) (Resize)",
         "asControlLayer": "As $t(controlLayers.controlLayer)",
@@ -1798,6 +1789,21 @@
         "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.",
+        "warnings": {
+            "problemsFound": "Problems found",
+            "unsupportedModel": "layer not supported for selected base model",
+            "controlAdapterNoModelSelected": "no Control Layer model selected",
+            "controlAdapterIncompatibleBaseModel": "incompatible Control Layer base model",
+            "controlAdapterNoControl": "no control selected/drawn",
+            "ipAdapterNoModelSelected": "no Reference Image model selected",
+            "ipAdapterIncompatibleBaseModel": "incompatible Reference Image base model",
+            "ipAdapterNoImageSelected": "no Reference Image image selected",
+            "rgNoPromptsOrIPAdapters": "no text prompts or Reference Images",
+            "rgNegativePromptNotSupported": "Negative Prompt not supported for selected base model",
+            "rgReferenceImagesNotSupported": "regional Reference Images not supported for selected base model",
+            "rgAutoNegativeNotSupported": "Auto-Negative not supported for selected base model",
+            "rgNoRegion": "no region drawn"
+        },
         "controlMode": {
             "controlMode": "Control Mode",
             "balanced": "Balanced (recommended)",
@@ -2128,8 +2134,8 @@
     "whatsNew": {
         "whatsNewInInvoke": "What's New in Invoke",
         "items": [
-            "<StrongComponent>Workflows</StrongComponent>: Run a workflow for a collection of images using the new <StrongComponent>Image Batch</StrongComponent> node.",
-            "<StrongComponent>FLUX</StrongComponent>: Support for XLabs IP Adapter v2."
+            "<StrongComponent>FLUX Regional Guidance (beta)</StrongComponent>: Our beta release of FLUX Regional Guidance is live for regional prompt control.",
+            "<StrongComponent>Various UX Improvements</StrongComponent>: A number of small UX and Quality of Life improvements throughout the app."
         ],
         "readReleaseNotes": "Read Release Notes",
         "watchRecentReleaseVideos": "Watch Recent Release Videos",

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

@@ -317,19 +317,6 @@
         "info": "Info",
         "showOptionsPanel": "Afficher le panneau latéral (O ou T)",
         "invoke": {
-            "layer": {
-                "rgNoPromptsOrIPAdapters": "aucun prompts ou IP Adapters",
-                "t2iAdapterIncompatibleScaledBboxWidth": "$t(parameters.invoke.layer.t2iAdapterRequiresDimensionsToBeMultipleOf) {{multiple}}, la largeur de la bounding box mise à l'échelle est {{width}}",
-                "t2iAdapterIncompatibleScaledBboxHeight": "$t(parameters.invoke.layer.t2iAdapterRequiresDimensionsToBeMultipleOf) {{multiple}}, la hauteur de la bounding box mise à l'échelle est {{height}}",
-                "ipAdapterNoModelSelected": "aucun IP adapter sélectionné",
-                "ipAdapterNoImageSelected": "aucune image d'IP adapter sélectionnée",
-                "controlAdapterIncompatibleBaseModel": "modèle de base de Control Adapter incompatible",
-                "t2iAdapterIncompatibleBboxHeight": "$t(parameters.invoke.layer.t2iAdapterRequiresDimensionsToBeMultipleOf) {{multiple}}, la hauteur de la bounding box est {{height}}",
-                "t2iAdapterIncompatibleBboxWidth": "$t(parameters.invoke.layer.t2iAdapterRequiresDimensionsToBeMultipleOf) {{multiple}}, la largeur de la bounding box est {{width}}",
-                "ipAdapterIncompatibleBaseModel": "modèle de base d'IP adapter incompatible",
-                "rgNoRegion": "aucune zone sélectionnée",
-                "controlAdapterNoModelSelected": "aucun modèle de Control Adapter sélectionné"
-            },
             "noPrompts": "Aucun prompts généré",
             "missingInputForField": "{{nodeLabel}} -> {{fieldLabel}} entrée manquante",
             "missingFieldTemplate": "Modèle de champ manquant",

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

@@ -96,7 +96,8 @@
         "clipboard": "Appunti",
         "ok": "Ok",
         "generating": "Generazione",
-        "loadingModel": "Caricamento del modello"
+        "loadingModel": "Caricamento del modello",
+        "warnings": "Avvisi"
     },
     "gallery": {
         "galleryImageSize": "Dimensione dell'immagine",
@@ -662,21 +663,8 @@
             "addingImagesTo": "Aggiungi immagini a",
             "systemDisconnected": "Sistema disconnesso",
             "missingNodeTemplate": "Modello di nodo mancante",
-            "missingInputForField": "{{nodeLabel}} -> {{fieldLabel}} ingresso mancante",
+            "missingInputForField": "{{nodeLabel}} -> {{fieldLabel}}: ingresso mancante",
             "missingFieldTemplate": "Modello di campo mancante",
-            "layer": {
-                "controlAdapterNoModelSelected": "Nessun modello di adattatore di controllo selezionato",
-                "controlAdapterIncompatibleBaseModel": "Il modello base dell'adattatore di controllo non è compatibile",
-                "ipAdapterNoModelSelected": "Nessun adattatore IP selezionato",
-                "ipAdapterIncompatibleBaseModel": "Il modello base dell'adattatore IP non è compatibile",
-                "ipAdapterNoImageSelected": "Nessuna immagine dell'adattatore IP selezionata",
-                "rgNoPromptsOrIPAdapters": "Nessun prompt o adattatore IP",
-                "rgNoRegion": "Nessuna regione selezionata",
-                "t2iAdapterIncompatibleBboxWidth": "$t(parameters.invoke.layer.t2iAdapterRequiresDimensionsToBeMultipleOf) {{multiple}}, larghezza riquadro è {{width}}",
-                "t2iAdapterIncompatibleBboxHeight": "$t(parameters.invoke.layer.t2iAdapterRequiresDimensionsToBeMultipleOf) {{multiple}}, altezza riquadro è {{height}}",
-                "t2iAdapterIncompatibleScaledBboxWidth": "$t(parameters.invoke.layer.t2iAdapterRequiresDimensionsToBeMultipleOf) {{multiple}}, larghezza del riquadro scalato {{width}}",
-                "t2iAdapterIncompatibleScaledBboxHeight": "$t(parameters.invoke.layer.t2iAdapterRequiresDimensionsToBeMultipleOf) {{multiple}}, altezza del riquadro scalato {{height}}"
-            },
             "fluxModelIncompatibleBboxHeight": "$t(parameters.invoke.fluxRequiresDimensionsToBeMultipleOf16), altezza riquadro è {{height}}",
             "fluxModelIncompatibleBboxWidth": "$t(parameters.invoke.fluxRequiresDimensionsToBeMultipleOf16), larghezza riquadro è {{width}}",
             "fluxModelIncompatibleScaledBboxWidth": "$t(parameters.invoke.fluxRequiresDimensionsToBeMultipleOf16), larghezza del riquadro scalato è {{width}}",
@@ -684,10 +672,14 @@
             "noT5EncoderModelSelected": "Nessun modello di encoder T5 selezionato per la generazione con FLUX",
             "noCLIPEmbedModelSelected": "Nessun modello CLIP Embed selezionato per la generazione con FLUX",
             "noFLUXVAEModelSelected": "Nessun modello VAE selezionato per la generazione con FLUX",
-            "canvasIsTransforming": "La tela sta trasformando",
-            "canvasIsRasterizing": "La tela sta rasterizzando",
-            "canvasIsCompositing": "La tela è in fase di composizione",
-            "canvasIsFiltering": "La tela sta filtrando"
+            "canvasIsTransforming": "La tela è occupata (sta trasformando)",
+            "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}}",
+            "canvasIsSelectingObject": "La tela è occupata (selezione dell'oggetto)",
+            "collectionTooFewItems": "{{nodeLabel}} -> {{fieldLabel}}: troppi pochi elementi, minimo {{minItems}}",
+            "collectionEmpty": "{{nodeLabel}} -> {{fieldLabel}} raccolta vuota"
         },
         "useCpuNoise": "Usa la CPU per generare rumore",
         "iterations": "Iterazioni",
@@ -972,7 +964,9 @@
         "saveToGallery": "Salva nella Galleria",
         "noMatchingWorkflows": "Nessun flusso di lavoro corrispondente",
         "noWorkflows": "Nessun flusso di lavoro",
-        "workflowHelpText": "Hai bisogno di aiuto? Consulta la nostra guida <LinkComponent>Introduzione ai flussi di lavoro</LinkComponent>."
+        "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."
     },
     "boards": {
         "autoAddBoard": "Aggiungi automaticamente bacheca",
@@ -1093,7 +1087,8 @@
         "workflows": "Flussi di lavoro",
         "generation": "Generazione",
         "other": "Altro",
-        "gallery": "Galleria"
+        "gallery": "Galleria",
+        "batchSize": "Dimensione del lotto"
     },
     "models": {
         "noMatchingModels": "Nessun modello corrispondente",
@@ -1195,8 +1190,9 @@
         "controlNetBeginEnd": {
             "heading": "Percentuale passi Inizio / Fine",
             "paragraphs": [
-                "La parte del processo di rimozione del rumore in cui verrà applicato l'adattatore di controllo.",
-                "In genere, gli adattatori di controllo applicati all'inizio del processo guidano la composizione, mentre quelli applicati alla fine guidano i dettagli."
+                "Questa impostazione determina quale parte del processo di rimozione del rumore (generazione) incorpora la guida da questo livello.",
+                "• Passo iniziale (%): specifica quando iniziare ad applicare la guida da questo livello durante il processo di generazione.",
+                "• Passo finale (%): specifica quando interrompere l'applicazione della guida di questo livello e ripristinare la guida generale dal modello e altre impostazioni."
             ]
         },
         "noiseUseCPU": {
@@ -1300,7 +1296,9 @@
         "controlNetWeight": {
             "heading": "Peso",
             "paragraphs": [
-                "Peso dell'adattatore di controllo. Un peso maggiore porterà a impatti maggiori sull'immagine finale."
+                "Regola la forza con cui il livello influenza il processo di generazione",
+                "• Peso maggiore (0.75-2): crea un impatto più significativo sul risultato finale.",
+                "• Peso inferiore (0-0.75): crea un impatto minore sul risultato finale."
             ]
         },
         "paramCFGScale": {
@@ -1477,9 +1475,9 @@
             ]
         },
         "ipAdapterMethod": {
-            "heading": "Metodo",
+            "heading": "Modalità",
             "paragraphs": [
-                "Metodo con cui applicare l'adattatore IP corrente."
+                "La modalità definisce il modo in cui l'immagine di riferimento guiderà il processo di generazione."
             ]
         },
         "scale": {
@@ -1801,7 +1799,10 @@
             "full": "Stile e Composizione",
             "style": "Solo Stile",
             "composition": "Solo Composizione",
-            "ipAdapterMethod": "Metodo Adattatore IP"
+            "ipAdapterMethod": "Modalità",
+            "fullDesc": "Applica lo stile visivo (colori, texture) e la composizione (disposizione, struttura).",
+            "styleDesc": "Applica lo stile visivo (colori, texture) senza considerare la disposizione.",
+            "compositionDesc": "Replica disposizione e struttura ignorando lo stile di riferimento."
         },
         "showingType": "Mostra {{type}}",
         "dynamicGrid": "Griglia dinamica",
@@ -2044,7 +2045,33 @@
         "replaceCurrent": "Sostituisci corrente",
         "mergeDown": "Unire in basso",
         "mergingLayers": "Unione dei livelli",
-        "controlLayerEmptyState": "<UploadButton>Carica un'immagine</UploadButton>, trascina un'immagine dalla <GalleryButton>galleria</GalleryButton> su questo livello oppure disegna sulla tela per iniziare."
+        "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.",
+        "asRasterLayer": "Come $t(controlLayers.rasterLayer)",
+        "asRasterLayerResize": "Come $t(controlLayers.rasterLayer) (Ridimensiona)",
+        "asControlLayer": "Come $t(controlLayers.controlLayer)",
+        "asControlLayerResize": "Come $t(controlLayers.controlLayer) (Ridimensiona)",
+        "newSession": "Nuova sessione",
+        "resetCanvasLayers": "Ripristina livelli Tela",
+        "referenceImageRegional": "Immagine di riferimento (regionale)",
+        "referenceImageGlobal": "Immagine di riferimento (globale)",
+        "warnings": {
+            "controlAdapterNoModelSelected": "nessun modello selezionato per il livello di controllo",
+            "controlAdapterNoControl": "nessun controllo selezionato/disegnato",
+            "ipAdapterNoModelSelected": "nessun modello di immagine di riferimento selezionato",
+            "rgNoPromptsOrIPAdapters": "nessun prompt testuale o immagini di riferimento",
+            "rgReferenceImagesNotSupported": "Immagini di riferimento regionali non supportate per il modello base selezionato",
+            "rgNoRegion": "nessuna regione disegnata",
+            "problemsFound": "Problemi riscontrati",
+            "unsupportedModel": "livello non supportato per il modello base selezionato",
+            "controlAdapterIncompatibleBaseModel": "modello di base del livello di controllo incompatibile",
+            "rgNegativePromptNotSupported": "Prompt negativo non supportato per il modello base selezionato",
+            "ipAdapterIncompatibleBaseModel": "modello base dell'immagine di riferimento incompatibile",
+            "ipAdapterNoImageSelected": "nessuna immagine di riferimento selezionata",
+            "rgAutoNegativeNotSupported": "Auto-Negativo non supportato per il modello base selezionato"
+        }
     },
     "ui": {
         "tabs": {
@@ -2144,8 +2171,8 @@
         "watchRecentReleaseVideos": "Guarda i video su questa versione",
         "watchUiUpdatesOverview": "Guarda le novità dell'interfaccia",
         "items": [
-            "<StrongComponent>SD 3.5</StrongComponent>: supporto per SD 3.5 Medium e Large.",
-            "<StrongComponent>Tela</StrongComponent>: elaborazione semplificata del livello di controllo e impostazioni di controllo predefinite migliorate."
+            "<StrongComponent>FLUX Regional Guidance (beta)</StrongComponent>: la nostra versione beta di FLUX Regional Guidance è attiva per il controllo dei prompt regionali.",
+            "<StrongComponent>Vari miglioramenti dell'esperienza utente</StrongComponent>: numerosi piccoli miglioramenti dell'esperienza utente e della qualità della vita in tutta l'app."
         ]
     },
     "system": {
@@ -2172,5 +2199,67 @@
             "logNamespaces": "Elementi del registro"
         },
         "enableLogging": "Abilita la registrazione"
+    },
+    "supportVideos": {
+        "gettingStarted": "Iniziare",
+        "supportVideos": "Video di supporto",
+        "videos": {
+            "usingControlLayersAndReferenceGuides": {
+                "title": "Utilizzo di livelli di controllo e guide di riferimento",
+                "description": "Scopri come guidare la creazione delle tue immagini con livelli di controllo e immagini di riferimento."
+            },
+            "creatingYourFirstImage": {
+                "description": "Introduzione alla creazione di un'immagine da zero utilizzando gli strumenti di Invoke.",
+                "title": "Creazione della tua prima immagine"
+            },
+            "understandingImageToImageAndDenoising": {
+                "description": "Panoramica delle trasformazioni immagine-a-immagine e della riduzione del rumore in Invoke.",
+                "title": "Comprendere immagine-a-immagine e riduzione del rumore"
+            },
+            "howDoIDoImageToImageTransformation": {
+                "description": "Tutorial su come eseguire trasformazioni da immagine a immagine in Invoke.",
+                "title": "Come si esegue la trasformazione da immagine-a-immagine?"
+            },
+            "howDoIUseInpaintMasks": {
+                "title": "Come si usano le maschere Inpaint?",
+                "description": "Come applicare maschere inpaint per la correzione e la variazione delle immagini."
+            },
+            "howDoIOutpaint": {
+                "description": "Guida all'outpainting oltre i confini dell'immagine originale.",
+                "title": "Come posso eseguire l'outpainting?"
+            },
+            "exploringAIModelsAndConceptAdapters": {
+                "description": "Approfondisci i modelli di intelligenza artificiale e scopri come utilizzare gli adattatori concettuali per il controllo creativo.",
+                "title": "Esplorazione dei modelli di IA e degli adattatori concettuali"
+            },
+            "upscaling": {
+                "title": "Ampliamento",
+                "description": "Come ampliare le immagini con gli strumenti di Invoke per migliorarne la risoluzione."
+            },
+            "creatingAndComposingOnInvokesControlCanvas": {
+                "description": "Impara a comporre immagini utilizzando la tela di controllo di Invoke.",
+                "title": "Creare e comporre sulla tela di controllo di Invoke"
+            },
+            "howDoIGenerateAndSaveToTheGallery": {
+                "description": "Passaggi per generare e salvare le immagini nella galleria.",
+                "title": "Come posso generare e salvare nella Galleria?"
+            },
+            "howDoIEditOnTheCanvas": {
+                "title": "Come posso apportare modifiche sulla tela?",
+                "description": "Guida alla modifica delle immagini direttamente sulla tela."
+            },
+            "howDoIUseControlNetsAndControlLayers": {
+                "title": "Come posso utilizzare le Reti di Controllo e i Livelli di Controllo?",
+                "description": "Impara ad applicare livelli di controllo e reti di controllo alle tue immagini."
+            },
+            "howDoIUseGlobalIPAdaptersAndReferenceImages": {
+                "title": "Come si utilizzano gli adattatori IP globali e le immagini di riferimento?",
+                "description": "Introduzione all'aggiunta di immagini di riferimento e adattatori IP globali."
+            }
+        },
+        "controlCanvas": "Tela di Controllo",
+        "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."
     }
 }

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

@@ -230,16 +230,7 @@
             "systemDisconnected": "Systeem is niet verbonden",
             "missingNodeTemplate": "Knooppuntsjabloon ontbreekt",
             "missingFieldTemplate": "Veldsjabloon ontbreekt",
-            "addingImagesTo": "Bezig met toevoegen van afbeeldingen aan",
-            "layer": {
-                "controlAdapterNoModelSelected": "geen controle-adaptermodel geselecteerd",
-                "controlAdapterIncompatibleBaseModel": "niet-compatibele basismodel voor controle-adapter",
-                "ipAdapterIncompatibleBaseModel": "niet-compatibele basismodel voor IP-adapter",
-                "ipAdapterNoImageSelected": "geen afbeelding voor IP-adapter geselecteerd",
-                "rgNoRegion": "geen gebied geselecteerd",
-                "rgNoPromptsOrIPAdapters": "geen tekstprompts of IP-adapters",
-                "ipAdapterNoModelSelected": "geen IP-adapter geselecteerd"
-            }
+            "addingImagesTo": "Bezig met toevoegen van afbeeldingen aan"
         },
         "patchmatchDownScaleSize": "Verklein",
         "useCpuNoise": "Gebruik CPU-ruis",

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

@@ -10,7 +10,24 @@
         "load": "Załaduj",
         "statusDisconnected": "Odłączono od serwera",
         "githubLabel": "GitHub",
-        "discordLabel": "Discord"
+        "discordLabel": "Discord",
+        "clipboard": "Schowek",
+        "aboutDesc": "Wykorzystujesz Invoke do pracy? Sprawdź:",
+        "ai": "SI",
+        "areYouSure": "Czy jesteś pewien?",
+        "copyError": "$t(gallery.copy) Błąd",
+        "apply": "Zastosuj",
+        "copy": "Kopiuj",
+        "or": "albo",
+        "add": "Dodaj",
+        "off": "Wyłączony",
+        "accept": "Zaakceptuj",
+        "cancel": "Anuluj",
+        "advanced": "Zawansowane",
+        "back": "Do tyłu",
+        "auto": "Automatyczny",
+        "beta": "Beta",
+        "close": "Wyjdź"
     },
     "gallery": {
         "galleryImageSize": "Rozmiar obrazów",
@@ -65,6 +82,42 @@
         "uploadImage": "Wgrywanie obrazu",
         "previousImage": "Poprzedni obraz",
         "nextImage": "Następny obraz",
-        "menu": "Menu"
+        "menu": "Menu",
+        "mode": "Tryb"
+    },
+    "boards": {
+        "cancel": "Anuluj",
+        "noBoards": "Brak tablic typu {{boardType}}",
+        "imagesWithCount_one": "{{count}} zdjęcie",
+        "imagesWithCount_few": "{{count}} zdjęcia",
+        "imagesWithCount_many": "{{count}} zdjęcia",
+        "private": "Prywatne tablice",
+        "updateBoardError": "Błąd aktualizacji tablicy",
+        "uncategorized": "Nieskategoryzowane",
+        "selectBoard": "Wybierz tablicę",
+        "downloadBoard": "Pobierz tablice",
+        "loading": "Ładowanie...",
+        "move": "Przenieś",
+        "noMatching": "Brak pasujących tablic"
+    },
+    "accordions": {
+        "compositing": {
+            "title": "Kompozycja",
+            "infillTab": "Inskrypcja",
+            "coherenceTab": "Przebieg Koherencji"
+        },
+        "generation": {
+            "title": "Generowanie"
+        },
+        "image": {
+            "title": "Zdjęcie"
+        },
+        "advanced": {
+            "options": "$t(accordions.advanced.title) Opcje",
+            "title": "Zaawansowane"
+        },
+        "control": {
+            "title": "Kontrola"
+        }
     }
 }

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

@@ -648,19 +648,6 @@
             "missingFieldTemplate": "Отсутствует шаблон поля",
             "addingImagesTo": "Добавление изображений в",
             "invoke": "Создать",
-            "layer": {
-                "ipAdapterNoModelSelected": "IP адаптер не выбран",
-                "controlAdapterNoModelSelected": "не выбрана модель адаптера контроля",
-                "controlAdapterIncompatibleBaseModel": "несовместимая базовая модель адаптера контроля",
-                "rgNoRegion": "регион не выбран",
-                "rgNoPromptsOrIPAdapters": "нет текстовых запросов или IP-адаптеров",
-                "ipAdapterIncompatibleBaseModel": "несовместимая базовая модель IP-адаптера",
-                "ipAdapterNoImageSelected": "изображение IP-адаптера не выбрано",
-                "t2iAdapterIncompatibleScaledBboxWidth": "$t(parameters.invoke.layer.t2iAdapterRequiresDimensionsToBeMultipleOf) {{multiple}}, масштабированная ширина рамки {{width}}",
-                "t2iAdapterIncompatibleBboxHeight": "$t(parameters.invoke.layer.t2iAdapterRequiresDimensionsToBeMultipleOf) {{multiple}}, высота рамки {{height}}",
-                "t2iAdapterIncompatibleBboxWidth": "$t(parameters.invoke.layer.t2iAdapterRequiresDimensionsToBeMultipleOf) {{multiple}}, ширина рамки {{width}}",
-                "t2iAdapterIncompatibleScaledBboxHeight": "$t(parameters.invoke.layer.t2iAdapterRequiresDimensionsToBeMultipleOf) {{multiple}}, масштабированная высота рамки {{height}}"
-            },
             "fluxModelIncompatibleBboxWidth": "$t(parameters.invoke.fluxRequiresDimensionsToBeMultipleOf16), ширина рамки {{width}}",
             "fluxModelIncompatibleBboxHeight": "$t(parameters.invoke.fluxRequiresDimensionsToBeMultipleOf16), высота рамки {{height}}",
             "fluxModelIncompatibleScaledBboxHeight": "$t(parameters.invoke.fluxRequiresDimensionsToBeMultipleOf16), масштабированная высота рамки {{height}}",

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

@@ -217,7 +217,10 @@
         "direction": "Phương Hướng",
         "unknownError": "Lỗi Không Rõ",
         "selected": "Đã chọn",
-        "tab": "Tab"
+        "tab": "Tab",
+        "loadingModel": "Đang Tải Model",
+        "generating": "Đang Tạo Sinh",
+        "warnings": "Cảnh Báo"
     },
     "prompt": {
         "addPromptTrigger": "Thêm Prompt Trigger",
@@ -290,7 +293,8 @@
         "cancelSucceeded": "Mục Đã Huỷ Bỏ",
         "completedIn": "Hoàn tất trong",
         "graphQueued": "Đồ Thị Đã Vào Hàng",
-        "batchQueuedDesc_other": "Thêm {{count}} phiên vào {{direction}} của hàng"
+        "batchQueuedDesc_other": "Thêm {{count}} phiên vào {{direction}} của hàng",
+        "batchSize": "Kích Thước Vùng Hàng Loạt"
     },
     "hotkeys": {
         "canvas": {
@@ -733,7 +737,9 @@
         "textualInversions": "Bộ Đảo Ngược Văn Bản",
         "loraTriggerPhrases": "Từ Ngữ Kích Hoạt Cho LoRA",
         "width": "Chiều Rộng",
-        "starterModelsInModelManager": "Model khởi đầu có thể tìm thấy ở Trình Quản Lý Model"
+        "starterModelsInModelManager": "Model khởi đầu có thể tìm thấy ở Trình Quản Lý Model",
+        "clipLEmbed": "CLIP-L Embed",
+        "clipGEmbed": "CLIP-G Embed"
     },
     "metadata": {
         "guidance": "Hướng Dẫn",
@@ -905,7 +911,7 @@
         "unknownNode": "Node Không Rõ",
         "unknownNodeType": "Loại Node Không Rõ",
         "unknownTemplate": "Mẫu Trình Bày Không Rõ",
-        "cannotConnectOutputToOutput": "Không thế kết nối đầu ra với đầu vào",
+        "cannotConnectOutputToOutput": "Không thế kết nối đầu ra với đầu ra",
         "cannotConnectToSelf": "Không thể kết nối với chính nó",
         "workflow": "Workflow",
         "addNodeToolTip": "Thêm Node (Shift+A, Space)",
@@ -952,7 +958,9 @@
         "executionStateInProgress": "Đang Xử Lý",
         "showLegendNodes": "Hiển Thị Vùng Nhập",
         "outputFieldTypeParseError": "Không thể phân tích loại dữ liệu đầu ra của {{node}}.{{field}} ({{message}})",
-        "modelAccessError": "Không thể tìm thấy model {{key}}, chuyển về mặc định"
+        "modelAccessError": "Không thể tìm thấy model {{key}}, chuyển về mặc định",
+        "internalDesc": "Trình kích hoạt này được dùng bên trong bởi Invoke. Nó có thể phá hỏng thay đổi trong khi cập nhật ứng dụng và có thể bị xoá bất cứ lúc nào.",
+        "specialDesc": "Trình kích hoạt này có một số xử lý đặc biệt trong ứng dụng. Ví dụ, Node Hàng Loạt được dùng để xếp vào nhiều đồ thị từ một workflow."
     },
     "popovers": {
         "paramCFGRescaleMultiplier": {
@@ -1105,7 +1113,9 @@
         },
         "controlNetWeight": {
             "paragraphs": [
-                "Trọng lượng của Control Adapter. Trọng lượng càng cao sẽ dẫn đến tác động càng lớn lên ảnh cuối cùng."
+                "Điều chỉnh mức độ layer ảnh hưởng đến quá trình xử lý tạo sinh.",
+                "• Trọng Lượng Lớn Hơn (.75-2): Gây ra ảnh hưởng lớn hơn lên kết quả cuối cùng.",
+                "• Trọng Lượng Nhỏ Hơn (0-.75): Gây ra ảnh hưởng nhỏ hơn lên kết quả cuối cùng."
             ],
             "heading": "Trọng Lượng"
         },
@@ -1149,7 +1159,7 @@
         },
         "ipAdapterMethod": {
             "paragraphs": [
-                "Cách thức dùng để áp dụng IP Adapter hiện tại."
+                "Phương thức định nghĩa cách ảnh mẫu sẽ chỉ dẫn quá trình xử lý tạo sinh."
             ],
             "heading": "Cách Thức"
         },
@@ -1196,8 +1206,9 @@
         },
         "controlNetBeginEnd": {
             "paragraphs": [
-                "Một phần trong quá trình xử lý khử nhiễu mà sẽ được Control Adapter áp dụng.",
-                "Nói chung, Control Adapter áp dụng vào lúc bắt đầu của quá trình hướng dẫn thành phần, và cũng áp dụng vào lúc kết thúc hướng dẫn chi tiết."
+                "Cài đặt này xác định phần xử lý khử nhiễu (trong khi tạo sinh) kết hợp với chỉ dẫn từ layer này.",
+                "• Bước Bắt Đầu (%): Chỉ định lúc bắt đầu áp dụng chỉ dẫn từ layer này trong quá trình tạo sinh.",
+                "• Bước Kết Thúc (%): Chỉ định lúc dừng áp dụng chỉ dẫn của layer này và trở về chỉ dẫn chung từ model và các thiết lập khác."
             ],
             "heading": "Phần Trăm Tham Số Bước Khi Bắt Đầu/Kết Thúc"
         },
@@ -1399,26 +1410,13 @@
         "processImage": "Xử Lý Hình Ảnh",
         "useSize": "Dùng Kích Thước",
         "invoke": {
-            "layer": {
-                "t2iAdapterIncompatibleScaledBboxHeight": "$t(parameters.invoke.layer.t2iAdapterRequiresDimensionsToBeMultipleOf) {{multiple}}, tỉ lệ chiều dài hộp giới hạn là {{height}}",
-                "rgNoRegion": "không có vùng được chọn",
-                "ipAdapterNoModelSelected": "không có IP Adapter được lựa chọn",
-                "ipAdapterNoImageSelected": "không có ảnh IP Adapter được lựa chọn",
-                "t2iAdapterIncompatibleBboxHeight": "$t(parameters.invoke.layer.t2iAdapterRequiresDimensionsToBeMultipleOf) {{multiple}}, chiều dài hộp giới hạn là {{height}}",
-                "t2iAdapterIncompatibleScaledBboxWidth": "$t(parameters.invoke.layer.t2iAdapterRequiresDimensionsToBeMultipleOf) {{multiple}}, tỉ lệ chiều rộng hộp giới hạn là {{width}}",
-                "t2iAdapterIncompatibleBboxWidth": "$t(parameters.invoke.layer.t2iAdapterRequiresDimensionsToBeMultipleOf) {{multiple}}, chiều rộng hộp giới hạn là {{width}}",
-                "rgNoPromptsOrIPAdapters": "không có lệnh chữ hoặc IP Adapter",
-                "controlAdapterIncompatibleBaseModel": "model cơ sở của Control Adapter không tương thích",
-                "ipAdapterIncompatibleBaseModel": "dạng model cơ sở của IP Adapter không tương thích",
-                "controlAdapterNoModelSelected": "không có model Control Adapter được chọn"
-            },
             "fluxModelIncompatibleBboxWidth": "$t(parameters.invoke.fluxRequiresDimensionsToBeMultipleOf16), chiều rộng hộp giới hạn là {{width}}",
             "noModelSelected": "Không có model được lựa chọn",
             "fluxModelIncompatibleScaledBboxHeight": "$t(parameters.invoke.fluxRequiresDimensionsToBeMultipleOf16), tỉ lệ chiều dài hộp giới hạn là {{height}}",
-            "canvasIsFiltering": "Canvas đang được lọc",
-            "canvasIsRasterizing": "Canvas đang được raster hoá",
-            "canvasIsTransforming": "Canvas đang được biến đổi",
-            "canvasIsCompositing": "Canvas đang được kết hợp",
+            "canvasIsFiltering": "Canvas đang bận (đang lọc)",
+            "canvasIsRasterizing": "Canvas đang bận (đang raster hoá)",
+            "canvasIsTransforming": "Canvas đang bận (đang biến đổi)",
+            "canvasIsCompositing": "Canvas đang bận (đang kết hợp)",
             "noPrompts": "Không có lệnh được tạo",
             "noNodesInGraph": "Không có node trong đồ thị",
             "addingImagesTo": "Thêm ảnh vào",
@@ -1430,8 +1428,12 @@
             "missingNodeTemplate": "Thiếu mẫu trình bày node",
             "fluxModelIncompatibleBboxHeight": "$t(parameters.invoke.fluxRequiresDimensionsToBeMultipleOf16), chiều dài hộp giới hạn là {{height}}",
             "fluxModelIncompatibleScaledBboxWidth": "$t(parameters.invoke.fluxRequiresDimensionsToBeMultipleOf16), tỉ lệ chiều rộng hộp giới hạn là {{width}}",
-            "missingInputForField": "{{nodeLabel}} -> {{fieldLabel}} thiếu đầu ra",
-            "missingFieldTemplate": "Thiếu vùng mẫu trình bày"
+            "missingInputForField": "{{nodeLabel}} -> {{fieldLabel}}: thiếu đầu vào",
+            "missingFieldTemplate": "Thiếu vùng mẫu trình bày",
+            "collectionEmpty": "{{nodeLabel}} -> {{fieldLabel}} tài nguyên trống",
+            "collectionTooFewItems": "{{nodeLabel}} -> {{fieldLabel}}: quá ít mục, tối thiểu {{minItems}}",
+            "collectionTooManyItems": "{{nodeLabel}} -> {{fieldLabel}}: quá nhiều mục, tối đa {{maxItems}}",
+            "canvasIsSelectingObject": "Canvas đang bận (đang chọn đồ vật)"
         },
         "cfgScale": "Thước Đo CFG",
         "useSeed": "Dùng Tham Số Hạt Giống",
@@ -1542,7 +1544,8 @@
         "resetWebUIDesc2": "Nếu ảnh không được xuất hiện trong thư viện hoặc điều gì đó không ổn đang diễn ra, hãy thử khởi động lại trước khi báo lỗi trên Github.",
         "displayInProgress": "Hiển Thị Hình Ảnh Đang Xử Lý",
         "intermediatesClearedFailed": "Có Vấn Đề Khi Dọn Sạch Sản Phẩm Trung Gian",
-        "enableInvisibleWatermark": "Bật Chế Độ Ẩn Watermark"
+        "enableInvisibleWatermark": "Bật Chế Độ Ẩn Watermark",
+        "showDetailedInvocationProgress": "Hiện Dữ Liệu Xử Lý"
     },
     "sdxl": {
         "loading": "Đang Tải...",
@@ -1594,7 +1597,7 @@
         "pullBboxIntoLayerError": "Có Vấn Đề Khi Chuyển Hộp Giới Hạn Thành Layer",
         "pullBboxIntoReferenceImageOk": "Chuyển Hộp Giới Hạn Thành Ảnh Mẫu",
         "clearCaches": "Xoá Bộ Nhớ Đệm",
-        "outputOnlyMaskedRegions": "Chỉ Xuất Đầu Ra Ở Vùng Phủ",
+        "outputOnlyMaskedRegions": "Chỉ Xuất Đầu Ra Ở Vùng Tạo Sinh",
         "addLayer": "Thêm Layer",
         "regional": "Khu Vực",
         "regionIsEmpty": "Vùng được chọn trống",
@@ -1608,10 +1611,13 @@
         "moveForward": "Chuyển Lên Đầu",
         "fitBboxToLayers": "Xếp Vừa Hộp Giới Hạn Vào Layer",
         "ipAdapterMethod": {
-            "full": "Đầy Đủ",
+            "full": "Phong Cách Và Thành Phần",
             "style": "Chỉ Lấy Phong Cách",
             "composition": "Chỉ Lấy Thành Phần",
-            "ipAdapterMethod": "Cách Thức IP Adapter"
+            "ipAdapterMethod": "Cách Thức",
+            "compositionDesc": "Áp dụng cách trình bày và bỏ qua phong cách mẫu.",
+            "fullDesc": "Áp dụng phong cách trực quan (màu, cấu tạo) & thành phần (cách trình bày).",
+            "styleDesc": "Áp dụng phong cách trực quan (màu, cấu tạo) và bỏ qua cách trình bày."
         },
         "deletePrompt": "Xoá Lệnh",
         "rasterLayer": "Layer Dạng Raster",
@@ -1899,7 +1905,33 @@
             "colorPicker": "Chọn Màu"
         },
         "mergingLayers": "Đang gộp layer",
-        "controlLayerEmptyState": "<UploadButton>Tải lên ảnh</UploadButton>, kéo thả ảnh từ <GalleryButton>thư viện</GalleryButton> vào layer này, hoặc vẽ trên canvas để bắt đầu."
+        "controlLayerEmptyState": "<UploadButton>Tải lên ảnh</UploadButton>, kéo thả ảnh từ <GalleryButton>thư viện</GalleryButton> vào layer này, hoặc vẽ trên canvas để bắt đầu.",
+        "referenceImageEmptyState": "<UploadButton>Tải lên ảnh</UploadButton> hoặc kéo thả ảnh từ <GalleryButton>thư viện</GalleryButton> vào layer này để bắt đầu.",
+        "useImage": "Dùng Hình Ảnh",
+        "resetCanvasLayers": "Khởi Động Lại Layer Canvas",
+        "asRasterLayer": "Như $t(controlLayers.rasterLayer)",
+        "asRasterLayerResize": "Như $t(controlLayers.rasterLayer) (Thay Đổi Kích Thước)",
+        "asControlLayer": "Như $t(controlLayers.controlLayer)",
+        "asControlLayerResize": "Như $t(controlLayers.controlLayer) (Thay Đổi Kích Thước)",
+        "newSession": "Phiên Làm Việc Mới",
+        "resetGenerationSettings": "Khởi Động Lại Cài Đặt Tạo Sinh",
+        "referenceImageRegional": "Ảnh Mẫu (Khu Vực)",
+        "referenceImageGlobal": "Ảnh Mẫu (Toàn Vùng)",
+        "warnings": {
+            "problemsFound": "Phát hiện vấn đề",
+            "unsupportedModel": "layer không được hỗ trợ cho model cơ sở này",
+            "controlAdapterNoModelSelected": "không có model được chọn cho Layer Chỉnh Sửa Được",
+            "controlAdapterNoControl": "chưa chọn/vẽ điều khiển",
+            "ipAdapterIncompatibleBaseModel": "model cơ sở cho Ảnh Mẫu không tương thích",
+            "ipAdapterNoImageSelected": "chưa chọn Ảnh Mẫu",
+            "controlAdapterIncompatibleBaseModel": "model cơ sở cho Layer Chỉnh Sửa Được không tương thích",
+            "ipAdapterNoModelSelected": "không có model được chọn cho Ảnh Mẫu",
+            "rgNoPromptsOrIPAdapters": "không có lệnh hoặc Ảnh Mẫu",
+            "rgNegativePromptNotSupported": "Lệnh Tiêu Cực không được hỗ trợ cho model cơ sở được chọn",
+            "rgReferenceImagesNotSupported": "Ảnh Mẫu Khu Vực không được hỗ trợ cho model cơ sở được chọn",
+            "rgAutoNegativeNotSupported": "Tự Động Đảo Chiều không được hỗ trợ cho model cơ sở được chọn",
+            "rgNoRegion": "không có khu vực được vẽ"
+        }
     },
     "stylePresets": {
         "negativePrompt": "Lệnh Tiêu Cực",
@@ -2124,8 +2156,8 @@
         "watchRecentReleaseVideos": "Xem Video Phát Hành Mới Nhất",
         "watchUiUpdatesOverview": "Xem Tổng Quan Về Những Cập Nhật Cho Giao Diện Người Dùng",
         "items": [
-            "<StrongComponent>SD 3.5</StrongComponent>: Hỗ trợ cho Từ ngữ Sang Hình Ảnh trong Workflow với phiên bản SD 3.5 Medium hoặc Large.",
-            "<StrongComponent>Canvas</StrongComponent>: Hợp lý hoá cách xử lý Layer Điều Khiển Được và cải thiện thiết lập điều khiển mặc định."
+            "<StrongComponent>Hướng Dẫn Khu Vực FLUX (beta)</StrongComponent>: Bản beta của Hướng Dẫn Khu Vực FLUX của chúng ta đã có mắt tại bảng điều khiển lệnh khu vực.",
+            "<StrongComponent>Nhiều Cải Tiến Ở UX</StrongComponent>: Một số nâng cấp nhỏ ở trải nghiệm và chất lượng người dùng trên toàn bộ ứng dụng."
         ]
     },
     "upsell": {
@@ -2133,5 +2165,67 @@
         "inviteTeammates": "Thêm Đồng Đội",
         "shareAccess": "Chia Sẻ Quyền Truy Cập",
         "professionalUpsell": "Không có sẵn Phiên Bản Chuyên Nghiệp cho Invoke. Bấm vào đây hoặc đến invoke.com/pricing để thêm chi tiết."
+    },
+    "supportVideos": {
+        "supportVideos": "Video Hỗ Trợ",
+        "gettingStarted": "Bắt Đầu Làm Quen",
+        "studioSessionsDesc1": "Xem thử <StudioSessionsPlaylistLink /> để hiểu rõ Invoke hơn.",
+        "studioSessionsDesc2": "Đến <DiscordLink /> để tham gia vào phiên trực tiếp và hỏi câu hỏi. Các phiên được tải lên danh sách phát vào các tuần.",
+        "videos": {
+            "howDoIDoImageToImageTransformation": {
+                "title": "Làm Sao Để Tôi Dùng Trình Biến Đổi Hình Ảnh Sang Hình Ảnh?",
+                "description": "Hướng dẫn cách thực hiện biến đổi ảnh sang ảnh trong Invoke."
+            },
+            "howDoIUseGlobalIPAdaptersAndReferenceImages": {
+                "description": "Giới thiệu về ảnh mẫu và IP adapter toàn vùng.",
+                "title": "Làm Sao Để Tôi Dùng IP Adapter Toàn Vùng Và Ảnh Mẫu?"
+            },
+            "creatingAndComposingOnInvokesControlCanvas": {
+                "description": "Học cách sáng tạo ảnh bằng trình điều khiển canvas của Invoke.",
+                "title": "Sáng Tạo Trong Trình Kiểm Soát Canvas Của Invoke"
+            },
+            "upscaling": {
+                "description": "Cách upscale ảnh bằng bộ công cụ của Invoke để nâng cấp độ phân giải.",
+                "title": "Upscale (Nâng Cấp Chất Lượng Hình Ảnh)"
+            },
+            "howDoIGenerateAndSaveToTheGallery": {
+                "title": "Làm Sao Để Tôi Tạo Sinh Và Lưu Vào Thư Viện?",
+                "description": "Các bước để tạo sinh và lưu ảnh vào thư viện."
+            },
+            "howDoIEditOnTheCanvas": {
+                "description": "Hướng dẫn chỉnh sửa ảnh trực tiếp trên canvas.",
+                "title": "Làm Sao Để Tôi Chỉnh Sửa Trên Canvas?"
+            },
+            "howDoIUseControlNetsAndControlLayers": {
+                "title": "Làm Sao Để Tôi Dùng ControlNet và Layer Điều Khiển Được?",
+                "description": "Học cách áp dụng layer điều khiển được và controlnet vào ảnh của bạn."
+            },
+            "howDoIUseInpaintMasks": {
+                "title": "Làm Sao Để Tôi Dùng Lớp Phủ Inpaint?",
+                "description": "Cách áp dụng lớp phủ inpaint vào chỉnh sửa và thay đổi ảnh."
+            },
+            "howDoIOutpaint": {
+                "title": "Làm Sao Để Tôi Outpaint?",
+                "description": "Hướng dẫn outpaint bên ngoài viền ảnh gốc."
+            },
+            "creatingYourFirstImage": {
+                "description": "Giới thiệu về cách tạo ảnh từ ban đầu bằng công cụ Invoke.",
+                "title": "Tạo Hình Ảnh Đầu Tiên Của Bạn"
+            },
+            "usingControlLayersAndReferenceGuides": {
+                "description": "Học cách chỉ dẫn ảnh được tạo ra bằng layer điều khiển được và ảnh mẫu.",
+                "title": "Dùng Layer Điều Khiển Được và Chỉ Dẫn Mẫu"
+            },
+            "understandingImageToImageAndDenoising": {
+                "title": "Hiểu Rõ Trình Hình Ảnh Sang Hình Ảnh Và Trình Khử Nhiễu",
+                "description": "Tổng quan về trình biến đổi ảnh sang ảnh và trình khử nhiễu trong Invoke."
+            },
+            "exploringAIModelsAndConceptAdapters": {
+                "title": "Khám Phá Model AI Và Khái Niệm Về Adapter",
+                "description": "Đào sâu vào model AI và cách dùng những adapter để điều khiển một cách sáng tạo."
+            }
+        },
+        "controlCanvas": "Điều Khiển Canvas",
+        "watch": "Xem"
     }
 }

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

@@ -661,19 +661,6 @@
             "missingFieldTemplate": "缺失模板",
             "addingImagesTo": "添加图像到",
             "noPrompts": "没有已生成的提示词",
-            "layer": {
-                "ipAdapterNoModelSelected": "未选择IP adapter",
-                "controlAdapterNoModelSelected": "未选择Control Adapter模型",
-                "rgNoPromptsOrIPAdapters": "无文本提示或IP Adapters",
-                "controlAdapterIncompatibleBaseModel": "Control Adapter的基础模型不兼容",
-                "ipAdapterIncompatibleBaseModel": "IP Adapter的基础模型不兼容",
-                "ipAdapterNoImageSelected": "未选择IP Adapter图像",
-                "rgNoRegion": "未选择区域",
-                "t2iAdapterIncompatibleBboxWidth": "$t(parameters.invoke.layer.t2iAdapterRequiresDimensionsToBeMultipleOf) {{multiple}}，边界框宽度为 {{width}}",
-                "t2iAdapterIncompatibleScaledBboxHeight": "$t(parameters.invoke.layer.t2iAdapterRequiresDimensionsToBeMultipleOf) {{multiple}}，缩放后的边界框高度为 {{height}}",
-                "t2iAdapterIncompatibleBboxHeight": "$t(parameters.invoke.layer.t2iAdapterRequiresDimensionsToBeMultipleOf) {{multiple}}，边界框高度为 {{height}}",
-                "t2iAdapterIncompatibleScaledBboxWidth": "$t(parameters.invoke.layer.t2iAdapterRequiresDimensionsToBeMultipleOf) {{multiple}}，缩放后的边界框宽度为 {{width}}"
-            },
             "canvasIsFiltering": "画布正在过滤",
             "fluxModelIncompatibleScaledBboxHeight": "$t(parameters.invoke.fluxRequiresDimensionsToBeMultipleOf16)，缩放后的边界框高度为 {{height}}",
             "noCLIPEmbedModelSelected": "未为FLUX生成选择CLIP嵌入模型",

invokeai/frontend/web/src/common/hooks/useGlobalHotkeys.ts

@@ -1,6 +1,6 @@
 import { useAppDispatch } from 'app/store/storeHooks';
-import { useClearQueue } from 'features/queue/components/ClearQueueConfirmationAlertDialog';
 import { useCancelCurrentQueueItem } from 'features/queue/hooks/useCancelCurrentQueueItem';
+import { useClearQueue } from 'features/queue/hooks/useClearQueue';
 import { useInvoke } from 'features/queue/hooks/useInvoke';
 import { useRegisteredHotkeys } from 'features/system/components/HotkeysModal/useHotkeyData';
 import { useFeatureStatus } from 'features/system/hooks/useFeatureStatus';

invokeai/frontend/web/src/features/controlLayers/components/CanvasAddEntityButtons.tsx

@@ -63,7 +63,7 @@ export const CanvasAddEntityButtons = memo(() => {
               justifyContent="flex-start"
               leftIcon={<PiPlusBold />}
               onClick={addRegionalGuidance}
-              isDisabled={isFLUX || isSD3}
+              isDisabled={isSD3}
             >
               {t('controlLayers.regionalGuidance')}
             </Button>

invokeai/frontend/web/src/features/controlLayers/components/CanvasEntityList/EntityListGlobalActionBarAddLayerMenu.tsx

@@ -49,7 +49,7 @@ export const EntityListGlobalActionBarAddLayerMenu = memo(() => {
           <MenuItem icon={<PiPlusBold />} onClick={addInpaintMask}>
             {t('controlLayers.inpaintMask')}
           </MenuItem>
-          <MenuItem icon={<PiPlusBold />} onClick={addRegionalGuidance} isDisabled={isFLUX || isSD3}>
+          <MenuItem icon={<PiPlusBold />} onClick={addRegionalGuidance} isDisabled={isSD3}>
             {t('controlLayers.regionalGuidance')}
           </MenuItem>
           <MenuItem icon={<PiPlusBold />} onClick={addRegionalReferenceImage} isDisabled={isFLUX || isSD3}>

invokeai/frontend/web/src/features/controlLayers/components/RegionalGuidance/RegionalGuidanceDeletePromptButton.tsx

@@ -1,27 +1,28 @@
-import { IconButton, Tooltip } from '@invoke-ai/ui-library';
+import type { IconButtonProps } from '@invoke-ai/ui-library';
+import { IconButton } from '@invoke-ai/ui-library';
 import { memo } from 'react';
 import { useTranslation } from 'react-i18next';
-import { PiTrashSimpleFill } from 'react-icons/pi';
+import { PiXBold } from 'react-icons/pi';
 
-type Props = {
+type Props = Omit<IconButtonProps, 'aria-label'> & {
   onDelete: () => void;
 };
 
-export const RegionalGuidanceDeletePromptButton = memo(({ onDelete }: Props) => {
+export const RegionalGuidanceDeletePromptButton = memo(({ onDelete, ...rest }: Props) => {
   const { t } = useTranslation();
   return (
-    <Tooltip label={t('controlLayers.deletePrompt')}>
-      <IconButton
-        variant="link"
-        aria-label={t('controlLayers.deletePrompt')}
-        icon={<PiTrashSimpleFill />}
-        onClick={onDelete}
-        flexGrow={0}
-        size="sm"
-        p={0}
-        colorScheme="error"
-      />
-    </Tooltip>
+    <IconButton
+      tooltip={t('common.delete')}
+      variant="link"
+      aria-label={t('common.delete')}
+      icon={<PiXBold />}
+      onClick={onDelete}
+      flexGrow={0}
+      size="sm"
+      p={0}
+      colorScheme="error"
+      {...rest}
+    />
   );
 });
 

invokeai/frontend/web/src/features/controlLayers/components/RegionalGuidance/RegionalGuidanceIPAdapterSettingsEmptyState.tsx

@@ -1,8 +1,9 @@
-import { Button, Flex, Text } from '@invoke-ai/ui-library';
+import { Button, Flex, IconButton, Spacer, Text } from '@invoke-ai/ui-library';
 import { useAppDispatch } from 'app/store/storeHooks';
 import { useImageUploadButton } from 'common/hooks/useImageUploadButton';
 import { useEntityIdentifierContext } from 'features/controlLayers/contexts/EntityIdentifierContext';
 import { useCanvasIsBusy } from 'features/controlLayers/hooks/useCanvasIsBusy';
+import { rgIPAdapterDeleted } from 'features/controlLayers/store/canvasSlice';
 import type { SetRegionalGuidanceReferenceImageDndTargetData } from 'features/dnd/dnd';
 import { setRegionalGuidanceReferenceImageDndTarget } from 'features/dnd/dnd';
 import { DndDropTarget } from 'features/dnd/DndDropTarget';
@@ -10,6 +11,7 @@ import { setRegionalGuidanceReferenceImage } from 'features/imageActions/actions
 import { activeTabCanvasRightPanelChanged } from 'features/ui/store/uiSlice';
 import { memo, useCallback, useMemo } from 'react';
 import { Trans, useTranslation } from 'react-i18next';
+import { PiXBold } from 'react-icons/pi';
 import type { ImageDTO } from 'services/api/types';
 
 type Props = {
@@ -31,6 +33,9 @@ export const RegionalGuidanceIPAdapterSettingsEmptyState = memo(({ referenceImag
   const onClickGalleryButton = useCallback(() => {
     dispatch(activeTabCanvasRightPanelChanged('gallery'));
   }, [dispatch]);
+  const onDeleteIPAdapter = useCallback(() => {
+    dispatch(rgIPAdapterDeleted({ entityIdentifier, referenceImageId }));
+  }, [dispatch, entityIdentifier, referenceImageId]);
 
   const dndTargetData = useMemo<SetRegionalGuidanceReferenceImageDndTargetData>(
     () =>
@@ -42,26 +47,44 @@ export const RegionalGuidanceIPAdapterSettingsEmptyState = memo(({ referenceImag
   );
 
   return (
-    <Flex flexDir="column" gap={3} position="relative" w="full" p={4}>
-      <Text textAlign="center" color="base.300">
-        <Trans
-          i18nKey="controlLayers.referenceImageEmptyState"
-          components={{
-            UploadButton: (
-              <Button
-                isDisabled={isBusy}
-                size="sm"
-                variant="link"
-                color="base.300"
-                {...uploadApi.getUploadButtonProps()}
-              />
-            ),
-            GalleryButton: (
-              <Button onClick={onClickGalleryButton} isDisabled={isBusy} size="sm" variant="link" color="base.300" />
-            ),
-          }}
+    <Flex flexDir="column" gap={2} position="relative" w="full">
+      <Flex alignItems="center" gap={2}>
+        <Text fontWeight="semibold" color="base.400">
+          {t('controlLayers.referenceImage')}
+        </Text>
+        <Spacer />
+        <IconButton
+          size="sm"
+          variant="link"
+          alignSelf="stretch"
+          icon={<PiXBold />}
+          tooltip={t('controlLayers.deleteReferenceImage')}
+          aria-label={t('controlLayers.deleteReferenceImage')}
+          onClick={onDeleteIPAdapter}
+          colorScheme="error"
         />
-      </Text>
+      </Flex>
+      <Flex alignItems="center" gap={2} p={4}>
+        <Text textAlign="center" color="base.300">
+          <Trans
+            i18nKey="controlLayers.referenceImageEmptyState"
+            components={{
+              UploadButton: (
+                <Button
+                  isDisabled={isBusy}
+                  size="sm"
+                  variant="link"
+                  color="base.300"
+                  {...uploadApi.getUploadButtonProps()}
+                />
+              ),
+              GalleryButton: (
+                <Button onClick={onClickGalleryButton} isDisabled={isBusy} size="sm" variant="link" color="base.300" />
+              ),
+            }}
+          />
+        </Text>
+      </Flex>
       <input {...uploadApi.getUploadInputProps()} />
       <DndDropTarget
         dndTarget={setRegionalGuidanceReferenceImageDndTarget}

invokeai/frontend/web/src/features/controlLayers/components/StagingArea/StagingAreaToolbar.tsx

@@ -5,6 +5,7 @@ import { StagingAreaToolbarDiscardSelectedButton } from 'features/controlLayers/
 import { StagingAreaToolbarImageCountButton } from 'features/controlLayers/components/StagingArea/StagingAreaToolbarImageCountButton';
 import { StagingAreaToolbarNextButton } from 'features/controlLayers/components/StagingArea/StagingAreaToolbarNextButton';
 import { StagingAreaToolbarPrevButton } from 'features/controlLayers/components/StagingArea/StagingAreaToolbarPrevButton';
+import { StagingAreaToolbarSaveAsMenu } from 'features/controlLayers/components/StagingArea/StagingAreaToolbarSaveAsMenu';
 import { StagingAreaToolbarSaveSelectedToGalleryButton } from 'features/controlLayers/components/StagingArea/StagingAreaToolbarSaveSelectedToGalleryButton';
 import { StagingAreaToolbarToggleShowResultsButton } from 'features/controlLayers/components/StagingArea/StagingAreaToolbarToggleShowResultsButton';
 import { memo } from 'react';
@@ -21,6 +22,7 @@ export const StagingAreaToolbar = memo(() => {
         <StagingAreaToolbarAcceptButton />
         <StagingAreaToolbarToggleShowResultsButton />
         <StagingAreaToolbarSaveSelectedToGalleryButton />
+        <StagingAreaToolbarSaveAsMenu />
         <StagingAreaToolbarDiscardSelectedButton />
         <StagingAreaToolbarDiscardAllButton />
       </ButtonGroup>

invokeai/frontend/web/src/features/controlLayers/components/StagingArea/StagingAreaToolbarSaveAsMenu.tsx

@@ -0,0 +1,136 @@
+import { IconButton, Menu, MenuButton, MenuItem, MenuList } from '@invoke-ai/ui-library';
+import { useAppStore } from 'app/store/nanostores/store';
+import { useAppSelector } from 'app/store/storeHooks';
+import { NewLayerIcon } from 'features/controlLayers/components/common/icons';
+import { selectSelectedImage } from 'features/controlLayers/store/canvasStagingAreaSlice';
+import { createNewCanvasEntityFromImage } from 'features/imageActions/actions';
+import { toast } from 'features/toast/toast';
+import { memo, useCallback } from 'react';
+import { useTranslation } from 'react-i18next';
+import { PiDotsThreeBold } from 'react-icons/pi';
+import { imageDTOToFile, uploadImage } from 'services/api/endpoints/images';
+
+const uploadImageArg = { image_category: 'general', is_intermediate: true, silent: true } as const;
+
+export const StagingAreaToolbarSaveAsMenu = memo(() => {
+  const { t } = useTranslation();
+  const selectedImage = useAppSelector(selectSelectedImage);
+  const store = useAppStore();
+
+  const onClickNewRasterLayerFromImage = useCallback(async () => {
+    if (!selectedImage) {
+      return;
+    }
+    const { dispatch, getState } = store;
+    const file = await imageDTOToFile(selectedImage.imageDTO);
+    const imageDTO = await uploadImage({ file, ...uploadImageArg });
+    createNewCanvasEntityFromImage({
+      imageDTO,
+      type: 'raster_layer',
+      dispatch,
+      getState,
+      overrides: { isEnabled: false }, // We are adding the layer while staging, it should be disabled by default
+    });
+    toast({
+      id: 'SENT_TO_CANVAS',
+      title: t('toast.sentToCanvas'),
+      status: 'success',
+    });
+  }, [selectedImage, store, t]);
+
+  const onClickNewControlLayerFromImage = useCallback(async () => {
+    if (!selectedImage) {
+      return;
+    }
+    const { dispatch, getState } = store;
+    const file = await imageDTOToFile(selectedImage.imageDTO);
+    const imageDTO = await uploadImage({ file, ...uploadImageArg });
+    createNewCanvasEntityFromImage({
+      imageDTO,
+      type: 'control_layer',
+      dispatch,
+      getState,
+      overrides: { isEnabled: false }, // We are adding the layer while staging, it should be disabled by default
+    });
+    toast({
+      id: 'SENT_TO_CANVAS',
+      title: t('toast.sentToCanvas'),
+      status: 'success',
+    });
+  }, [selectedImage, store, t]);
+
+  const onClickNewInpaintMaskFromImage = useCallback(async () => {
+    if (!selectedImage) {
+      return;
+    }
+    const { dispatch, getState } = store;
+    const file = await imageDTOToFile(selectedImage.imageDTO);
+    const imageDTO = await uploadImage({ file, ...uploadImageArg });
+    createNewCanvasEntityFromImage({
+      imageDTO,
+      type: 'inpaint_mask',
+      dispatch,
+      getState,
+      overrides: { isEnabled: false }, // We are adding the layer while staging, it should be disabled by default
+    });
+    toast({
+      id: 'SENT_TO_CANVAS',
+      title: t('toast.sentToCanvas'),
+      status: 'success',
+    });
+  }, [selectedImage, store, t]);
+
+  const onClickNewRegionalGuidanceFromImage = useCallback(async () => {
+    if (!selectedImage) {
+      return;
+    }
+    const { dispatch, getState } = store;
+    const file = await imageDTOToFile(selectedImage.imageDTO);
+    const imageDTO = await uploadImage({ file, ...uploadImageArg });
+    createNewCanvasEntityFromImage({
+      imageDTO,
+      type: 'regional_guidance',
+      dispatch,
+      getState,
+      overrides: { isEnabled: false }, // We are adding the layer while staging, it should be disabled by default
+    });
+    toast({
+      id: 'SENT_TO_CANVAS',
+      title: t('toast.sentToCanvas'),
+      status: 'success',
+    });
+  }, [selectedImage, store, t]);
+
+  return (
+    <Menu>
+      <MenuButton
+        as={IconButton}
+        aria-label={t('controlLayers.newLayerFromImage')}
+        tooltip={t('controlLayers.newLayerFromImage')}
+        icon={<PiDotsThreeBold />}
+        colorScheme="invokeBlue"
+        isDisabled={!selectedImage}
+      />
+      <MenuList>
+        <MenuItem icon={<NewLayerIcon />} onClickCapture={onClickNewInpaintMaskFromImage} isDisabled={!selectedImage}>
+          {t('controlLayers.inpaintMask')}
+        </MenuItem>
+        <MenuItem
+          icon={<NewLayerIcon />}
+          onClickCapture={onClickNewRegionalGuidanceFromImage}
+          isDisabled={!selectedImage}
+        >
+          {t('controlLayers.regionalGuidance')}
+        </MenuItem>
+        <MenuItem icon={<NewLayerIcon />} onClickCapture={onClickNewControlLayerFromImage} isDisabled={!selectedImage}>
+          {t('controlLayers.controlLayer')}
+        </MenuItem>
+        <MenuItem icon={<NewLayerIcon />} onClickCapture={onClickNewRasterLayerFromImage} isDisabled={!selectedImage}>
+          {t('controlLayers.rasterLayer')}
+        </MenuItem>
+      </MenuList>
+    </Menu>
+  );
+});
+
+StagingAreaToolbarSaveAsMenu.displayName = 'StagingAreaToolbarSaveAsMenu';

invokeai/frontend/web/src/features/controlLayers/components/StagingArea/StagingAreaToolbarSaveSelectedToGalleryButton.tsx

@@ -1,6 +1,4 @@
 import { IconButton } from '@invoke-ai/ui-library';
-import { useStore } from '@nanostores/react';
-import { $authToken } from 'app/store/nanostores/authToken';
 import { useAppSelector } from 'app/store/storeHooks';
 import { withResultAsync } from 'common/util/result';
 import { selectSelectedImage } from 'features/controlLayers/store/canvasStagingAreaSlice';
@@ -9,14 +7,13 @@ import { toast } from 'features/toast/toast';
 import { memo, useCallback } from 'react';
 import { useTranslation } from 'react-i18next';
 import { PiFloppyDiskBold } from 'react-icons/pi';
-import { uploadImage } from 'services/api/endpoints/images';
+import { imageDTOToFile, uploadImage } from 'services/api/endpoints/images';
 
 const TOAST_ID = 'SAVE_STAGING_AREA_IMAGE_TO_GALLERY';
 
 export const StagingAreaToolbarSaveSelectedToGalleryButton = memo(() => {
   const autoAddBoardId = useAppSelector(selectAutoAddBoardId);
   const selectedImage = useAppSelector(selectSelectedImage);
-  const authToken = useStore($authToken);
 
   const { t } = useTranslation();
 
@@ -28,18 +25,8 @@ export const StagingAreaToolbarSaveSelectedToGalleryButton = memo(() => {
     // To save the image to gallery, we will download it and re-upload it. This allows the user to delete the image
     // the gallery without borking the canvas, which may need this image to exist.
     const result = await withResultAsync(async () => {
-      // Download the image
-      const requestOpts = authToken
-        ? {
-            headers: {
-              Authorization: `Bearer ${authToken}`,
-            },
-          }
-        : {};
-      const res = await fetch(selectedImage.imageDTO.image_url, requestOpts);
-      const blob = await res.blob();
       // Create a new file with the same name, which we will upload
-      const file = new File([blob], `copy_of_${selectedImage.imageDTO.image_name}`, { type: 'image/png' });
+      const file = await imageDTOToFile(selectedImage.imageDTO);
 
       await uploadImage({
         file,
@@ -66,7 +53,7 @@ export const StagingAreaToolbarSaveSelectedToGalleryButton = memo(() => {
         status: 'error',
       });
     }
-  }, [autoAddBoardId, selectedImage, t, authToken]);
+  }, [autoAddBoardId, selectedImage, t]);
 
   return (
     <IconButton

invokeai/frontend/web/src/features/controlLayers/components/common/CanvasEntityHeaderCommonActions.tsx

@@ -1,6 +1,7 @@
 import { Flex } from '@invoke-ai/ui-library';
 import { CanvasEntityDeleteButton } from 'features/controlLayers/components/common/CanvasEntityDeleteButton';
 import { CanvasEntityEnabledToggle } from 'features/controlLayers/components/common/CanvasEntityEnabledToggle';
+import { CanvasEntityHeaderWarnings } from 'features/controlLayers/components/common/CanvasEntityHeaderWarnings';
 import { CanvasEntityIsBookmarkedForQuickSwitchToggle } from 'features/controlLayers/components/common/CanvasEntityIsBookmarkedForQuickSwitchToggle';
 import { CanvasEntityIsLockedToggle } from 'features/controlLayers/components/common/CanvasEntityIsLockedToggle';
 import { useEntityIdentifierContext } from 'features/controlLayers/contexts/EntityIdentifierContext';
@@ -11,6 +12,7 @@ export const CanvasEntityHeaderCommonActions = memo(() => {
 
   return (
     <Flex alignSelf="stretch">
+      <CanvasEntityHeaderWarnings />
       <CanvasEntityIsBookmarkedForQuickSwitchToggle />
       {entityIdentifier.type !== 'reference_image' && <CanvasEntityIsLockedToggle />}
       <CanvasEntityEnabledToggle />

invokeai/frontend/web/src/features/controlLayers/components/common/CanvasEntityHeaderWarnings.tsx

@@ -0,0 +1,101 @@
+import { Flex, IconButton, ListItem, Text, UnorderedList } from '@invoke-ai/ui-library';
+import { createSelector } from '@reduxjs/toolkit';
+import { EMPTY_ARRAY } from 'app/store/constants';
+import { useAppSelector } from 'app/store/storeHooks';
+import { useEntityIdentifierContext } from 'features/controlLayers/contexts/EntityIdentifierContext';
+import { useEntityIsEnabled } from 'features/controlLayers/hooks/useEntityIsEnabled';
+import { selectModel } from 'features/controlLayers/store/paramsSlice';
+import { selectCanvasSlice, selectEntityOrThrow } from 'features/controlLayers/store/selectors';
+import type { CanvasEntityIdentifier } from 'features/controlLayers/store/types';
+import {
+  getControlLayerWarnings,
+  getGlobalReferenceImageWarnings,
+  getInpaintMaskWarnings,
+  getRasterLayerWarnings,
+  getRegionalGuidanceWarnings,
+} from 'features/controlLayers/store/validators';
+import type { TFunction } from 'i18next';
+import { upperFirst } from 'lodash-es';
+import { memo, useMemo } from 'react';
+import { useTranslation } from 'react-i18next';
+import { PiWarningBold } from 'react-icons/pi';
+import type { Equals } from 'tsafe';
+import { assert } from 'tsafe';
+
+const buildSelectWarnings = (entityIdentifier: CanvasEntityIdentifier, t: TFunction) => {
+  return createSelector(selectCanvasSlice, selectModel, (canvas, model) => {
+    // This component is used within a <CanvasEntityStateGate /> so we can safely assume that the entity exists.
+    // Should never throw.
+    const entity = selectEntityOrThrow(canvas, entityIdentifier, 'CanvasEntityHeaderWarnings');
+
+    let warnings: string[] = [];
+
+    const entityType = entity.type;
+
+    if (entityType === 'control_layer') {
+      warnings = getControlLayerWarnings(entity, model);
+    } else if (entityType === 'regional_guidance') {
+      warnings = getRegionalGuidanceWarnings(entity, model);
+    } else if (entityType === 'inpaint_mask') {
+      warnings = getInpaintMaskWarnings(entity, model);
+    } else if (entityType === 'raster_layer') {
+      warnings = getRasterLayerWarnings(entity, model);
+    } else if (entityType === 'reference_image') {
+      warnings = getGlobalReferenceImageWarnings(entity, model);
+    } else {
+      assert<Equals<typeof entityType, never>>(false, 'Unexpected entity type');
+    }
+
+    // Return a stable reference if there are no warnings
+    if (warnings.length === 0) {
+      return EMPTY_ARRAY;
+    }
+
+    return warnings.map((w) => t(w)).map(upperFirst);
+  });
+};
+
+export const CanvasEntityHeaderWarnings = memo(() => {
+  const entityIdentifier = useEntityIdentifierContext();
+  const { t } = useTranslation();
+  const isEnabled = useEntityIsEnabled(entityIdentifier);
+  const selectWarnings = useMemo(() => buildSelectWarnings(entityIdentifier, t), [entityIdentifier, t]);
+  const warnings = useAppSelector(selectWarnings);
+
+  if (warnings.length === 0) {
+    return null;
+  }
+
+  return (
+    // Using IconButton here bc it matches the styling of the actual buttons in the header without any fanagling, but
+    // it's not a button
+    <IconButton
+      as="span"
+      size="sm"
+      variant="link"
+      alignSelf="stretch"
+      aria-label="warnings"
+      tooltip={<TooltipContent warnings={warnings} />}
+      icon={<PiWarningBold />}
+      colorScheme="warning"
+      isDisabled={!isEnabled}
+    />
+  );
+});
+
+CanvasEntityHeaderWarnings.displayName = 'CanvasEntityHeaderWarnings';
+
+const TooltipContent = memo((props: { warnings: string[] }) => {
+  const { t } = useTranslation();
+  return (
+    <Flex flexDir="column">
+      <Text>{t('controlLayers.warnings.problemsFound')}:</Text>
+      <UnorderedList>
+        {props.warnings.map((warning, index) => (
+          <ListItem key={index}>{warning}</ListItem>
+        ))}
+      </UnorderedList>
+    </Flex>
+  );
+});
+TooltipContent.displayName = 'TooltipContent';

invokeai/frontend/web/src/features/controlLayers/hooks/addLayerHooks.ts

@@ -29,7 +29,13 @@ import { modelConfigsAdapterSelectors, selectModelConfigsQuery } from 'services/
 import type { ControlNetModelConfig, IPAdapterModelConfig, T2IAdapterModelConfig } from 'services/api/types';
 import { isControlNetOrT2IAdapterModelConfig, isIPAdapterModelConfig } from 'services/api/types';
 
-/** @knipignore */
+/**
+ * Selects the default control adapter configuration based on the model configurations and the base.
+ *
+ * Be sure to clone the output of this selector before modifying it!
+ *
+ * @knipignore
+ */
 export const selectDefaultControlAdapter = createSelector(
   selectModelConfigsQuery,
   selectBase,
@@ -52,6 +58,11 @@ export const selectDefaultControlAdapter = createSelector(
   }
 );
 
+/**
+ * Selects the default IP adapter configuration based on the model configurations and the base.
+ *
+ * Be sure to clone the output of this selector before modifying it!
+ */
 export const selectDefaultIPAdapter = createSelector(
   selectModelConfigsQuery,
   selectBase,
@@ -117,7 +128,9 @@ export const useAddRegionalReferenceImage = () => {
 
   const func = useCallback(() => {
     const overrides: Partial<CanvasRegionalGuidanceState> = {
-      referenceImages: [{ id: getPrefixedId('regional_guidance_reference_image'), ipAdapter: defaultIPAdapter }],
+      referenceImages: [
+        { id: getPrefixedId('regional_guidance_reference_image'), ipAdapter: deepClone(defaultIPAdapter) },
+      ],
     };
     dispatch(rgAdded({ isSelected: true, overrides }));
   }, [defaultIPAdapter, dispatch]);
@@ -129,7 +142,7 @@ export const useAddGlobalReferenceImage = () => {
   const dispatch = useAppDispatch();
   const defaultIPAdapter = useAppSelector(selectDefaultIPAdapter);
   const func = useCallback(() => {
-    const overrides = { ipAdapter: defaultIPAdapter };
+    const overrides = { ipAdapter: deepClone(defaultIPAdapter) };
     dispatch(referenceImageAdded({ isSelected: true, overrides }));
   }, [defaultIPAdapter, dispatch]);
 
@@ -140,7 +153,7 @@ export const useAddRegionalGuidanceIPAdapter = (entityIdentifier: CanvasEntityId
   const dispatch = useAppDispatch();
   const defaultIPAdapter = useAppSelector(selectDefaultIPAdapter);
   const func = useCallback(() => {
-    dispatch(rgIPAdapterAdded({ entityIdentifier, overrides: { ipAdapter: defaultIPAdapter } }));
+    dispatch(rgIPAdapterAdded({ entityIdentifier, overrides: { ipAdapter: deepClone(defaultIPAdapter) } }));
   }, [defaultIPAdapter, dispatch, entityIdentifier]);
 
   return func;

invokeai/frontend/web/src/features/controlLayers/store/paramsSlice.ts

@@ -24,6 +24,7 @@ import type {
   ParameterSeed,
   ParameterSteps,
   ParameterStrength,
+  ParameterStructuralLoRAModel,
   ParameterT5EncoderModel,
   ParameterVAEModel,
 } from 'features/parameters/types/parameterSchemas';
@@ -75,6 +76,7 @@ export type ParamsState = {
   clipEmbedModel: ParameterCLIPEmbedModel | null;
   clipLEmbedModel: ParameterCLIPLEmbedModel | null;
   clipGEmbedModel: ParameterCLIPGEmbedModel | null;
+  structuralLora: ParameterStructuralLoRAModel | null;
 };
 
 const initialState: ParamsState = {
@@ -121,6 +123,7 @@ const initialState: ParamsState = {
   clipEmbedModel: null,
   clipLEmbedModel: null,
   clipGEmbedModel: null,
+  structuralLora: null,
 };
 
 export const paramsSlice = createSlice({
@@ -195,6 +198,9 @@ export const paramsSlice = createSlice({
     t5EncoderModelSelected: (state, action: PayloadAction<ParameterT5EncoderModel | null>) => {
       state.t5EncoderModel = action.payload;
     },
+    structuralLoRAModelSelected: (state, action: PayloadAction<ParameterStructuralLoRAModel | null>) => {
+      state.structuralLora = action.payload;
+    },
     clipEmbedModelSelected: (state, action: PayloadAction<ParameterCLIPEmbedModel | null>) => {
       state.clipEmbedModel = action.payload;
     },

invokeai/frontend/web/src/features/controlLayers/store/validators.ts

@@ -0,0 +1,153 @@
+import type {
+  CanvasControlLayerState,
+  CanvasInpaintMaskState,
+  CanvasRasterLayerState,
+  CanvasReferenceImageState,
+  CanvasRegionalGuidanceState,
+} from 'features/controlLayers/store/types';
+import type { ParameterModel } from 'features/parameters/types/parameterSchemas';
+
+const WARNINGS = {
+  UNSUPPORTED_MODEL: 'controlLayers.warnings.unsupportedModel',
+  RG_NO_PROMPTS_OR_IP_ADAPTERS: 'controlLayers.warnings.rgNoPromptsOrIPAdapters',
+  RG_NEGATIVE_PROMPT_NOT_SUPPORTED: 'controlLayers.warnings.rgNegativePromptNotSupported',
+  RG_REFERENCE_IMAGES_NOT_SUPPORTED: 'controlLayers.warnings.rgReferenceImagesNotSupported',
+  RG_AUTO_NEGATIVE_NOT_SUPPORTED: 'controlLayers.warnings.rgAutoNegativeNotSupported',
+  RG_NO_REGION: 'controlLayers.warnings.rgNoRegion',
+  IP_ADAPTER_NO_MODEL_SELECTED: 'controlLayers.warnings.ipAdapterNoModelSelected',
+  IP_ADAPTER_INCOMPATIBLE_BASE_MODEL: 'controlLayers.warnings.ipAdapterIncompatibleBaseModel',
+  IP_ADAPTER_NO_IMAGE_SELECTED: 'controlLayers.warnings.ipAdapterNoImageSelected',
+  CONTROL_ADAPTER_NO_MODEL_SELECTED: 'controlLayers.warnings.controlAdapterNoModelSelected',
+  CONTROL_ADAPTER_INCOMPATIBLE_BASE_MODEL: 'controlLayers.warnings.controlAdapterIncompatibleBaseModel',
+  CONTROL_ADAPTER_NO_CONTROL: 'controlLayers.warnings.controlAdapterNoControl',
+} as const;
+
+type WarningTKey = (typeof WARNINGS)[keyof typeof WARNINGS];
+
+export const getRegionalGuidanceWarnings = (
+  entity: CanvasRegionalGuidanceState,
+  model: ParameterModel | null
+): WarningTKey[] => {
+  const warnings: WarningTKey[] = [];
+
+  if (entity.objects.length === 0) {
+    // Layer is in empty state
+    warnings.push(WARNINGS.RG_NO_REGION);
+  }
+
+  if (entity.positivePrompt === null && entity.negativePrompt === null && entity.referenceImages.length === 0) {
+    // Must have at least 1 prompt or IP Adapter
+    warnings.push(WARNINGS.RG_NO_PROMPTS_OR_IP_ADAPTERS);
+  }
+
+  if (model) {
+    if (model.base === 'sd-3' || model.base === 'sd-2') {
+      // Unsupported model architecture
+      warnings.push(WARNINGS.UNSUPPORTED_MODEL);
+    } else if (model.base === 'flux') {
+      // Some features are not supported for flux models
+      if (entity.negativePrompt !== null) {
+        warnings.push(WARNINGS.RG_NEGATIVE_PROMPT_NOT_SUPPORTED);
+      }
+      if (entity.referenceImages.length > 0) {
+        warnings.push(WARNINGS.RG_REFERENCE_IMAGES_NOT_SUPPORTED);
+      }
+      if (entity.autoNegative) {
+        warnings.push(WARNINGS.RG_AUTO_NEGATIVE_NOT_SUPPORTED);
+      }
+    } else {
+      entity.referenceImages.forEach(({ ipAdapter }) => {
+        if (!ipAdapter.model) {
+          // No model selected
+          warnings.push(WARNINGS.IP_ADAPTER_NO_MODEL_SELECTED);
+        } else if (ipAdapter.model.base !== model.base) {
+          // Supported model architecture but doesn't match
+          warnings.push(WARNINGS.IP_ADAPTER_INCOMPATIBLE_BASE_MODEL);
+        }
+
+        if (!ipAdapter.image) {
+          // No image selected
+          warnings.push(WARNINGS.IP_ADAPTER_NO_IMAGE_SELECTED);
+        }
+      });
+    }
+  }
+
+  return warnings;
+};
+
+export const getGlobalReferenceImageWarnings = (
+  entity: CanvasReferenceImageState,
+  model: ParameterModel | null
+): WarningTKey[] => {
+  const warnings: WarningTKey[] = [];
+
+  if (!entity.ipAdapter.model) {
+    // No model selected
+    warnings.push(WARNINGS.IP_ADAPTER_NO_MODEL_SELECTED);
+  } else if (model) {
+    if (model.base === 'sd-3' || model.base === 'sd-2') {
+      // Unsupported model architecture
+      warnings.push(WARNINGS.UNSUPPORTED_MODEL);
+    } else if (entity.ipAdapter.model.base !== model.base) {
+      // Supported model architecture but doesn't match
+      warnings.push(WARNINGS.IP_ADAPTER_INCOMPATIBLE_BASE_MODEL);
+    }
+  }
+
+  if (!entity.ipAdapter.image) {
+    // No image selected
+    warnings.push(WARNINGS.IP_ADAPTER_NO_IMAGE_SELECTED);
+  }
+
+  return warnings;
+};
+
+export const getControlLayerWarnings = (
+  entity: CanvasControlLayerState,
+  model: ParameterModel | null
+): WarningTKey[] => {
+  const warnings: WarningTKey[] = [];
+
+  if (entity.objects.length === 0) {
+    // Layer is in empty state
+    warnings.push(WARNINGS.CONTROL_ADAPTER_NO_CONTROL);
+  }
+
+  if (!entity.controlAdapter.model) {
+    // No model selected
+    warnings.push(WARNINGS.CONTROL_ADAPTER_NO_MODEL_SELECTED);
+  } else if (model) {
+    if (model.base === 'sd-3' || model.base === 'sd-2') {
+      // Unsupported model architecture
+      warnings.push(WARNINGS.UNSUPPORTED_MODEL);
+    } else if (entity.controlAdapter.model.base !== model.base) {
+      // Supported model architecture but doesn't match
+      warnings.push(WARNINGS.CONTROL_ADAPTER_INCOMPATIBLE_BASE_MODEL);
+    }
+  }
+
+  return warnings;
+};
+
+export const getRasterLayerWarnings = (
+  _entity: CanvasRasterLayerState,
+  _model: ParameterModel | null
+): WarningTKey[] => {
+  const warnings: WarningTKey[] = [];
+
+  // There are no warnings at the moment for raster layers.
+
+  return warnings;
+};
+
+export const getInpaintMaskWarnings = (
+  _entity: CanvasInpaintMaskState,
+  _model: ParameterModel | null
+): WarningTKey[] => {
+  const warnings: WarningTKey[] = [];
+
+  // There are no warnings at the moment for inpaint masks.
+
+  return warnings;
+};

invokeai/frontend/web/src/features/gallery/components/ImageContextMenu/ImageMenuItemNewLayerFromImageSubMenu.tsx

@@ -77,6 +77,32 @@ export const ImageMenuItemNewLayerFromImageSubMenu = memo(() => {
     });
   }, [imageDTO, imageViewer, store, t]);
 
+  const onClickNewRegionalReferenceImageFromImage = useCallback(() => {
+    const { dispatch, getState } = store;
+    createNewCanvasEntityFromImage({ imageDTO, type: 'reference_image', dispatch, getState });
+    dispatch(sentImageToCanvas());
+    dispatch(setActiveTab('canvas'));
+    imageViewer.close();
+    toast({
+      id: 'SENT_TO_CANVAS',
+      title: t('toast.sentToCanvas'),
+      status: 'success',
+    });
+  }, [imageDTO, imageViewer, store, t]);
+
+  const onClickNewGlobalReferenceImageFromImage = useCallback(() => {
+    const { dispatch, getState } = store;
+    createNewCanvasEntityFromImage({ imageDTO, type: 'regional_guidance_with_reference_image', dispatch, getState });
+    dispatch(sentImageToCanvas());
+    dispatch(setActiveTab('canvas'));
+    imageViewer.close();
+    toast({
+      id: 'SENT_TO_CANVAS',
+      title: t('toast.sentToCanvas'),
+      status: 'success',
+    });
+  }, [imageDTO, imageViewer, store, t]);
+
   return (
     <MenuItem {...subMenu.parentMenuItemProps} icon={<PiPlusBold />}>
       <Menu {...subMenu.menuProps}>
@@ -104,6 +130,20 @@ export const ImageMenuItemNewLayerFromImageSubMenu = memo(() => {
           <MenuItem icon={<NewLayerIcon />} onClickCapture={onClickNewRasterLayerFromImage} isDisabled={isBusy}>
             {t('controlLayers.rasterLayer')}
           </MenuItem>
+          <MenuItem
+            icon={<NewLayerIcon />}
+            onClickCapture={onClickNewRegionalReferenceImageFromImage}
+            isDisabled={isBusy}
+          >
+            {t('controlLayers.referenceImageRegional')}
+          </MenuItem>
+          <MenuItem
+            icon={<NewLayerIcon />}
+            onClickCapture={onClickNewGlobalReferenceImageFromImage}
+            isDisabled={isBusy}
+          >
+            {t('controlLayers.referenceImageGlobal')}
+          </MenuItem>
         </MenuList>
       </Menu>
     </MenuItem>

invokeai/frontend/web/src/features/imageActions/actions.ts

@@ -20,6 +20,7 @@ import { selectBboxModelBase, selectBboxRect } from 'features/controlLayers/stor
 import type {
   CanvasControlLayerState,
   CanvasEntityIdentifier,
+  CanvasEntityState,
   CanvasEntityType,
   CanvasInpaintMaskState,
   CanvasRasterLayerState,
@@ -134,14 +135,16 @@ export const createNewCanvasEntityFromImage = (arg: {
   type: CanvasEntityType | 'regional_guidance_with_reference_image';
   dispatch: AppDispatch;
   getState: () => RootState;
+  overrides?: Partial<Pick<CanvasEntityState, 'isEnabled' | 'isLocked' | 'name'>>;
 }) => {
-  const { type, imageDTO, dispatch, getState } = arg;
+  const { type, imageDTO, dispatch, getState, overrides: _overrides } = arg;
   const state = getState();
   const imageObject = imageDTOToImageObject(imageDTO);
   const { x, y } = selectBboxRect(state);
   const overrides = {
     objects: [imageObject],
     position: { x, y },
+    ..._overrides,
   };
   switch (type) {
     case 'raster_layer': {
@@ -166,13 +169,13 @@ export const createNewCanvasEntityFromImage = (arg: {
       break;
     }
     case 'reference_image': {
-      const ipAdapter = selectDefaultIPAdapter(getState());
+      const ipAdapter = deepClone(selectDefaultIPAdapter(getState()));
       ipAdapter.image = imageDTOToImageWithDims(imageDTO);
       dispatch(referenceImageAdded({ overrides: { ipAdapter }, isSelected: true }));
       break;
     }
     case 'regional_guidance_with_reference_image': {
-      const ipAdapter = selectDefaultIPAdapter(getState());
+      const ipAdapter = deepClone(selectDefaultIPAdapter(getState()));
       ipAdapter.image = imageDTOToImageWithDims(imageDTO);
       const referenceImages = [{ id: getPrefixedId('regional_guidance_reference_image'), ipAdapter }];
       dispatch(rgAdded({ overrides: { referenceImages }, isSelected: true }));
@@ -288,14 +291,14 @@ export const newCanvasFromImage = (arg: {
       break;
     }
     case 'reference_image': {
-      const ipAdapter = selectDefaultIPAdapter(getState());
+      const ipAdapter = deepClone(selectDefaultIPAdapter(getState()));
       ipAdapter.image = imageDTOToImageWithDims(imageDTO);
       dispatch(canvasReset());
       dispatch(referenceImageAdded({ overrides: { ipAdapter }, isSelected: true }));
       break;
     }
     case 'regional_guidance_with_reference_image': {
-      const ipAdapter = selectDefaultIPAdapter(getState());
+      const ipAdapter = deepClone(selectDefaultIPAdapter(getState()));
       ipAdapter.image = imageDTOToImageWithDims(imageDTO);
       const referenceImages = [{ id: getPrefixedId('regional_guidance_reference_image'), ipAdapter }];
       dispatch(canvasReset());

invokeai/frontend/web/src/features/metadata/util/parsers.ts

@@ -46,6 +46,7 @@ import type {
   ParameterSeed,
   ParameterSteps,
   ParameterStrength,
+  ParameterStructuralLoRAModel,
   ParameterVAEModel,
   ParameterWidth,
 } from 'features/parameters/types/parameterSchemas';
@@ -80,6 +81,7 @@ import {
   isLoRAModelConfig,
   isNonRefinerMainModelConfig,
   isRefinerMainModelModelConfig,
+  isStructuralLoRAModelConfig,
   isT2IAdapterModelConfig,
   isVAEModelConfig,
 } from 'services/api/types';
@@ -226,6 +228,14 @@ const parseVAEModel: MetadataParseFunc<ParameterVAEModel> = async (metadata) =>
   return modelIdentifier;
 };
 
+const parseStructuralLoRAModel: MetadataParseFunc<ParameterStructuralLoRAModel> = async (metadata) => {
+  const slora = await getProperty(metadata, 'structural_lora', undefined);
+  const key = await getModelKey(slora, 'structural_lora');
+  const sloraModelConfig = await fetchModelConfigWithTypeGuard(key, isStructuralLoRAModelConfig);
+  const modelIdentifier = zModelIdentifierField.parse(sloraModelConfig);
+  return modelIdentifier;
+};
+
 const parseLoRA: MetadataParseFunc<LoRA> = async (metadataItem) => {
   // Previously, the LoRA model identifier parts were stored in the LoRA metadata: `{key: ..., weight: 0.75}`
   const modelV1 = await getProperty(metadataItem, 'lora', undefined);
@@ -671,6 +681,7 @@ export const parsers = {
   mainModel: parseMainModel,
   refinerModel: parseRefinerModel,
   vaeModel: parseVAEModel,
+  structuralLora: parseStructuralLoRAModel,
   lora: parseLoRA,
   loras: parseAllLoRAs,
   controlNet: parseControlNet,

invokeai/frontend/web/src/features/modelManagerV2/subpanels/ModelManagerPanel/ModelList.tsx

@@ -18,6 +18,7 @@ import {
   useMainModels,
   useRefinerModels,
   useSpandrelImageToImageModels,
+  useStructuralLoRAModel,
   useT2IAdapterModels,
   useT5EncoderModels,
   useVAEModels,
@@ -92,6 +93,12 @@ const ModelList = () => {
     [t5EncoderModels, searchTerm, filteredModelType]
   );
 
+  const [structuralLoRAModels, { isLoading: isLoadingStructuralLoRAModels }] = useStructuralLoRAModel();
+  const filteredStructuralLoRAModels = useMemo(
+    () => modelsFilter(structuralLoRAModels, searchTerm, filteredModelType),
+    [structuralLoRAModels, searchTerm, filteredModelType]
+  );
+
   const [clipEmbedModels, { isLoading: isLoadingClipEmbedModels }] = useCLIPEmbedModels({ excludeSubmodels: true });
   const filteredClipEmbedModels = useMemo(
     () => modelsFilter(clipEmbedModels, searchTerm, filteredModelType),
@@ -118,7 +125,8 @@ const ModelList = () => {
       filteredVAEModels.length +
       filteredSpandrelImageToImageModels.length +
       t5EncoderModels.length +
-      clipEmbedModels.length
+      clipEmbedModels.length +
+      structuralLoRAModels.length
     );
   }, [
     filteredControlNetModels.length,
@@ -133,6 +141,7 @@ const ModelList = () => {
     filteredSpandrelImageToImageModels.length,
     t5EncoderModels.length,
     clipEmbedModels.length,
+    structuralLoRAModels.length,
   ]);
 
   return (
@@ -195,6 +204,15 @@ const ModelList = () => {
         {!isLoadingT5EncoderModels && filteredT5EncoderModels.length > 0 && (
           <ModelListWrapper title={t('modelManager.t5Encoder')} modelList={filteredT5EncoderModels} key="t5-encoder" />
         )}
+        {/* Structural Lora List */}
+        {isLoadingStructuralLoRAModels && <FetchingModelsLoader loadingMessage="Loading Structural Loras..." />}
+        {!isLoadingStructuralLoRAModels && filteredStructuralLoRAModels.length > 0 && (
+          <ModelListWrapper
+            title={t('modelManager.structuralLora')}
+            modelList={filteredStructuralLoRAModels}
+            key="structural-lora"
+          />
+        )}
         {/* Clip Embed List */}
         {isLoadingClipEmbedModels && <FetchingModelsLoader loadingMessage="Loading Clip Embed Models..." />}
         {!isLoadingClipEmbedModels && filteredClipEmbedModels.length > 0 && (

invokeai/frontend/web/src/features/modelManagerV2/subpanels/ModelManagerPanel/ModelTypeFilter.tsx

@@ -24,6 +24,7 @@ export const ModelTypeFilter = memo(() => {
       ip_adapter: t('common.ipAdapter'),
       clip_vision: 'CLIP Vision',
       spandrel_image_to_image: t('modelManager.spandrelImageToImage'),
+      structural_lora: t('modelManager.structuralLora'),
     }),
     [t]
   );

invokeai/frontend/web/src/features/nodes/components/flow/nodes/Invocation/fields/InputFieldRenderer.tsx

@@ -51,6 +51,8 @@ import {
   isSpandrelImageToImageModelFieldInputTemplate,
   isStringFieldInputInstance,
   isStringFieldInputTemplate,
+  isStructuralLoRAModelFieldInputInstance,
+  isStructuralLoRAModelFieldInputTemplate,
   isT2IAdapterModelFieldInputInstance,
   isT2IAdapterModelFieldInputTemplate,
   isT5EncoderModelFieldInputInstance,
@@ -81,6 +83,7 @@ import SD3MainModelFieldInputComponent from './inputs/SD3MainModelFieldInputComp
 import SDXLMainModelFieldInputComponent from './inputs/SDXLMainModelFieldInputComponent';
 import SpandrelImageToImageModelFieldInputComponent from './inputs/SpandrelImageToImageModelFieldInputComponent';
 import StringFieldInputComponent from './inputs/StringFieldInputComponent';
+import StructuralLoRAModelFieldInputComponent from './inputs/StructuralLoraModelFieldInputComponent';
 import T2IAdapterModelFieldInputComponent from './inputs/T2IAdapterModelFieldInputComponent';
 import T5EncoderModelFieldInputComponent from './inputs/T5EncoderModelFieldInputComponent';
 import VAEModelFieldInputComponent from './inputs/VAEModelFieldInputComponent';
@@ -156,6 +159,15 @@ const InputFieldRenderer = ({ nodeId, fieldName }: InputFieldProps) => {
     return <CLIPGEmbedModelFieldInputComponent nodeId={nodeId} field={fieldInstance} fieldTemplate={fieldTemplate} />;
   }
 
+  if (
+    isStructuralLoRAModelFieldInputInstance(fieldInstance) &&
+    isStructuralLoRAModelFieldInputTemplate(fieldTemplate)
+  ) {
+    return (
+      <StructuralLoRAModelFieldInputComponent nodeId={nodeId} field={fieldInstance} fieldTemplate={fieldTemplate} />
+    );
+  }
+
   if (isFluxVAEModelFieldInputInstance(fieldInstance) && isFluxVAEModelFieldInputTemplate(fieldTemplate)) {
     return <FluxVAEModelFieldInputComponent nodeId={nodeId} field={fieldInstance} fieldTemplate={fieldTemplate} />;
   }

invokeai/frontend/web/src/features/nodes/components/flow/nodes/Invocation/fields/inputs/StructuralLoraModelFieldInputComponent.tsx

@@ -0,0 +1,65 @@
+import { Combobox, Flex, FormControl, Tooltip } from '@invoke-ai/ui-library';
+import { useAppDispatch, useAppSelector } from 'app/store/storeHooks';
+import { useGroupedModelCombobox } from 'common/hooks/useGroupedModelCombobox';
+import { fieldStructuralLoRAModelValueChanged } from 'features/nodes/store/nodesSlice';
+import type {
+  StructuralLoRAModelFieldInputInstance,
+  StructuralLoRAModelFieldInputTemplate,
+} from 'features/nodes/types/field';
+import { memo, useCallback } from 'react';
+import { useTranslation } from 'react-i18next';
+import { useStructuralLoRAModel } from 'services/api/hooks/modelsByType';
+import { isStructuralLoRAModelConfig, type StructuralLoRAModelConfig } from 'services/api/types';
+
+import type { FieldComponentProps } from './types';
+
+type Props = FieldComponentProps<StructuralLoRAModelFieldInputInstance, StructuralLoRAModelFieldInputTemplate>;
+
+const StructuralLoRAModelFieldInputComponent = (props: Props) => {
+  const { nodeId, field } = props;
+  const { t } = useTranslation();
+  const disabledTabs = useAppSelector((s) => s.config.disabledTabs);
+  const dispatch = useAppDispatch();
+  const [modelConfigs, { isLoading }] = useStructuralLoRAModel();
+
+  const _onChange = useCallback(
+    (value: StructuralLoRAModelConfig | null) => {
+      if (!value) {
+        return;
+      }
+      dispatch(
+        fieldStructuralLoRAModelValueChanged({
+          nodeId,
+          fieldName: field.name,
+          value,
+        })
+      );
+    },
+    [dispatch, field.name, nodeId]
+  );
+  const { options, value, onChange, placeholder, noOptionsMessage } = useGroupedModelCombobox({
+    modelConfigs: modelConfigs.filter((config) => isStructuralLoRAModelConfig(config)),
+    onChange: _onChange,
+    isLoading,
+    selectedModel: field.value,
+  });
+  const required = props.fieldTemplate.required;
+
+  return (
+    <Flex w="full" alignItems="center" gap={2}>
+      <Tooltip label={!disabledTabs.includes('models') && t('modelManager.starterModelsInModelManager')}>
+        <FormControl className="nowheel nodrag" isDisabled={!options.length} isInvalid={!value && required}>
+          <Combobox
+            value={value}
+            placeholder={required ? placeholder : `(Optional) ${placeholder}`}
+            options={options}
+            onChange={onChange}
+            noOptionsMessage={noOptionsMessage}
+          />
+        </FormControl>
+      </Tooltip>
+    </Flex>
+  );
+};
+
+export default memo(StructuralLoRAModelFieldInputComponent);

invokeai/frontend/web/src/features/nodes/store/nodesSlice.ts

@@ -28,6 +28,7 @@ import type {
   SpandrelImageToImageModelFieldValue,
   StatefulFieldValue,
   StringFieldValue,
+  StructuralLoRAModelFieldValue,
   T2IAdapterModelFieldValue,
   T5EncoderModelFieldValue,
   VAEModelFieldValue,
@@ -55,6 +56,7 @@ import {
   zSpandrelImageToImageModelFieldValue,
   zStatefulFieldValue,
   zStringFieldValue,
+  zStructuralLoRAModelFieldValue,
   zT2IAdapterModelFieldValue,
   zT5EncoderModelFieldValue,
   zVAEModelFieldValue,
@@ -369,6 +371,9 @@ export const nodesSlice = createSlice({
     fieldCLIPGEmbedValueChanged: (state, action: FieldValueAction<CLIPGEmbedModelFieldValue>) => {
       fieldValueReducer(state, action, zCLIPGEmbedModelFieldValue);
     },
+    fieldStructuralLoRAModelValueChanged: (state, action: FieldValueAction<StructuralLoRAModelFieldValue>) => {
+      fieldValueReducer(state, action, zStructuralLoRAModelFieldValue);
+    },
     fieldFluxVAEModelValueChanged: (state, action: FieldValueAction<FluxVAEModelFieldValue>) => {
       fieldValueReducer(state, action, zFluxVAEModelFieldValue);
     },
@@ -438,6 +443,7 @@ export const {
   fieldCLIPEmbedValueChanged,
   fieldCLIPLEmbedValueChanged,
   fieldCLIPGEmbedValueChanged,
+  fieldStructuralLoRAModelValueChanged,
   fieldFluxVAEModelValueChanged,
   nodeEditorReset,
   nodeIsIntermediateChanged,

invokeai/frontend/web/src/features/nodes/types/common.ts

@@ -69,6 +69,7 @@ const zModelType = z.enum([
   'main',
   'vae',
   'lora',
+  'structural_lora',
   'controlnet',
   't2i_adapter',
   'ip_adapter',

invokeai/frontend/web/src/features/nodes/types/field.ts

@@ -178,6 +178,10 @@ const zCLIPGEmbedModelFieldType = zFieldTypeBase.extend({
   name: z.literal('CLIPGEmbedModelField'),
   originalType: zStatelessFieldType.optional(),
 });
+const zStructuralLoRAModelFieldType = zFieldTypeBase.extend({
+  name: z.literal('StructuralLoRAModelField'),
+  originalType: zStatelessFieldType.optional(),
+});
 const zFluxVAEModelFieldType = zFieldTypeBase.extend({
   name: z.literal('FluxVAEModelField'),
   originalType: zStatelessFieldType.optional(),
@@ -210,6 +214,7 @@ const zStatefulFieldType = z.union([
   zCLIPEmbedModelFieldType,
   zCLIPLEmbedModelFieldType,
   zCLIPGEmbedModelFieldType,
+  zStructuralLoRAModelFieldType,
   zFluxVAEModelFieldType,
   zColorFieldType,
   zSchedulerFieldType,
@@ -864,6 +869,29 @@ export const isCLIPGEmbedModelFieldInputTemplate = (val: unknown): val is CLIPGE
 
 // #endregion
 
+// #region StructuralLoRAModelField
+
+export const zStructuralLoRAModelFieldValue = zModelIdentifierField.optional();
+const zStructuralLoRAModelFieldInputInstance = zFieldInputInstanceBase.extend({
+  value: zStructuralLoRAModelFieldValue,
+});
+const zStructuralLoRAModelFieldInputTemplate = zFieldInputTemplateBase.extend({
+  type: zStructuralLoRAModelFieldType,
+  originalType: zFieldType.optional(),
+  default: zStructuralLoRAModelFieldValue,
+});
+
+export type StructuralLoRAModelFieldValue = z.infer<typeof zCLIPLEmbedModelFieldValue>;
+
+export type StructuralLoRAModelFieldInputInstance = z.infer<typeof zStructuralLoRAModelFieldInputInstance>;
+export type StructuralLoRAModelFieldInputTemplate = z.infer<typeof zStructuralLoRAModelFieldInputTemplate>;
+export const isStructuralLoRAModelFieldInputInstance = (val: unknown): val is StructuralLoRAModelFieldInputInstance =>
+  zStructuralLoRAModelFieldInputInstance.safeParse(val).success;
+export const isStructuralLoRAModelFieldInputTemplate = (val: unknown): val is StructuralLoRAModelFieldInputTemplate =>
+  zStructuralLoRAModelFieldInputTemplate.safeParse(val).success;
+
+// #endregion
+
 // #region SchedulerField
 
 export const zSchedulerFieldValue = zSchedulerField.optional();
@@ -959,6 +987,7 @@ export const zStatefulFieldValue = z.union([
   zCLIPEmbedModelFieldValue,
   zCLIPLEmbedModelFieldValue,
   zCLIPGEmbedModelFieldValue,
+  zStructuralLoRAModelFieldValue,
   zColorFieldValue,
   zSchedulerFieldValue,
 ]);
@@ -1030,6 +1059,7 @@ const zStatefulFieldInputTemplate = z.union([
   zCLIPEmbedModelFieldInputTemplate,
   zCLIPLEmbedModelFieldInputTemplate,
   zCLIPGEmbedModelFieldInputTemplate,
+  zStructuralLoRAModelFieldInputTemplate,
   zColorFieldInputTemplate,
   zSchedulerFieldInputTemplate,
   zStatelessFieldInputTemplate,

invokeai/frontend/web/src/features/nodes/util/graph/generation/addControlAdapters.ts

@@ -1,35 +1,41 @@
 import { logger } from 'app/logging/logger';
 import { withResultAsync } from 'common/util/result';
 import type { CanvasManager } from 'features/controlLayers/konva/CanvasManager';
-import type {
-  CanvasControlLayerState,
-  ControlNetConfig,
-  Rect,
-  T2IAdapterConfig,
-} from 'features/controlLayers/store/types';
+import type { CanvasControlLayerState, Rect } from 'features/controlLayers/store/types';
+import { getControlLayerWarnings } from 'features/controlLayers/store/validators';
 import type { Graph } from 'features/nodes/util/graph/generation/Graph';
+import type { ParameterModel } from 'features/parameters/types/parameterSchemas';
 import { serializeError } from 'serialize-error';
-import type { BaseModelType, ImageDTO, Invocation } from 'services/api/types';
+import type { ImageDTO, Invocation } from 'services/api/types';
 import { assert } from 'tsafe';
 
 const log = logger('system');
 
+type AddControlNetsArg = {
+  manager: CanvasManager;
+  entities: CanvasControlLayerState[];
+  g: Graph;
+  rect: Rect;
+  collector: Invocation<'collect'>;
+  model: ParameterModel;
+};
+
 type AddControlNetsResult = {
   addedControlNets: number;
 };
 
-export const addControlNets = async (
-  manager: CanvasManager,
-  layers: CanvasControlLayerState[],
-  g: Graph,
-  rect: Rect,
-  collector: Invocation<'collect'>,
-  base: BaseModelType
-): Promise<AddControlNetsResult> => {
-  const validControlLayers = layers
-    .filter((layer) => layer.isEnabled)
-    .filter((layer) => isValidControlAdapter(layer.controlAdapter, base))
-    .filter((layer) => layer.controlAdapter.type === 'controlnet');
+export const addControlNets = async ({
+  manager,
+  entities,
+  g,
+  rect,
+  collector,
+  model,
+}: AddControlNetsArg): Promise<AddControlNetsResult> => {
+  const validControlLayers = entities
+    .filter((entity) => entity.isEnabled)
+    .filter((entity) => entity.controlAdapter.type === 'controlnet')
+    .filter((entity) => getControlLayerWarnings(entity, model).length === 0);
 
   const result: AddControlNetsResult = {
     addedControlNets: 0,
@@ -54,22 +60,31 @@ export const addControlNets = async (
   return result;
 };
 
+type AddT2IAdaptersArg = {
+  manager: CanvasManager;
+  entities: CanvasControlLayerState[];
+  g: Graph;
+  rect: Rect;
+  collector: Invocation<'collect'>;
+  model: ParameterModel;
+};
+
 type AddT2IAdaptersResult = {
   addedT2IAdapters: number;
 };
 
-export const addT2IAdapters = async (
-  manager: CanvasManager,
-  layers: CanvasControlLayerState[],
-  g: Graph,
-  rect: Rect,
-  collector: Invocation<'collect'>,
-  base: BaseModelType
-): Promise<AddT2IAdaptersResult> => {
-  const validControlLayers = layers
-    .filter((layer) => layer.isEnabled)
-    .filter((layer) => isValidControlAdapter(layer.controlAdapter, base))
-    .filter((layer) => layer.controlAdapter.type === 't2i_adapter');
+export const addT2IAdapters = async ({
+  manager,
+  entities,
+  g,
+  rect,
+  collector,
+  model,
+}: AddT2IAdaptersArg): Promise<AddT2IAdaptersResult> => {
+  const validControlLayers = entities
+    .filter((entity) => entity.isEnabled)
+    .filter((entity) => entity.controlAdapter.type === 't2i_adapter')
+    .filter((entity) => getControlLayerWarnings(entity, model).length === 0);
 
   const result: AddT2IAdaptersResult = {
     addedT2IAdapters: 0,
@@ -145,11 +160,3 @@ const addT2IAdapterToGraph = (
 
   g.addEdge(t2iAdapter, 't2i_adapter', collector, 'item');
 };
-
-const isValidControlAdapter = (controlAdapter: ControlNetConfig | T2IAdapterConfig, base: BaseModelType): boolean => {
-  // Must be have a model
-  const hasModel = Boolean(controlAdapter.model);
-  // Model must match the current base model
-  const modelMatchesBase = controlAdapter.model?.base === base;
-  return hasModel && modelMatchesBase;
-};

invokeai/frontend/web/src/features/nodes/util/graph/generation/addIPAdapters.ts

@@ -1,19 +1,25 @@
 import type { CanvasReferenceImageState } from 'features/controlLayers/store/types';
+import { getGlobalReferenceImageWarnings } from 'features/controlLayers/store/validators';
 import type { Graph } from 'features/nodes/util/graph/generation/Graph';
-import type { BaseModelType, Invocation } from 'services/api/types';
+import type { ParameterModel } from 'features/parameters/types/parameterSchemas';
+import type { Invocation } from 'services/api/types';
 import { assert } from 'tsafe';
 
 type AddIPAdaptersResult = {
   addedIPAdapters: number;
 };
 
-export const addIPAdapters = (
-  ipAdapters: CanvasReferenceImageState[],
-  g: Graph,
-  collector: Invocation<'collect'>,
-  base: BaseModelType
-): AddIPAdaptersResult => {
-  const validIPAdapters = ipAdapters.filter((entity) => isValidIPAdapter(entity, base));
+type AddIPAdaptersArg = {
+  entities: CanvasReferenceImageState[];
+  g: Graph;
+  collector: Invocation<'collect'>;
+  model: ParameterModel;
+};
+
+export const addIPAdapters = ({ entities, g, collector, model }: AddIPAdaptersArg): AddIPAdaptersResult => {
+  const validIPAdapters = entities
+    .filter((entity) => entity.isEnabled)
+    .filter((entity) => getGlobalReferenceImageWarnings(entity, model).length === 0);
 
   const result: AddIPAdaptersResult = {
     addedIPAdapters: 0,
@@ -76,11 +82,3 @@ const addIPAdapter = (entity: CanvasReferenceImageState, g: Graph, collector: In
 
   g.addEdge(ipAdapterNode, 'ip_adapter', collector, 'item');
 };
-
-const isValidIPAdapter = ({ isEnabled, ipAdapter }: CanvasReferenceImageState, base: BaseModelType): boolean => {
-  // Must be have a model that matches the current base and must have a control image
-  const hasModel = Boolean(ipAdapter.model);
-  const modelMatchesBase = ipAdapter.model?.base === base;
-  const hasImage = Boolean(ipAdapter.image);
-  return isEnabled && hasModel && modelMatchesBase && hasImage;
-};

invokeai/frontend/web/src/features/nodes/util/graph/generation/addRegions.ts

@@ -3,15 +3,12 @@ import { deepClone } from 'common/util/deepClone';
 import { withResultAsync } from 'common/util/result';
 import type { CanvasManager } from 'features/controlLayers/konva/CanvasManager';
 import { getPrefixedId } from 'features/controlLayers/konva/util';
-import type {
-  CanvasRegionalGuidanceState,
-  IPAdapterConfig,
-  Rect,
-  RegionalGuidanceReferenceImageState,
-} from 'features/controlLayers/store/types';
+import type { CanvasRegionalGuidanceState, Rect } from 'features/controlLayers/store/types';
+import { getRegionalGuidanceWarnings } from 'features/controlLayers/store/validators';
 import type { Graph } from 'features/nodes/util/graph/generation/Graph';
+import type { ParameterModel } from 'features/parameters/types/parameterSchemas';
 import { serializeError } from 'serialize-error';
-import type { BaseModelType, Invocation } from 'services/api/types';
+import type { Invocation } from 'services/api/types';
 import { assert } from 'tsafe';
 
 const log = logger('system');
@@ -23,19 +20,26 @@ type AddedRegionResult = {
   addedIPAdapters: number;
 };
 
-const isValidRegion = (rg: CanvasRegionalGuidanceState, base: BaseModelType) => {
-  const isEnabled = rg.isEnabled;
-  const hasTextPrompt = Boolean(rg.positivePrompt || rg.negativePrompt);
-  const hasIPAdapter = rg.referenceImages.filter(({ ipAdapter }) => isValidIPAdapter(ipAdapter, base)).length > 0;
-  return isEnabled && (hasTextPrompt || hasIPAdapter);
+type AddRegionsArg = {
+  manager: CanvasManager;
+  regions: CanvasRegionalGuidanceState[];
+  g: Graph;
+  bbox: Rect;
+  model: ParameterModel;
+  posCond: Invocation<'compel' | 'sdxl_compel_prompt' | 'flux_text_encoder'>;
+  negCond: Invocation<'compel' | 'sdxl_compel_prompt' | 'flux_text_encoder'> | null;
+  posCondCollect: Invocation<'collect'>;
+  negCondCollect: Invocation<'collect'> | null;
+  ipAdapterCollect: Invocation<'collect'>;
 };
 
 /**
  * Adds regional guidance to the graph
+ * @param manager The canvas manager
  * @param regions Array of regions to add
  * @param g The graph to add the layers to
- * @param base The base model type
- * @param denoise The main denoise node
+ * @param bbox The bounding box
+ * @param model The main model
  * @param posCond The positive conditioning node
  * @param negCond The negative conditioning node
  * @param posCondCollect The positive conditioning collector
@@ -44,22 +48,25 @@ const isValidRegion = (rg: CanvasRegionalGuidanceState, base: BaseModelType) =>
  * @returns A promise that resolves to the regions that were successfully added to the graph
  */
 
-export const addRegions = async (
-  manager: CanvasManager,
-  regions: CanvasRegionalGuidanceState[],
-  g: Graph,
-  bbox: Rect,
-  base: BaseModelType,
-  denoise: Invocation<'denoise_latents'>,
-  posCond: Invocation<'compel'> | Invocation<'sdxl_compel_prompt'>,
-  negCond: Invocation<'compel'> | Invocation<'sdxl_compel_prompt'>,
-  posCondCollect: Invocation<'collect'>,
-  negCondCollect: Invocation<'collect'>,
-  ipAdapterCollect: Invocation<'collect'>
-): Promise<AddedRegionResult[]> => {
-  const isSDXL = base === 'sdxl';
+export const addRegions = async ({
+  manager,
+  regions,
+  g,
+  bbox,
+  model,
+  posCond,
+  negCond,
+  posCondCollect,
+  negCondCollect,
+  ipAdapterCollect,
+}: AddRegionsArg): Promise<AddedRegionResult[]> => {
+  const isSDXL = model.base === 'sdxl';
+  const isFLUX = model.base === 'flux';
+
+  const validRegions = regions
+    .filter((entity) => entity.isEnabled)
+    .filter((entity) => getRegionalGuidanceWarnings(entity, model).length === 0);
 
-  const validRegions = regions.filter((rg) => isValidRegion(rg, base));
   const results: AddedRegionResult[] = [];
 
   for (const region of validRegions) {
@@ -94,20 +101,27 @@ export const addRegions = async (
     if (region.positivePrompt) {
       // The main positive conditioning node
       result.addedPositivePrompt = true;
-      const regionalPosCond = g.addNode(
-        isSDXL
-          ? {
-              type: 'sdxl_compel_prompt',
-              id: getPrefixedId('prompt_region_positive_cond'),
-              prompt: region.positivePrompt,
-              style: region.positivePrompt, // TODO: Should we put the positive prompt in both fields?
-            }
-          : {
-              type: 'compel',
-              id: getPrefixedId('prompt_region_positive_cond'),
-              prompt: region.positivePrompt,
-            }
-      );
+      let regionalPosCond: Invocation<'compel' | 'sdxl_compel_prompt' | 'flux_text_encoder'>;
+      if (isSDXL) {
+        regionalPosCond = g.addNode({
+          type: 'sdxl_compel_prompt',
+          id: getPrefixedId('prompt_region_positive_cond'),
+          prompt: region.positivePrompt,
+          style: region.positivePrompt, // TODO: Should we put the positive prompt in both fields?
+        });
+      } else if (isFLUX) {
+        regionalPosCond = g.addNode({
+          type: 'flux_text_encoder',
+          id: getPrefixedId('prompt_region_positive_cond'),
+          prompt: region.positivePrompt,
+        });
+      } else {
+        regionalPosCond = g.addNode({
+          type: 'compel',
+          id: getPrefixedId('prompt_region_positive_cond'),
+          prompt: region.positivePrompt,
+        });
+      }
       // Connect the mask to the conditioning
       g.addEdge(maskToTensor, 'mask', regionalPosCond, 'mask');
       // Connect the conditioning to the collector
@@ -115,38 +129,55 @@ export const addRegions = async (
       // Copy the connections to the "global" positive conditioning node to the regional cond
       if (posCond.type === 'compel') {
         for (const edge of g.getEdgesTo(posCond, ['clip', 'mask'])) {
-          // Clone the edge, but change the destination node to the regional conditioning node
+          const clone = deepClone(edge);
+          clone.destination.node_id = regionalPosCond.id;
+          g.addEdgeFromObj(clone);
+        }
+      } else if (posCond.type === 'sdxl_compel_prompt') {
+        for (const edge of g.getEdgesTo(posCond, ['clip', 'clip2', 'mask'])) {
+          const clone = deepClone(edge);
+          clone.destination.node_id = regionalPosCond.id;
+          g.addEdgeFromObj(clone);
+        }
+      } else if (posCond.type === 'flux_text_encoder') {
+        for (const edge of g.getEdgesTo(posCond, ['clip', 't5_encoder', 't5_max_seq_len', 'mask'])) {
           const clone = deepClone(edge);
           clone.destination.node_id = regionalPosCond.id;
           g.addEdgeFromObj(clone);
         }
       } else {
-        for (const edge of g.getEdgesTo(posCond, ['clip', 'clip2', 'mask'])) {
-          // Clone the edge, but change the destination node to the regional conditioning node
-          const clone = deepClone(edge);
-          clone.destination.node_id = regionalPosCond.id;
-          g.addEdgeFromObj(clone);
-        }
+        assert(false, 'Unsupported positive conditioning node type.');
       }
     }
 
     if (region.negativePrompt) {
-      result.addedNegativePrompt = true;
+      assert(negCond, 'Negative conditioning node is required if there is a negative prompt');
+      assert(negCondCollect, 'Negative conditioning collector is required if there is a negative prompt');
+
       // The main negative conditioning node
-      const regionalNegCond = g.addNode(
-        isSDXL
-          ? {
-              type: 'sdxl_compel_prompt',
-              id: getPrefixedId('prompt_region_negative_cond'),
-              prompt: region.negativePrompt,
-              style: region.negativePrompt,
-            }
-          : {
-              type: 'compel',
-              id: getPrefixedId('prompt_region_negative_cond'),
-              prompt: region.negativePrompt,
-            }
-      );
+      result.addedNegativePrompt = true;
+      let regionalNegCond: Invocation<'compel' | 'sdxl_compel_prompt' | 'flux_text_encoder'>;
+      if (isSDXL) {
+        regionalNegCond = g.addNode({
+          type: 'sdxl_compel_prompt',
+          id: getPrefixedId('prompt_region_negative_cond'),
+          prompt: region.negativePrompt,
+          style: region.negativePrompt,
+        });
+      } else if (isFLUX) {
+        regionalNegCond = g.addNode({
+          type: 'flux_text_encoder',
+          id: getPrefixedId('prompt_region_negative_cond'),
+          prompt: region.negativePrompt,
+        });
+      } else {
+        regionalNegCond = g.addNode({
+          type: 'compel',
+          id: getPrefixedId('prompt_region_negative_cond'),
+          prompt: region.negativePrompt,
+        });
+      }
+
       // Connect the mask to the conditioning
       g.addEdge(maskToTensor, 'mask', regionalNegCond, 'mask');
       // Connect the conditioning to the collector
@@ -158,17 +189,27 @@ export const addRegions = async (
           clone.destination.node_id = regionalNegCond.id;
           g.addEdgeFromObj(clone);
         }
-      } else {
+      } else if (negCond.type === 'sdxl_compel_prompt') {
         for (const edge of g.getEdgesTo(negCond, ['clip', 'clip2', 'mask'])) {
           const clone = deepClone(edge);
           clone.destination.node_id = regionalNegCond.id;
           g.addEdgeFromObj(clone);
         }
+      } else if (negCond.type === 'flux_text_encoder') {
+        for (const edge of g.getEdgesTo(negCond, ['clip', 't5_encoder', 't5_max_seq_len', 'mask'])) {
+          const clone = deepClone(edge);
+          clone.destination.node_id = regionalNegCond.id;
+          g.addEdgeFromObj(clone);
+        }
+      } else {
+        assert(false, 'Unsupported negative conditioning node type.');
       }
     }
 
     // If we are using the "invert" auto-negative setting, we need to add an additional negative conditioning node
     if (region.autoNegative && region.positivePrompt) {
+      assert(negCondCollect, 'Negative conditioning collector is required if there is an auto-negative setting');
+
       result.addedAutoNegativePositivePrompt = true;
       // We re-use the mask image, but invert it when converting to tensor
       const invertTensorMask = g.addNode({
@@ -178,20 +219,27 @@ export const addRegions = async (
       // Connect the OG mask image to the inverted mask-to-tensor node
       g.addEdge(maskToTensor, 'mask', invertTensorMask, 'mask');
       // Create the conditioning node. It's going to be connected to the negative cond collector, but it uses the positive prompt
-      const regionalPosCondInverted = g.addNode(
-        isSDXL
-          ? {
-              type: 'sdxl_compel_prompt',
-              id: getPrefixedId('prompt_region_positive_cond_inverted'),
-              prompt: region.positivePrompt,
-              style: region.positivePrompt,
-            }
-          : {
-              type: 'compel',
-              id: getPrefixedId('prompt_region_positive_cond_inverted'),
-              prompt: region.positivePrompt,
-            }
-      );
+      let regionalPosCondInverted: Invocation<'compel' | 'sdxl_compel_prompt' | 'flux_text_encoder'>;
+      if (isSDXL) {
+        regionalPosCondInverted = g.addNode({
+          type: 'sdxl_compel_prompt',
+          id: getPrefixedId('prompt_region_positive_cond_inverted'),
+          prompt: region.positivePrompt,
+          style: region.positivePrompt,
+        });
+      } else if (isFLUX) {
+        regionalPosCondInverted = g.addNode({
+          type: 'flux_text_encoder',
+          id: getPrefixedId('prompt_region_positive_cond_inverted'),
+          prompt: region.positivePrompt,
+        });
+      } else {
+        regionalPosCondInverted = g.addNode({
+          type: 'compel',
+          id: getPrefixedId('prompt_region_positive_cond_inverted'),
+          prompt: region.positivePrompt,
+        });
+      }
       // Connect the inverted mask to the conditioning
       g.addEdge(invertTensorMask, 'mask', regionalPosCondInverted, 'mask');
       // Connect the conditioning to the negative collector
@@ -203,20 +251,26 @@ export const addRegions = async (
           clone.destination.node_id = regionalPosCondInverted.id;
           g.addEdgeFromObj(clone);
         }
-      } else {
+      } else if (posCond.type === 'sdxl_compel_prompt') {
         for (const edge of g.getEdgesTo(posCond, ['clip', 'clip2', 'mask'])) {
           const clone = deepClone(edge);
           clone.destination.node_id = regionalPosCondInverted.id;
           g.addEdgeFromObj(clone);
         }
+      } else if (posCond.type === 'flux_text_encoder') {
+        for (const edge of g.getEdgesTo(posCond, ['clip', 't5_encoder', 't5_max_seq_len', 'mask'])) {
+          const clone = deepClone(edge);
+          clone.destination.node_id = regionalPosCondInverted.id;
+          g.addEdgeFromObj(clone);
+        }
+      } else {
+        assert(false, 'Unsupported positive conditioning node type.');
       }
     }
 
-    const validRGIPAdapters: RegionalGuidanceReferenceImageState[] = region.referenceImages.filter(({ ipAdapter }) =>
-      isValidIPAdapter(ipAdapter, base)
-    );
+    for (const { id, ipAdapter } of region.referenceImages) {
+      assert(!isFLUX, 'Regional IP adapters are not supported for FLUX.');
 
-    for (const { id, ipAdapter } of validRGIPAdapters) {
       result.addedIPAdapters++;
       const { weight, model, clipVisionModel, method, beginEndStepPct, image } = ipAdapter;
       assert(model, 'IP Adapter model is required');
@@ -248,11 +302,3 @@ export const addRegions = async (
 
   return results;
 };
-
-const isValidIPAdapter = (ipAdapter: IPAdapterConfig, base: BaseModelType): boolean => {
-  // Must be have a model that matches the current base and must have a control image
-  const hasModel = Boolean(ipAdapter.model);
-  const modelMatchesBase = ipAdapter.model?.base === base;
-  const hasImage = Boolean(ipAdapter.image);
-  return hasModel && modelMatchesBase && hasImage;
-};

invokeai/frontend/web/src/features/nodes/util/graph/generation/buildFLUXGraph.ts

@@ -11,6 +11,7 @@ import { addImageToImage } from 'features/nodes/util/graph/generation/addImageTo
 import { addInpaint } from 'features/nodes/util/graph/generation/addInpaint';
 import { addNSFWChecker } from 'features/nodes/util/graph/generation/addNSFWChecker';
 import { addOutpaint } from 'features/nodes/util/graph/generation/addOutpaint';
+import { addRegions } from 'features/nodes/util/graph/generation/addRegions';
 import { addTextToImage } from 'features/nodes/util/graph/generation/addTextToImage';
 import { addWatermarker } from 'features/nodes/util/graph/generation/addWatermarker';
 import { Graph } from 'features/nodes/util/graph/generation/Graph';
@@ -79,7 +80,10 @@ export const buildFLUXGraph = async (
     id: getPrefixedId('flux_text_encoder'),
     prompt: positivePrompt,
   });
-
+  const posCondCollect = g.addNode({
+    type: 'collect',
+    id: getPrefixedId('pos_cond_collect'),
+  });
   const denoise = g.addNode({
     type: 'flux_denoise',
     id: getPrefixedId('flux_denoise'),
@@ -104,13 +108,12 @@ export const buildFLUXGraph = async (
   g.addEdge(modelLoader, 'clip', posCond, 'clip');
   g.addEdge(modelLoader, 't5_encoder', posCond, 't5_encoder');
   g.addEdge(modelLoader, 'max_seq_len', posCond, 't5_max_seq_len');
+  g.addEdge(posCond, 'conditioning', posCondCollect, 'item');
+  g.addEdge(posCondCollect, 'collection', denoise, 'positive_text_conditioning');
+  g.addEdge(denoise, 'latents', l2i, 'latents');
 
   addFLUXLoRAs(state, g, denoise, modelLoader, posCond);
 
-  g.addEdge(posCond, 'conditioning', denoise, 'positive_text_conditioning');
-
-  g.addEdge(denoise, 'latents', l2i, 'latents');
-
   const modelConfig = await fetchModelConfigWithTypeGuard(model.key, isNonRefinerMainModelConfig);
   assert(modelConfig.base === 'flux');
 
@@ -196,31 +199,50 @@ export const buildFLUXGraph = async (
     type: 'collect',
     id: getPrefixedId('control_net_collector'),
   });
-  const controlNetResult = await addControlNets(
+  const controlNetResult = await addControlNets({
     manager,
-    canvas.controlLayers.entities,
+    entities: canvas.controlLayers.entities,
     g,
-    canvas.bbox.rect,
-    controlNetCollector,
-    modelConfig.base
-  );
+    rect: canvas.bbox.rect,
+    collector: controlNetCollector,
+    model: modelConfig,
+  });
   if (controlNetResult.addedControlNets > 0) {
     g.addEdge(controlNetCollector, 'collection', denoise, 'control');
   } else {
     g.deleteNode(controlNetCollector.id);
   }
 
-  const ipAdapterCollector = g.addNode({
+  const ipAdapterCollect = g.addNode({
     type: 'collect',
     id: getPrefixedId('ip_adapter_collector'),
   });
-  const ipAdapterResult = addIPAdapters(canvas.referenceImages.entities, g, ipAdapterCollector, modelConfig.base);
+  const ipAdapterResult = addIPAdapters({
+    entities: canvas.referenceImages.entities,
+    g,
+    collector: ipAdapterCollect,
+    model: modelConfig,
+  });
 
-  const totalIPAdaptersAdded = ipAdapterResult.addedIPAdapters;
+  const regionsResult = await addRegions({
+    manager,
+    regions: canvas.regionalGuidance.entities,
+    g,
+    bbox: canvas.bbox.rect,
+    model: modelConfig,
+    posCond,
+    negCond: null,
+    posCondCollect,
+    negCondCollect: null,
+    ipAdapterCollect,
+  });
+
+  const totalIPAdaptersAdded =
+    ipAdapterResult.addedIPAdapters + regionsResult.reduce((acc, r) => acc + r.addedIPAdapters, 0);
   if (totalIPAdaptersAdded > 0) {
-    g.addEdge(ipAdapterCollector, 'collection', denoise, 'ip_adapter');
+    g.addEdge(ipAdapterCollect, 'collection', denoise, 'ip_adapter');
   } else {
-    g.deleteNode(ipAdapterCollector.id);
+    g.deleteNode(ipAdapterCollect.id);
   }
 
   if (state.system.shouldUseNSFWChecker) {

invokeai/frontend/web/src/features/nodes/util/graph/generation/buildSD1Graph.ts

@@ -227,14 +227,14 @@ export const buildSD1Graph = async (
     type: 'collect',
     id: getPrefixedId('control_net_collector'),
   });
-  const controlNetResult = await addControlNets(
+  const controlNetResult = await addControlNets({
     manager,
-    canvas.controlLayers.entities,
+    entities: canvas.controlLayers.entities,
     g,
-    canvas.bbox.rect,
-    controlNetCollector,
-    modelConfig.base
-  );
+    rect: canvas.bbox.rect,
+    collector: controlNetCollector,
+    model: modelConfig,
+  });
   if (controlNetResult.addedControlNets > 0) {
     g.addEdge(controlNetCollector, 'collection', denoise, 'control');
   } else {
@@ -245,46 +245,50 @@ export const buildSD1Graph = async (
     type: 'collect',
     id: getPrefixedId('t2i_adapter_collector'),
   });
-  const t2iAdapterResult = await addT2IAdapters(
+  const t2iAdapterResult = await addT2IAdapters({
     manager,
-    canvas.controlLayers.entities,
+    entities: canvas.controlLayers.entities,
     g,
-    canvas.bbox.rect,
-    t2iAdapterCollector,
-    modelConfig.base
-  );
+    rect: canvas.bbox.rect,
+    collector: t2iAdapterCollector,
+    model: modelConfig,
+  });
   if (t2iAdapterResult.addedT2IAdapters > 0) {
     g.addEdge(t2iAdapterCollector, 'collection', denoise, 't2i_adapter');
   } else {
     g.deleteNode(t2iAdapterCollector.id);
   }
 
-  const ipAdapterCollector = g.addNode({
+  const ipAdapterCollect = g.addNode({
     type: 'collect',
     id: getPrefixedId('ip_adapter_collector'),
   });
-  const ipAdapterResult = addIPAdapters(canvas.referenceImages.entities, g, ipAdapterCollector, modelConfig.base);
-
-  const regionsResult = await addRegions(
-    manager,
-    canvas.regionalGuidance.entities,
+  const ipAdapterResult = addIPAdapters({
+    entities: canvas.referenceImages.entities,
     g,
-    canvas.bbox.rect,
-    modelConfig.base,
-    denoise,
+    collector: ipAdapterCollect,
+    model: modelConfig,
+  });
+
+  const regionsResult = await addRegions({
+    manager,
+    regions: canvas.regionalGuidance.entities,
+    g,
+    bbox: canvas.bbox.rect,
+    model: modelConfig,
     posCond,
     negCond,
     posCondCollect,
     negCondCollect,
-    ipAdapterCollector
-  );
+    ipAdapterCollect,
+  });
 
   const totalIPAdaptersAdded =
     ipAdapterResult.addedIPAdapters + regionsResult.reduce((acc, r) => acc + r.addedIPAdapters, 0);
   if (totalIPAdaptersAdded > 0) {
-    g.addEdge(ipAdapterCollector, 'collection', denoise, 'ip_adapter');
+    g.addEdge(ipAdapterCollect, 'collection', denoise, 'ip_adapter');
   } else {
-    g.deleteNode(ipAdapterCollector.id);
+    g.deleteNode(ipAdapterCollect.id);
   }
 
   if (state.system.shouldUseNSFWChecker) {

invokeai/frontend/web/src/features/nodes/util/graph/generation/buildSDXLGraph.ts

@@ -232,14 +232,14 @@ export const buildSDXLGraph = async (
     type: 'collect',
     id: getPrefixedId('control_net_collector'),
   });
-  const controlNetResult = await addControlNets(
+  const controlNetResult = await addControlNets({
     manager,
-    canvas.controlLayers.entities,
+    entities: canvas.controlLayers.entities,
     g,
-    canvas.bbox.rect,
-    controlNetCollector,
-    modelConfig.base
-  );
+    rect: canvas.bbox.rect,
+    collector: controlNetCollector,
+    model: modelConfig,
+  });
   if (controlNetResult.addedControlNets > 0) {
     g.addEdge(controlNetCollector, 'collection', denoise, 'control');
   } else {
@@ -250,46 +250,50 @@ export const buildSDXLGraph = async (
     type: 'collect',
     id: getPrefixedId('t2i_adapter_collector'),
   });
-  const t2iAdapterResult = await addT2IAdapters(
+  const t2iAdapterResult = await addT2IAdapters({
     manager,
-    canvas.controlLayers.entities,
+    entities: canvas.controlLayers.entities,
     g,
-    canvas.bbox.rect,
-    t2iAdapterCollector,
-    modelConfig.base
-  );
+    rect: canvas.bbox.rect,
+    collector: t2iAdapterCollector,
+    model: modelConfig,
+  });
   if (t2iAdapterResult.addedT2IAdapters > 0) {
     g.addEdge(t2iAdapterCollector, 'collection', denoise, 't2i_adapter');
   } else {
     g.deleteNode(t2iAdapterCollector.id);
   }
 
-  const ipAdapterCollector = g.addNode({
+  const ipAdapterCollect = g.addNode({
     type: 'collect',
     id: getPrefixedId('ip_adapter_collector'),
   });
-  const ipAdapterResult = addIPAdapters(canvas.referenceImages.entities, g, ipAdapterCollector, modelConfig.base);
-
-  const regionsResult = await addRegions(
-    manager,
-    canvas.regionalGuidance.entities,
+  const ipAdapterResult = addIPAdapters({
+    entities: canvas.referenceImages.entities,
     g,
-    canvas.bbox.rect,
-    modelConfig.base,
-    denoise,
+    collector: ipAdapterCollect,
+    model: modelConfig,
+  });
+
+  const regionsResult = await addRegions({
+    manager,
+    regions: canvas.regionalGuidance.entities,
+    g,
+    bbox: canvas.bbox.rect,
+    model: modelConfig,
     posCond,
     negCond,
     posCondCollect,
     negCondCollect,
-    ipAdapterCollector
-  );
+    ipAdapterCollect,
+  });
 
   const totalIPAdaptersAdded =
     ipAdapterResult.addedIPAdapters + regionsResult.reduce((acc, r) => acc + r.addedIPAdapters, 0);
   if (totalIPAdaptersAdded > 0) {
-    g.addEdge(ipAdapterCollector, 'collection', denoise, 'ip_adapter');
+    g.addEdge(ipAdapterCollect, 'collection', denoise, 'ip_adapter');
   } else {
-    g.deleteNode(ipAdapterCollector.id);
+    g.deleteNode(ipAdapterCollect.id);
   }
 
   if (state.system.shouldUseNSFWChecker) {

invokeai/frontend/web/src/features/nodes/util/schema/buildFieldInputInstance.ts

@@ -28,6 +28,7 @@ const FIELD_VALUE_FALLBACK_MAP: Record<StatefulFieldType['name'], FieldValue> =
   CLIPEmbedModelField: undefined,
   CLIPLEmbedModelField: undefined,
   CLIPGEmbedModelField: undefined,
+  StructuralLoRAModelField: undefined,
 };
 
 export const buildFieldInputInstance = (id: string, template: FieldInputTemplate): FieldInputInstance => {

invokeai/frontend/web/src/features/nodes/util/schema/buildFieldInputTemplate.ts

@@ -28,6 +28,7 @@ import type {
   StatefulFieldType,
   StatelessFieldInputTemplate,
   StringFieldInputTemplate,
+  StructuralLoRAModelFieldInputTemplate,
   T2IAdapterModelFieldInputTemplate,
   T5EncoderModelFieldInputTemplate,
   VAEModelFieldInputTemplate,
@@ -300,6 +301,20 @@ const buildCLIPGEmbedModelFieldInputTemplate: FieldInputTemplateBuilder<CLIPGEmb
   return template;
 };
 
+const buildStructuralLoRAModelFieldInputTemplate: FieldInputTemplateBuilder<StructuralLoRAModelFieldInputTemplate> = ({
+  schemaObject,
+  baseField,
+  fieldType,
+}) => {
+  const template: StructuralLoRAModelFieldInputTemplate = {
+    ...baseField,
+    type: fieldType,
+    default: schemaObject.default ?? undefined,
+  };
+
+  return template;
+};
+
 const buildFluxVAEModelFieldInputTemplate: FieldInputTemplateBuilder<FluxVAEModelFieldInputTemplate> = ({
   schemaObject,
   baseField,
@@ -526,6 +541,7 @@ export const TEMPLATE_BUILDER_MAP: Record<StatefulFieldType['name'], FieldInputT
   CLIPLEmbedModelField: buildCLIPLEmbedModelFieldInputTemplate,
   CLIPGEmbedModelField: buildCLIPGEmbedModelFieldInputTemplate,
   FluxVAEModelField: buildFluxVAEModelFieldInputTemplate,
+  StructuralLoRAModelField: buildStructuralLoRAModelFieldInputTemplate,
 } as const;
 
 export const buildFieldInputTemplate = (

invokeai/frontend/web/src/features/parameters/types/parameterSchemas.ts

@@ -113,6 +113,11 @@ export const zParameterVAEModel = zModelIdentifierField;
 export type ParameterVAEModel = z.infer<typeof zParameterVAEModel>;
 // #endregion
 
+// #region Structural Lora Model
+export const zParameterStructuralLoRAModel = zModelIdentifierField;
+export type ParameterStructuralLoRAModel = z.infer<typeof zParameterStructuralLoRAModel>;
+// #endregion
+
 // #region T5Encoder Model
 export const zParameterT5EncoderModel = zModelIdentifierField;
 export type ParameterT5EncoderModel = z.infer<typeof zParameterT5EncoderModel>;

invokeai/frontend/web/src/features/queue/components/ClearQueueButton.tsx

@@ -4,13 +4,13 @@ import { memo } from 'react';
 import { useTranslation } from 'react-i18next';
 import { PiTrashSimpleFill } from 'react-icons/pi';
 
-import { useClearQueue } from './ClearQueueConfirmationAlertDialog';
+import { useClearQueueDialog } from './ClearQueueConfirmationAlertDialog';
 
 type Props = ButtonProps;
 
 const ClearQueueButton = (props: Props) => {
   const { t } = useTranslation();
-  const clearQueue = useClearQueue();
+  const clearQueue = useClearQueueDialog();
 
   return (
     <>

invokeai/frontend/web/src/features/queue/components/ClearQueueConfirmationAlertDialog.tsx

@@ -1,51 +1,15 @@
 import { ConfirmationAlertDialog, Text } from '@invoke-ai/ui-library';
-import { useStore } from '@nanostores/react';
-import { useAppDispatch } from 'app/store/storeHooks';
 import { useAssertSingleton } from 'common/hooks/useAssertSingleton';
 import { buildUseBoolean } from 'common/hooks/useBoolean';
-import { listCursorChanged, listPriorityChanged } from 'features/queue/store/queueSlice';
-import { toast } from 'features/toast/toast';
-import { memo, useCallback, useMemo } from 'react';
+import { useClearQueue } from 'features/queue/hooks/useClearQueue';
+import { memo } from 'react';
 import { useTranslation } from 'react-i18next';
-import { useClearQueueMutation, useGetQueueStatusQuery } from 'services/api/endpoints/queue';
-import { $isConnected } from 'services/events/stores';
 
 const [useClearQueueConfirmationAlertDialog] = buildUseBoolean(false);
 
-export const useClearQueue = () => {
-  const { t } = useTranslation();
-  const dispatch = useAppDispatch();
+export const useClearQueueDialog = () => {
   const dialog = useClearQueueConfirmationAlertDialog();
-  const { data: queueStatus } = useGetQueueStatusQuery();
-  const isConnected = useStore($isConnected);
-  const [trigger, { isLoading }] = useClearQueueMutation({
-    fixedCacheKey: 'clearQueue',
-  });
-
-  const clearQueue = useCallback(async () => {
-    if (!queueStatus?.queue.total) {
-      return;
-    }
-
-    try {
-      await trigger().unwrap();
-      toast({
-        id: 'QUEUE_CLEAR_SUCCEEDED',
-        title: t('queue.clearSucceeded'),
-        status: 'success',
-      });
-      dispatch(listCursorChanged(undefined));
-      dispatch(listPriorityChanged(undefined));
-    } catch {
-      toast({
-        id: 'QUEUE_CLEAR_FAILED',
-        title: t('queue.clearFailed'),
-        status: 'error',
-      });
-    }
-  }, [queueStatus?.queue.total, trigger, dispatch, t]);
-
-  const isDisabled = useMemo(() => !isConnected || !queueStatus?.queue.total, [isConnected, queueStatus?.queue.total]);
+  const { clearQueue, isLoading, isDisabled, queueStatus } = useClearQueue();
 
   return {
     clearQueue,
@@ -61,7 +25,7 @@ export const useClearQueue = () => {
 export const ClearQueueConfirmationsAlertDialog = memo(() => {
   useAssertSingleton('ClearQueueConfirmationsAlertDialog');
   const { t } = useTranslation();
-  const clearQueue = useClearQueue();
+  const clearQueue = useClearQueueDialog();
 
   return (
     <ConfirmationAlertDialog

invokeai/frontend/web/src/features/queue/components/ClearQueueIconButton.tsx

@@ -4,11 +4,11 @@ import { memo } from 'react';
 import { useTranslation } from 'react-i18next';
 import { PiTrashSimpleBold, PiXBold } from 'react-icons/pi';
 
-import { useClearQueue } from './ClearQueueConfirmationAlertDialog';
+import { useClearQueueDialog } from './ClearQueueConfirmationAlertDialog';
 
 export const ClearQueueIconButton = memo((_) => {
   const { t } = useTranslation();
-  const clearQueue = useClearQueue();
+  const clearQueue = useClearQueueDialog();
   const cancelCurrentQueueItem = useCancelCurrentQueueItem();
 
   // Show the single item clear button when shift is pressed

invokeai/frontend/web/src/features/queue/components/InvokeButtonTooltip/InvokeButtonTooltip.tsx

@@ -147,8 +147,6 @@ const UpscaleTabTooltipContent = memo(({ prepend = false }: { prepend?: boolean
           <ReasonsList reasons={reasons} />
         </>
       )}
-      <StyledDivider />
-      <AddingToText />
     </Flex>
   );
 });
@@ -180,8 +178,6 @@ const WorkflowsTabTooltipContent = memo(({ prepend = false }: { prepend?: boolea
           <ReasonsList reasons={reasons} />
         </>
       )}
-      <StyledDivider />
-      <AddingToText />
     </Flex>
   );
 });

invokeai/frontend/web/src/features/queue/components/QueueActionsMenuButton.tsx

@@ -1,8 +1,9 @@
 import { IconButton, Menu, MenuButton, MenuGroup, MenuItem, MenuList } from '@invoke-ai/ui-library';
 import { useAppDispatch } from 'app/store/storeHooks';
 import { SessionMenuItems } from 'common/components/SessionMenuItems';
-import { useClearQueue } from 'features/queue/components/ClearQueueConfirmationAlertDialog';
+import { useClearQueueDialog } from 'features/queue/components/ClearQueueConfirmationAlertDialog';
 import { QueueCountBadge } from 'features/queue/components/QueueCountBadge';
+import { useCancelCurrentQueueItem } from 'features/queue/hooks/useCancelCurrentQueueItem';
 import { usePauseProcessor } from 'features/queue/hooks/usePauseProcessor';
 import { useResumeProcessor } from 'features/queue/hooks/useResumeProcessor';
 import { useFeatureStatus } from 'features/system/hooks/useFeatureStatus';
@@ -17,7 +18,8 @@ export const QueueActionsMenuButton = memo(() => {
   const { t } = useTranslation();
   const isPauseEnabled = useFeatureStatus('pauseQueue');
   const isResumeEnabled = useFeatureStatus('resumeQueue');
-  const clearQueue = useClearQueue();
+  const cancelCurrent = useCancelCurrentQueueItem();
+  const clearQueue = useClearQueueDialog();
   const {
     resumeProcessor,
     isLoading: isLoadingResumeProcessor,
@@ -44,9 +46,9 @@ export const QueueActionsMenuButton = memo(() => {
             <MenuItem
               isDestructive
               icon={<PiXBold />}
-              onClick={clearQueue.openDialog}
-              isLoading={clearQueue.isLoading}
-              isDisabled={clearQueue.isDisabled}
+              onClick={cancelCurrent.cancelQueueItem}
+              isLoading={cancelCurrent.isLoading}
+              isDisabled={cancelCurrent.isDisabled}
             >
               {t('queue.cancelTooltip')}
             </MenuItem>

invokeai/frontend/web/src/features/queue/hooks/useClearQueue.ts

@@ -0,0 +1,50 @@
+import { useStore } from '@nanostores/react';
+import { useAppDispatch } from 'app/store/storeHooks';
+import { listCursorChanged, listPriorityChanged } from 'features/queue/store/queueSlice';
+import { toast } from 'features/toast/toast';
+import { useCallback, useMemo } from 'react';
+import { useTranslation } from 'react-i18next';
+import { useClearQueueMutation, useGetQueueStatusQuery } from 'services/api/endpoints/queue';
+import { $isConnected } from 'services/events/stores';
+
+export const useClearQueue = () => {
+  const { t } = useTranslation();
+  const dispatch = useAppDispatch();
+  const { data: queueStatus } = useGetQueueStatusQuery();
+  const isConnected = useStore($isConnected);
+  const [trigger, { isLoading }] = useClearQueueMutation({
+    fixedCacheKey: 'clearQueue',
+  });
+
+  const clearQueue = useCallback(async () => {
+    if (!queueStatus?.queue.total) {
+      return;
+    }
+
+    try {
+      await trigger().unwrap();
+      toast({
+        id: 'QUEUE_CLEAR_SUCCEEDED',
+        title: t('queue.clearSucceeded'),
+        status: 'success',
+      });
+      dispatch(listCursorChanged(undefined));
+      dispatch(listPriorityChanged(undefined));
+    } catch {
+      toast({
+        id: 'QUEUE_CLEAR_FAILED',
+        title: t('queue.clearFailed'),
+        status: 'error',
+      });
+    }
+  }, [queueStatus?.queue.total, trigger, dispatch, t]);
+
+  const isDisabled = useMemo(() => !isConnected || !queueStatus?.queue.total, [isConnected, queueStatus?.queue.total]);
+
+  return {
+    clearQueue,
+    isLoading,
+    queueStatus,
+    isDisabled,
+  };
+};