make session_list api return a raw dict rather than pydantic object

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

- [ ] Refactor
- [ ] Feature
- [x] Bug Fix
- [ ] Optimization
- [ ] Documentation Update
- [ ] Community Node Submission


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

      
## Have you updated all relevant documentation?
- [ ] Yes
- [ ] No


## Description


## Related Tickets & Documents

<!--
For pull requests that relate or close an issue, please include them
below. 

For example having the text: "closes #1234" would connect the current
pull
request to issue 1234.  And when we merge the pull request, Github will
automatically close the issue.
-->

- Related Issue #
- Closes #

## QA Instructions, Screenshots, Recordings

<!-- 
Please provide steps on how to test changes, any hardware or 
software specifications as well as any other pertinent information. 
-->

## Added/updated tests?

- [ ] Yes
- [ ] No : _please replace this line with details on why tests
      have not been included_

## [optional] Are there any post deployment tasks we need to perform?
Should be removed when added in diffusers
https://github.com/huggingface/diffusers/pull/4599

docker/docker-entrypoint.sh

@@ -29,8 +29,8 @@ configure() {
         echo "To reconfigure InvokeAI, delete the above file."
         echo "======================================================================"
     else
-        mkdir -p ${INVOKEAI_ROOT}
-        chown --recursive ${USER} ${INVOKEAI_ROOT}
+        mkdir -p "${INVOKEAI_ROOT}"
+        chown --recursive ${USER} "${INVOKEAI_ROOT}"
         gosu ${USER} invokeai-configure --yes --default_only
     fi
 }
@@ -50,16 +50,16 @@ fi
 if [[ -v "PUBLIC_KEY" ]] && [[ ! -d "${HOME}/.ssh" ]]; then
     apt-get update
     apt-get install -y openssh-server
-    pushd $HOME
+    pushd "$HOME"
     mkdir -p .ssh
-    echo ${PUBLIC_KEY} > .ssh/authorized_keys
+    echo "${PUBLIC_KEY}" > .ssh/authorized_keys
     chmod -R 700 .ssh
     popd
     service ssh start
 fi
 
 
-cd ${INVOKEAI_ROOT}
+cd "${INVOKEAI_ROOT}"
 
 # Run the CMD as the Container User (not root).
 exec gosu ${USER} "$@"

docs/features/POSTPROCESS.md

@@ -4,35 +4,13 @@ title: Postprocessing
 
 # :material-image-edit: Postprocessing
 
-## Intro
-
-This extension provides the ability to restore faces and upscale images.
+This sections details the ability to improve faces and upscale images.
 
 ## Face Fixing
 
-The default face restoration module is GFPGAN. The default upscale is
-Real-ESRGAN. For an alternative face restoration module, see
-[CodeFormer Support](#codeformer-support) below.
+As of InvokeAI 3.0, the easiest way to improve faces created during image generation is through the Inpainting functionality of the Unified Canvas. Simply add the image containing the faces that you would like to improve to the canvas, mask the face to be improved and run the invocation. For best results, make sure to use an inpainting specific model; these are usually identified by the "-inpainting" term in the model name. 
 
-As of version 1.14, environment.yaml will install the Real-ESRGAN package into
-the standard install location for python packages, and will put GFPGAN into a
-subdirectory of "src" in the InvokeAI directory. Upscaling with Real-ESRGAN
-should "just work" without further intervention. Simply indicate the desired scale on
-the popup in the Web GUI.
-
-**GFPGAN** requires a series of downloadable model files to work. These are
-loaded when you run `invokeai-configure`. If GFPAN is failing with an
-error, please run the following from the InvokeAI directory:
-
-```bash
-invokeai-configure
-```
-
-If you do not run this script in advance, the GFPGAN module will attempt to
-download the models files the first time you try to perform facial
-reconstruction.
-
-### Upscaling
+## Upscaling
 
 Open the upscaling dialog by clicking on the "expand" icon located
 above the image display area in the Web UI:
@@ -41,82 +19,23 @@ above the image display area in the Web UI:
 ![upscale1](../assets/features/upscale-dialog.png)
 </figure>
 
-There are three different upscaling parameters that you can
-adjust. The first is the scale itself, either 2x or 4x.
+The default upscaling option is Real-ESRGAN x2 Plus, which will scale your image by a factor of two. This means upscaling a 512x512 image will result in a new 1024x1024 image.
 
-The second is the "Denoising Strength." Higher values will smooth out
-the image and remove digital chatter, but may lose fine detail at
-higher values.
+Other options are the x4 upscalers, which will scale your image by a factor of 4. 
 
-Third, "Upscale Strength" allows you to adjust how the You can set the
-scaling stength between `0` and `1.0` to control the intensity of the
-scaling. AI upscalers generally tend to smooth out texture details. If
-you wish to retain some of those for natural looking results, we
-recommend using values between `0.5 to 0.8`.
-
-[This figure](../assets/features/upscaling-montage.png) illustrates
-the effects of denoising and strength. The original image was 512x512,
-4x scaled to 2048x2048. The "original" version on the upper left was
-scaled using simple pixel averaging. The remainder use the ESRGAN
-upscaling algorithm at different levels of denoising and strength.
-
-<figure markdown>
-![upscaling](../assets/features/upscaling-montage.png){ width=720 }
-</figure>
-
-Both denoising and strength default to 0.75.
-
-### Face Restoration
-
-InvokeAI offers alternative two face restoration algorithms,
-[GFPGAN](https://github.com/TencentARC/GFPGAN) and
-[CodeFormer](https://huggingface.co/spaces/sczhou/CodeFormer). These
-algorithms improve the appearance of faces, particularly eyes and
-mouths. Issues with faces are less common with the latest set of
-Stable Diffusion models than with the original 1.4 release, but the
-restoration algorithms can still make a noticeable improvement in
-certain cases. You can also apply restoration to old photographs you
-upload.
-
-To access face restoration, click the "smiley face" icon in the
-toolbar above the InvokeAI image panel. You will be presented with a
-dialog that offers a choice between the two algorithm and sliders that
-allow you to adjust their parameters. Alternatively, you may open the
-left-hand accordion panel labeled "Face Restoration" and have the
-restoration algorithm of your choice applied to generated images
-automatically.
-
-
-Like upscaling, there are a number of parameters that adjust the face
-restoration output. GFPGAN has a single parameter, `strength`, which
-controls how much the algorithm is allowed to adjust the
-image. CodeFormer has two parameters, `strength`, and `fidelity`,
-which together control the quality of the output image as described in
-the [CodeFormer project
-page](https://shangchenzhou.com/projects/CodeFormer/). Default values
-are 0.75 for both parameters, which achieves a reasonable balance
-between changing the image too much and not enough.
-
-[This figure](../assets/features/restoration-montage.png) illustrates
-the effects of adjusting GFPGAN and CodeFormer parameters.
-
-<figure markdown>
-![upscaling](../assets/features/restoration-montage.png){ width=720 }
-</figure>
 
 !!! note
 
-    GFPGAN and Real-ESRGAN are both memory intensive. In order to avoid crashes and memory overloads
+    Real-ESRGAN is memory intensive. In order to avoid crashes and memory overloads
     during the Stable Diffusion process, these effects are applied after Stable Diffusion has completed
     its work.
 
     In single image generations, you will see the output right away but when you are using multiple
-    iterations, the images will first be generated and then upscaled and face restored after that
+    iterations, the images will first be generated and then upscaled after that
     process is complete. While the image generation is taking place, you will still be able to preview
     the base images.
 
 ## How to disable
 
-If, for some reason, you do not wish to load the GFPGAN and/or ESRGAN libraries,
-you can disable them on the invoke.py command line with the `--no_restore` and
-`--no_esrgan` options, respectively.
+If, for some reason, you do not wish to load the ESRGAN libraries,
+you can disable them on the invoke.py command line with the `--no_esrgan` options.

docs/installation/INSTALLATION.md

@@ -25,10 +25,10 @@ This method is recommended for experienced users and developers
 #### [Docker Installation](040_INSTALL_DOCKER.md)
 This method is recommended for those familiar with running Docker containers
 ### Other Installation Guides
-  - [PyPatchMatch](installation/060_INSTALL_PATCHMATCH.md)
-  - [XFormers](installation/070_INSTALL_XFORMERS.md)
-  - [CUDA and ROCm Drivers](installation/030_INSTALL_CUDA_AND_ROCM.md)
-  - [Installing New Models](installation/050_INSTALLING_MODELS.md)
+  - [PyPatchMatch](060_INSTALL_PATCHMATCH.md)
+  - [XFormers](070_INSTALL_XFORMERS.md)
+  - [CUDA and ROCm Drivers](030_INSTALL_CUDA_AND_ROCM.md)
+  - [Installing New Models](050_INSTALLING_MODELS.md)
 
 ## :fontawesome-solid-computer: Hardware Requirements
 

invokeai/app/api/routers/images.py

@@ -5,7 +5,7 @@ from PIL import Image
 from fastapi import Body, HTTPException, Path, Query, Request, Response, UploadFile
 from fastapi.responses import FileResponse
 from fastapi.routing import APIRouter
-from pydantic import BaseModel
+from pydantic import BaseModel, Field
 
 from invokeai.app.invocations.metadata import ImageMetadata
 from invokeai.app.models.image import ImageCategory, ResourceOrigin
@@ -19,6 +19,7 @@ from ..dependencies import ApiDependencies
 
 images_router = APIRouter(prefix="/v1/images", tags=["images"])
 
+
 # images are immutable; set a high max-age
 IMAGE_MAX_AGE = 31536000
 
@@ -286,3 +287,41 @@ async def delete_images_from_list(
         return DeleteImagesFromListResult(deleted_images=deleted_images)
     except Exception as e:
         raise HTTPException(status_code=500, detail="Failed to delete images")
+
+
+class ImagesUpdatedFromListResult(BaseModel):
+    updated_image_names: list[str] = Field(description="The image names that were updated")
+
+
+@images_router.post("/star", operation_id="star_images_in_list", response_model=ImagesUpdatedFromListResult)
+async def star_images_in_list(
+    image_names: list[str] = Body(description="The list of names of images to star", embed=True),
+) -> ImagesUpdatedFromListResult:
+    try:
+        updated_image_names: list[str] = []
+        for image_name in image_names:
+            try:
+                ApiDependencies.invoker.services.images.update(image_name, changes=ImageRecordChanges(starred=True))
+                updated_image_names.append(image_name)
+            except:
+                pass
+        return ImagesUpdatedFromListResult(updated_image_names=updated_image_names)
+    except Exception as e:
+        raise HTTPException(status_code=500, detail="Failed to star images")
+
+
+@images_router.post("/unstar", operation_id="unstar_images_in_list", response_model=ImagesUpdatedFromListResult)
+async def unstar_images_in_list(
+    image_names: list[str] = Body(description="The list of names of images to unstar", embed=True),
+) -> ImagesUpdatedFromListResult:
+    try:
+        updated_image_names: list[str] = []
+        for image_name in image_names:
+            try:
+                ApiDependencies.invoker.services.images.update(image_name, changes=ImageRecordChanges(starred=False))
+                updated_image_names.append(image_name)
+            except:
+                pass
+        return ImagesUpdatedFromListResult(updated_image_names=updated_image_names)
+    except Exception as e:
+        raise HTTPException(status_code=500, detail="Failed to unstar images")

invokeai/app/api/routers/sessions.py

@@ -40,7 +40,7 @@ async def create_session(
 @session_router.get(
     "/",
     operation_id="list_sessions",
-    responses={200: {"model": PaginatedResults[GraphExecutionState]}},
+    responses={200: {"model": PaginatedResults[dict]}},
 )
 async def list_sessions(
     page: int = Query(default=0, description="The page of results to get"),

invokeai/app/api_app.py

@@ -38,7 +38,7 @@ import mimetypes
 from .api.dependencies import ApiDependencies
 from .api.routers import sessions, models, images, boards, board_images, app_info
 from .api.sockets import SocketIO
-from .invocations.baseinvocation import BaseInvocation
+from .invocations.baseinvocation import BaseInvocation, _InputField, _OutputField, UIConfigBase
 
 
 import torch
@@ -134,6 +134,11 @@ def custom_openapi():
         # This could break in some cases, figure out a better way to do it
         output_type_titles[schema_key] = output_schema["title"]
 
+    # Add Node Editor UI helper schemas
+    ui_config_schemas = schema([UIConfigBase, _InputField, _OutputField], ref_prefix="#/components/schemas/")
+    for schema_key, output_schema in ui_config_schemas["definitions"].items():
+        openapi_schema["components"]["schemas"][schema_key] = output_schema
+
     # Add a reference to the output type to additionalProperties of the invoker schema
     for invoker in all_invocations:
         invoker_name = invoker.__name__

invokeai/app/invocations/baseinvocation.py

@@ -3,15 +3,366 @@
 from __future__ import annotations
 
 from abc import ABC, abstractmethod
+from enum import Enum
 from inspect import signature
-from typing import TYPE_CHECKING, Dict, List, Literal, TypedDict, get_args, get_type_hints
+from typing import (
+    TYPE_CHECKING,
+    AbstractSet,
+    Any,
+    Callable,
+    ClassVar,
+    Mapping,
+    Optional,
+    Type,
+    TypeVar,
+    Union,
+    get_args,
+    get_type_hints,
+)
 
-from pydantic import BaseConfig, BaseModel, Field
+from pydantic import BaseModel, Field
+from pydantic.fields import Undefined
+from pydantic.typing import NoArgAnyCallable
 
 if TYPE_CHECKING:
     from ..services.invocation_services import InvocationServices
 
 
+class FieldDescriptions:
+    denoising_start = "When to start denoising, expressed a percentage of total steps"
+    denoising_end = "When to stop denoising, expressed a percentage of total steps"
+    cfg_scale = "Classifier-Free Guidance scale"
+    scheduler = "Scheduler to use during inference"
+    positive_cond = "Positive conditioning tensor"
+    negative_cond = "Negative conditioning tensor"
+    noise = "Noise tensor"
+    clip = "CLIP (tokenizer, text encoder, LoRAs) and skipped layer count"
+    unet = "UNet (scheduler, LoRAs)"
+    vae = "VAE"
+    cond = "Conditioning tensor"
+    controlnet_model = "ControlNet model to load"
+    vae_model = "VAE model to load"
+    lora_model = "LoRA model to load"
+    main_model = "Main model (UNet, VAE, CLIP) to load"
+    sdxl_main_model = "SDXL Main model (UNet, VAE, CLIP1, CLIP2) to load"
+    sdxl_refiner_model = "SDXL Refiner Main Modde (UNet, VAE, CLIP2) to load"
+    onnx_main_model = "ONNX Main model (UNet, VAE, CLIP) to load"
+    lora_weight = "The weight at which the LoRA is applied to each model"
+    compel_prompt = "Prompt to be parsed by Compel to create a conditioning tensor"
+    raw_prompt = "Raw prompt text (no parsing)"
+    sdxl_aesthetic = "The aesthetic score to apply to the conditioning tensor"
+    skipped_layers = "Number of layers to skip in text encoder"
+    seed = "Seed for random number generation"
+    steps = "Number of steps to run"
+    width = "Width of output (px)"
+    height = "Height of output (px)"
+    control = "ControlNet(s) to apply"
+    denoised_latents = "Denoised latents tensor"
+    latents = "Latents tensor"
+    strength = "Strength of denoising (proportional to steps)"
+    core_metadata = "Optional core metadata to be written to image"
+    interp_mode = "Interpolation mode"
+    torch_antialias = "Whether or not to apply antialiasing (bilinear or bicubic only)"
+    fp32 = "Whether or not to use full float32 precision"
+    precision = "Precision to use"
+    tiled = "Processing using overlapping tiles (reduce memory consumption)"
+    detect_res = "Pixel resolution for detection"
+    image_res = "Pixel resolution for output image"
+    safe_mode = "Whether or not to use safe mode"
+    scribble_mode = "Whether or not to use scribble mode"
+    scale_factor = "The factor by which to scale"
+    num_1 = "The first number"
+    num_2 = "The second number"
+    mask = "The mask to use for the operation"
+
+
+class Input(str, Enum):
+    """
+    The type of input a field accepts.
+    - `Input.Direct`: The field must have its value provided directly, when the invocation and field \
+      are instantiated.
+    - `Input.Connection`: The field must have its value provided by a connection.
+    - `Input.Any`: The field may have its value provided either directly or by a connection.
+    """
+
+    Connection = "connection"
+    Direct = "direct"
+    Any = "any"
+
+
+class UIType(str, Enum):
+    """
+    Type hints for the UI.
+    If a field should be provided a data type that does not exactly match the python type of the field, \
+    use this to provide the type that should be used instead. See the node development docs for detail \
+    on adding a new field type, which involves client-side changes.
+    """
+
+    # region Primitives
+    Integer = "integer"
+    Float = "float"
+    Boolean = "boolean"
+    String = "string"
+    Array = "array"
+    Image = "ImageField"
+    Latents = "LatentsField"
+    Conditioning = "ConditioningField"
+    Control = "ControlField"
+    Color = "ColorField"
+    ImageCollection = "ImageCollection"
+    ConditioningCollection = "ConditioningCollection"
+    ColorCollection = "ColorCollection"
+    LatentsCollection = "LatentsCollection"
+    IntegerCollection = "IntegerCollection"
+    FloatCollection = "FloatCollection"
+    StringCollection = "StringCollection"
+    BooleanCollection = "BooleanCollection"
+    # endregion
+
+    # region Models
+    MainModel = "MainModelField"
+    SDXLMainModel = "SDXLMainModelField"
+    SDXLRefinerModel = "SDXLRefinerModelField"
+    ONNXModel = "ONNXModelField"
+    VaeModel = "VaeModelField"
+    LoRAModel = "LoRAModelField"
+    ControlNetModel = "ControlNetModelField"
+    UNet = "UNetField"
+    Vae = "VaeField"
+    CLIP = "ClipField"
+    # endregion
+
+    # region Iterate/Collect
+    Collection = "Collection"
+    CollectionItem = "CollectionItem"
+    # endregion
+
+    # region Misc
+    FilePath = "FilePath"
+    Enum = "enum"
+    # endregion
+
+
+class UIComponent(str, Enum):
+    """
+    The type of UI component to use for a field, used to override the default components, which are \
+    inferred from the field type.
+    """
+
+    None_ = "none"
+    Textarea = "textarea"
+    Slider = "slider"
+
+
+class _InputField(BaseModel):
+    """
+    *DO NOT USE*
+    This helper class is used to tell the client about our custom field attributes via OpenAPI
+    schema generation, and Typescript type generation from that schema. It serves no functional
+    purpose in the backend.
+    """
+
+    input: Input
+    ui_hidden: bool
+    ui_type: Optional[UIType]
+    ui_component: Optional[UIComponent]
+
+
+class _OutputField(BaseModel):
+    """
+    *DO NOT USE*
+    This helper class is used to tell the client about our custom field attributes via OpenAPI
+    schema generation, and Typescript type generation from that schema. It serves no functional
+    purpose in the backend.
+    """
+
+    ui_hidden: bool
+    ui_type: Optional[UIType]
+
+
+def InputField(
+    *args: Any,
+    default: Any = Undefined,
+    default_factory: Optional[NoArgAnyCallable] = None,
+    alias: Optional[str] = None,
+    title: Optional[str] = None,
+    description: Optional[str] = None,
+    exclude: Optional[Union[AbstractSet[Union[int, str]], Mapping[Union[int, str], Any], Any]] = None,
+    include: Optional[Union[AbstractSet[Union[int, str]], Mapping[Union[int, str], Any], Any]] = None,
+    const: Optional[bool] = None,
+    gt: Optional[float] = None,
+    ge: Optional[float] = None,
+    lt: Optional[float] = None,
+    le: Optional[float] = None,
+    multiple_of: Optional[float] = None,
+    allow_inf_nan: Optional[bool] = None,
+    max_digits: Optional[int] = None,
+    decimal_places: Optional[int] = None,
+    min_items: Optional[int] = None,
+    max_items: Optional[int] = None,
+    unique_items: Optional[bool] = None,
+    min_length: Optional[int] = None,
+    max_length: Optional[int] = None,
+    allow_mutation: bool = True,
+    regex: Optional[str] = None,
+    discriminator: Optional[str] = None,
+    repr: bool = True,
+    input: Input = Input.Any,
+    ui_type: Optional[UIType] = None,
+    ui_component: Optional[UIComponent] = None,
+    ui_hidden: bool = False,
+    **kwargs: Any,
+) -> Any:
+    """
+    Creates an input field for an invocation.
+
+    This is a wrapper for Pydantic's [Field](https://docs.pydantic.dev/1.10/usage/schema/#field-customization) \
+    that adds a few extra parameters to support graph execution and the node editor UI.
+
+    :param Input input: [Input.Any] The kind of input this field requires. \
+      `Input.Direct` means a value must be provided on instantiation. \
+      `Input.Connection` means the value must be provided by a connection. \
+      `Input.Any` means either will do.
+
+    :param UIType ui_type: [None] Optionally provides an extra type hint for the UI. \
+      In some situations, the field's type is not enough to infer the correct UI type. \
+      For example, model selection fields should render a dropdown UI component to select a model. \
+      Internally, there is no difference between SD-1, SD-2 and SDXL model fields, they all use \
+      `MainModelField`. So to ensure the base-model-specific UI is rendered, you can use \
+      `UIType.SDXLMainModelField` to indicate that the field is an SDXL main model field.
+      
+    :param UIComponent ui_component: [None] Optionally specifies a specific component to use in the UI. \
+      The UI will always render a suitable component, but sometimes you want something different than the default. \
+      For example, a `string` field will default to a single-line input, but you may want a multi-line textarea instead. \
+      For this case, you could provide `UIComponent.Textarea`.
+
+    : param bool ui_hidden: [False] Specifies whether or not this field should be hidden in the UI.
+    """
+    return Field(
+        *args,
+        default=default,
+        default_factory=default_factory,
+        alias=alias,
+        title=title,
+        description=description,
+        exclude=exclude,
+        include=include,
+        const=const,
+        gt=gt,
+        ge=ge,
+        lt=lt,
+        le=le,
+        multiple_of=multiple_of,
+        allow_inf_nan=allow_inf_nan,
+        max_digits=max_digits,
+        decimal_places=decimal_places,
+        min_items=min_items,
+        max_items=max_items,
+        unique_items=unique_items,
+        min_length=min_length,
+        max_length=max_length,
+        allow_mutation=allow_mutation,
+        regex=regex,
+        discriminator=discriminator,
+        repr=repr,
+        input=input,
+        ui_type=ui_type,
+        ui_component=ui_component,
+        ui_hidden=ui_hidden,
+        **kwargs,
+    )
+
+
+def OutputField(
+    *args: Any,
+    default: Any = Undefined,
+    default_factory: Optional[NoArgAnyCallable] = None,
+    alias: Optional[str] = None,
+    title: Optional[str] = None,
+    description: Optional[str] = None,
+    exclude: Optional[Union[AbstractSet[Union[int, str]], Mapping[Union[int, str], Any], Any]] = None,
+    include: Optional[Union[AbstractSet[Union[int, str]], Mapping[Union[int, str], Any], Any]] = None,
+    const: Optional[bool] = None,
+    gt: Optional[float] = None,
+    ge: Optional[float] = None,
+    lt: Optional[float] = None,
+    le: Optional[float] = None,
+    multiple_of: Optional[float] = None,
+    allow_inf_nan: Optional[bool] = None,
+    max_digits: Optional[int] = None,
+    decimal_places: Optional[int] = None,
+    min_items: Optional[int] = None,
+    max_items: Optional[int] = None,
+    unique_items: Optional[bool] = None,
+    min_length: Optional[int] = None,
+    max_length: Optional[int] = None,
+    allow_mutation: bool = True,
+    regex: Optional[str] = None,
+    discriminator: Optional[str] = None,
+    repr: bool = True,
+    ui_type: Optional[UIType] = None,
+    ui_hidden: bool = False,
+    **kwargs: Any,
+) -> Any:
+    """
+    Creates an output field for an invocation output.
+
+    This is a wrapper for Pydantic's [Field](https://docs.pydantic.dev/1.10/usage/schema/#field-customization) \
+    that adds a few extra parameters to support graph execution and the node editor UI.
+
+    :param UIType ui_type: [None] Optionally provides an extra type hint for the UI. \
+      In some situations, the field's type is not enough to infer the correct UI type. \
+      For example, model selection fields should render a dropdown UI component to select a model. \
+      Internally, there is no difference between SD-1, SD-2 and SDXL model fields, they all use \
+      `MainModelField`. So to ensure the base-model-specific UI is rendered, you can use \
+      `UIType.SDXLMainModelField` to indicate that the field is an SDXL main model field.
+
+    : param bool ui_hidden: [False] Specifies whether or not this field should be hidden in the UI. \
+    """
+    return Field(
+        *args,
+        default=default,
+        default_factory=default_factory,
+        alias=alias,
+        title=title,
+        description=description,
+        exclude=exclude,
+        include=include,
+        const=const,
+        gt=gt,
+        ge=ge,
+        lt=lt,
+        le=le,
+        multiple_of=multiple_of,
+        allow_inf_nan=allow_inf_nan,
+        max_digits=max_digits,
+        decimal_places=decimal_places,
+        min_items=min_items,
+        max_items=max_items,
+        unique_items=unique_items,
+        min_length=min_length,
+        max_length=max_length,
+        allow_mutation=allow_mutation,
+        regex=regex,
+        discriminator=discriminator,
+        repr=repr,
+        ui_type=ui_type,
+        ui_hidden=ui_hidden,
+        **kwargs,
+    )
+
+
+class UIConfigBase(BaseModel):
+    """
+    Provides additional node configuration to the UI.
+    This is used internally by the @tags and @title decorator logic. You probably want to use those
+    decorators, though you may add this class to a node definition to specify the title and tags.
+    """
+
+    tags: Optional[list[str]] = Field(default_factory=None, description="The tags to display in the UI")
+    title: Optional[str] = Field(default=None, description="The display name of the node")
+
+
 class InvocationContext:
     services: InvocationServices
     graph_execution_state_id: str
@@ -39,6 +390,20 @@ class BaseInvocationOutput(BaseModel):
         return tuple(subclasses)
 
 
+class RequiredConnectionException(Exception):
+    """Raised when an field which requires a connection did not receive a value."""
+
+    def __init__(self, node_id: str, field_name: str):
+        super().__init__(f"Node {node_id} missing connections for field {field_name}")
+
+
+class MissingInputException(Exception):
+    """Raised when an field which requires some input, but did not receive a value."""
+
+    def __init__(self, node_id: str, field_name: str):
+        super().__init__(f"Node {node_id} missing value or connection for field {field_name}")
+
+
 class BaseInvocation(ABC, BaseModel):
     """A node to process inputs and produce outputs.
     May use dependency injection in __init__ to receive providers.
@@ -76,70 +441,81 @@ class BaseInvocation(ABC, BaseModel):
     def get_output_type(cls):
         return signature(cls.invoke).return_annotation
 
+    class Config:
+        @staticmethod
+        def schema_extra(schema: dict[str, Any], model_class: Type[BaseModel]) -> None:
+            uiconfig = getattr(model_class, "UIConfig", None)
+            if uiconfig and hasattr(uiconfig, "title"):
+                schema["title"] = uiconfig.title
+            if uiconfig and hasattr(uiconfig, "tags"):
+                schema["tags"] = uiconfig.tags
+
     @abstractmethod
     def invoke(self, context: InvocationContext) -> BaseInvocationOutput:
         """Invoke with provided context and return outputs."""
         pass
 
-    # fmt: off
-    id: str = Field(description="The id of this node. Must be unique among all nodes.")
-    is_intermediate: bool = Field(default=False, description="Whether or not this node is an intermediate node.")
-    # fmt: on
+    def __init__(self, **data):
+        # nodes may have required fields, that can accept input from connections
+        # on instantiation of the model, we need to exclude these from validation
+        restore = dict()
+        try:
+            field_names = list(self.__fields__.keys())
+            for field_name in field_names:
+                # if the field is required and may get its value from a connection, exclude it from validation
+                field = self.__fields__[field_name]
+                _input = field.field_info.extra.get("input", None)
+                if _input in [Input.Connection, Input.Any] and field.required:
+                    if field_name not in data:
+                        restore[field_name] = self.__fields__.pop(field_name)
+            # instantiate the node, which will validate the data
+            super().__init__(**data)
+        finally:
+            # restore the removed fields
+            for field_name, field in restore.items():
+                self.__fields__[field_name] = field
+
+    def invoke_internal(self, context: InvocationContext) -> BaseInvocationOutput:
+        for field_name, field in self.__fields__.items():
+            _input = field.field_info.extra.get("input", None)
+            if field.required and not hasattr(self, field_name):
+                if _input == Input.Connection:
+                    raise RequiredConnectionException(self.__fields__["type"].default, field_name)
+                elif _input == Input.Any:
+                    raise MissingInputException(self.__fields__["type"].default, field_name)
+        return self.invoke(context)
+
+    id: str = InputField(description="The id of this node. Must be unique among all nodes.")
+    is_intermediate: bool = InputField(
+        default=False, description="Whether or not this node is an intermediate node.", input=Input.Direct
+    )
+    UIConfig: ClassVar[Type[UIConfigBase]]
 
 
-# TODO: figure out a better way to provide these hints
-# TODO: when we can upgrade to python 3.11, we can use the`NotRequired` type instead of `total=False`
-class UIConfig(TypedDict, total=False):
-    type_hints: Dict[
-        str,
-        Literal[
-            "integer",
-            "float",
-            "boolean",
-            "string",
-            "enum",
-            "image",
-            "latents",
-            "model",
-            "control",
-            "image_collection",
-            "vae_model",
-            "lora_model",
-        ],
-    ]
-    tags: List[str]
-    title: str
+T = TypeVar("T", bound=BaseInvocation)
 
 
-class CustomisedSchemaExtra(TypedDict):
-    ui: UIConfig
+def title(title: str) -> Callable[[Type[T]], Type[T]]:
+    """Adds a title to the invocation. Use this to override the default title generation, which is based on the class name."""
+
+    def wrapper(cls: Type[T]) -> Type[T]:
+        uiconf_name = cls.__qualname__ + ".UIConfig"
+        if not hasattr(cls, "UIConfig") or cls.UIConfig.__qualname__ != uiconf_name:
+            cls.UIConfig = type(uiconf_name, (UIConfigBase,), dict())
+        cls.UIConfig.title = title
+        return cls
+
+    return wrapper
 
 
-class InvocationConfig(BaseConfig):
-    """Customizes pydantic's BaseModel.Config class for use by Invocations.
+def tags(*tags: str) -> Callable[[Type[T]], Type[T]]:
+    """Adds tags to the invocation. Use this to improve the streamline finding the invocation in the UI."""
 
-    Provide `schema_extra` a `ui` dict to add hints for generated UIs.
+    def wrapper(cls: Type[T]) -> Type[T]:
+        uiconf_name = cls.__qualname__ + ".UIConfig"
+        if not hasattr(cls, "UIConfig") or cls.UIConfig.__qualname__ != uiconf_name:
+            cls.UIConfig = type(uiconf_name, (UIConfigBase,), dict())
+        cls.UIConfig.tags = list(tags)
+        return cls
 
-    `tags`
-    - A list of strings, used to categorise invocations.
-
-    `type_hints`
-    - A dict of field types which override the types in the invocation definition.
-    - Each key should be the name of one of the invocation's fields.
-    - Each value should be one of the valid types:
-      - `integer`, `float`, `boolean`, `string`, `enum`, `image`, `latents`, `model`
-
-    ```python
-    class Config(InvocationConfig):
-      schema_extra = {
-          "ui": {
-              "tags": ["stable-diffusion", "image"],
-              "type_hints": {
-                  "initial_image": "image",
-              },
-          },
-      }
-    ```
-    """
-
-    schema_extra: CustomisedSchemaExtra
+    return wrapper

invokeai/app/invocations/collections.py

@@ -3,58 +3,25 @@
 from typing import Literal
 
 import numpy as np
-from pydantic import Field, validator
+from pydantic import validator
 
-from invokeai.app.models.image import ImageField
+from invokeai.app.invocations.primitives import ImageCollectionOutput, ImageField, IntegerCollectionOutput
 from invokeai.app.util.misc import SEED_MAX, get_random_seed
 
-from .baseinvocation import BaseInvocation, BaseInvocationOutput, InvocationConfig, InvocationContext, UIConfig
-
-
-class IntCollectionOutput(BaseInvocationOutput):
-    """A collection of integers"""
-
-    type: Literal["int_collection"] = "int_collection"
-
-    # Outputs
-    collection: list[int] = Field(default=[], description="The int collection")
-
-
-class FloatCollectionOutput(BaseInvocationOutput):
-    """A collection of floats"""
-
-    type: Literal["float_collection"] = "float_collection"
-
-    # Outputs
-    collection: list[float] = Field(default=[], description="The float collection")
-
-
-class ImageCollectionOutput(BaseInvocationOutput):
-    """A collection of images"""
-
-    type: Literal["image_collection"] = "image_collection"
-
-    # Outputs
-    collection: list[ImageField] = Field(default=[], description="The output images")
-
-    class Config:
-        schema_extra = {"required": ["type", "collection"]}
+from .baseinvocation import BaseInvocation, InputField, InvocationContext, UIType, tags, title
 
 
+@title("Integer Range")
+@tags("collection", "integer", "range")
 class RangeInvocation(BaseInvocation):
     """Creates a range of numbers from start to stop with step"""
 
     type: Literal["range"] = "range"
 
     # Inputs
-    start: int = Field(default=0, description="The start of the range")
-    stop: int = Field(default=10, description="The stop of the range")
-    step: int = Field(default=1, description="The step of the range")
-
-    class Config(InvocationConfig):
-        schema_extra = {
-            "ui": {"title": "Range", "tags": ["range", "integer", "collection"]},
-        }
+    start: int = InputField(default=0, description="The start of the range")
+    stop: int = InputField(default=10, description="The stop of the range")
+    step: int = InputField(default=1, description="The step of the range")
 
     @validator("stop")
     def stop_gt_start(cls, v, values):
@@ -62,76 +29,44 @@ class RangeInvocation(BaseInvocation):
             raise ValueError("stop must be greater than start")
         return v
 
-    def invoke(self, context: InvocationContext) -> IntCollectionOutput:
-        return IntCollectionOutput(collection=list(range(self.start, self.stop, self.step)))
+    def invoke(self, context: InvocationContext) -> IntegerCollectionOutput:
+        return IntegerCollectionOutput(collection=list(range(self.start, self.stop, self.step)))
 
 
+@title("Integer Range of Size")
+@tags("range", "integer", "size", "collection")
 class RangeOfSizeInvocation(BaseInvocation):
     """Creates a range from start to start + size with step"""
 
     type: Literal["range_of_size"] = "range_of_size"
 
     # Inputs
-    start: int = Field(default=0, description="The start of the range")
-    size: int = Field(default=1, description="The number of values")
-    step: int = Field(default=1, description="The step of the range")
+    start: int = InputField(default=0, description="The start of the range")
+    size: int = InputField(default=1, description="The number of values")
+    step: int = InputField(default=1, description="The step of the range")
 
-    class Config(InvocationConfig):
-        schema_extra = {
-            "ui": {"title": "Sized Range", "tags": ["range", "integer", "size", "collection"]},
-        }
-
-    def invoke(self, context: InvocationContext) -> IntCollectionOutput:
-        return IntCollectionOutput(collection=list(range(self.start, self.start + self.size, self.step)))
+    def invoke(self, context: InvocationContext) -> IntegerCollectionOutput:
+        return IntegerCollectionOutput(collection=list(range(self.start, self.start + self.size, self.step)))
 
 
+@title("Random Range")
+@tags("range", "integer", "random", "collection")
 class RandomRangeInvocation(BaseInvocation):
     """Creates a collection of random numbers"""
 
     type: Literal["random_range"] = "random_range"
 
     # Inputs
-    low: int = Field(default=0, description="The inclusive low value")
-    high: int = Field(default=np.iinfo(np.int32).max, description="The exclusive high value")
-    size: int = Field(default=1, description="The number of values to generate")
-    seed: int = Field(
+    low: int = InputField(default=0, description="The inclusive low value")
+    high: int = InputField(default=np.iinfo(np.int32).max, description="The exclusive high value")
+    size: int = InputField(default=1, description="The number of values to generate")
+    seed: int = InputField(
         ge=0,
         le=SEED_MAX,
         description="The seed for the RNG (omit for random)",
         default_factory=get_random_seed,
     )
 
-    class Config(InvocationConfig):
-        schema_extra = {
-            "ui": {"title": "Random Range", "tags": ["range", "integer", "random", "collection"]},
-        }
-
-    def invoke(self, context: InvocationContext) -> IntCollectionOutput:
+    def invoke(self, context: InvocationContext) -> IntegerCollectionOutput:
         rng = np.random.default_rng(self.seed)
-        return IntCollectionOutput(collection=list(rng.integers(low=self.low, high=self.high, size=self.size)))
-
-
-class ImageCollectionInvocation(BaseInvocation):
-    """Load a collection of images and provide it as output."""
-
-    # fmt: off
-    type: Literal["image_collection"] = "image_collection"
-
-    # Inputs
-    images: list[ImageField] = Field(
-        default=[], description="The image collection to load"
-    )
-    # fmt: on
-
-    def invoke(self, context: InvocationContext) -> ImageCollectionOutput:
-        return ImageCollectionOutput(collection=self.images)
-
-    class Config(InvocationConfig):
-        schema_extra = {
-            "ui": {
-                "type_hints": {
-                    "title": "Image Collection",
-                    "images": "image_collection",
-                }
-            },
-        }
+        return IntegerCollectionOutput(collection=list(rng.integers(low=self.low, high=self.high, size=self.size)))

invokeai/app/invocations/compel.py

@@ -1,56 +1,40 @@
-from typing import Literal, Optional, Union, List, Annotated
-from pydantic import BaseModel, Field
 import re
-
-from .baseinvocation import BaseInvocation, BaseInvocationOutput, InvocationContext, InvocationConfig
-from .model import ClipField
-
-from ...backend.util.devices import torch_dtype
-from ...backend.stable_diffusion.diffusion import InvokeAIDiffuserComponent
-from ...backend.model_management import BaseModelType, ModelType, SubModelType, ModelPatcher
+from dataclasses import dataclass
+from typing import List, Literal, Union
 
 import torch
 from compel import Compel, ReturnedEmbeddingsType
 from compel.prompt_parser import Blend, Conjunction, CrossAttentionControlSubstitute, FlattenedPrompt, Fragment
-from ...backend.util.devices import torch_dtype
-from ...backend.model_management import ModelType
-from ...backend.model_management.models import ModelNotFoundException
+from invokeai.app.invocations.primitives import ConditioningField, ConditioningOutput
+
+from invokeai.backend.stable_diffusion.diffusion.shared_invokeai_diffusion import (
+    BasicConditioningInfo,
+    SDXLConditioningInfo,
+)
+
+from ...backend.model_management import ModelPatcher, ModelType
 from ...backend.model_management.lora import ModelPatcher
+from ...backend.model_management.models import ModelNotFoundException
 from ...backend.stable_diffusion.diffusion import InvokeAIDiffuserComponent
-from .baseinvocation import BaseInvocation, BaseInvocationOutput, InvocationConfig, InvocationContext
+from ...backend.util.devices import torch_dtype
+from .baseinvocation import (
+    BaseInvocation,
+    BaseInvocationOutput,
+    FieldDescriptions,
+    Input,
+    InputField,
+    InvocationContext,
+    OutputField,
+    UIComponent,
+    tags,
+    title,
+)
 from .model import ClipField
-from dataclasses import dataclass
-
-
-class ConditioningField(BaseModel):
-    conditioning_name: Optional[str] = Field(default=None, description="The name of conditioning data")
-
-    class Config:
-        schema_extra = {"required": ["conditioning_name"]}
-
-
-@dataclass
-class BasicConditioningInfo:
-    # type: Literal["basic_conditioning"] = "basic_conditioning"
-    embeds: torch.Tensor
-    extra_conditioning: Optional[InvokeAIDiffuserComponent.ExtraConditioningInfo]
-    # weight: float
-    # mode: ConditioningAlgo
-
-
-@dataclass
-class SDXLConditioningInfo(BasicConditioningInfo):
-    # type: Literal["sdxl_conditioning"] = "sdxl_conditioning"
-    pooled_embeds: torch.Tensor
-    add_time_ids: torch.Tensor
-
-
-ConditioningInfoType = Annotated[Union[BasicConditioningInfo, SDXLConditioningInfo], Field(discriminator="type")]
 
 
 @dataclass
 class ConditioningFieldData:
-    conditionings: List[Union[BasicConditioningInfo, SDXLConditioningInfo]]
+    conditionings: List[BasicConditioningInfo]
     # unconditioned: Optional[torch.Tensor]
 
 
@@ -60,32 +44,26 @@ class ConditioningFieldData:
 #    PerpNeg = "perp_neg"
 
 
-class CompelOutput(BaseInvocationOutput):
-    """Compel parser output"""
-
-    # fmt: off
-    type: Literal["compel_output"] = "compel_output"
-
-    conditioning: ConditioningField = Field(default=None, description="Conditioning")
-    # fmt: on
-
-
+@title("Compel Prompt")
+@tags("prompt", "compel")
 class CompelInvocation(BaseInvocation):
     """Parse prompt using compel package to conditioning."""
 
     type: Literal["compel"] = "compel"
 
-    prompt: str = Field(default="", description="Prompt")
-    clip: ClipField = Field(None, description="Clip to use")
-
-    # Schema customisation
-    class Config(InvocationConfig):
-        schema_extra = {
-            "ui": {"title": "Prompt (Compel)", "tags": ["prompt", "compel"], "type_hints": {"model": "model"}},
-        }
+    prompt: str = InputField(
+        default="",
+        description=FieldDescriptions.compel_prompt,
+        ui_component=UIComponent.Textarea,
+    )
+    clip: ClipField = InputField(
+        title="CLIP",
+        description=FieldDescriptions.clip,
+        input=Input.Connection,
+    )
 
     @torch.no_grad()
-    def invoke(self, context: InvocationContext) -> CompelOutput:
+    def invoke(self, context: InvocationContext) -> ConditioningOutput:
         tokenizer_info = context.services.model_manager.get_model(
             **self.clip.tokenizer.dict(),
             context=context,
@@ -168,7 +146,7 @@ class CompelInvocation(BaseInvocation):
         conditioning_name = f"{context.graph_execution_state_id}_{self.id}_conditioning"
         context.services.latents.save(conditioning_name, conditioning_data)
 
-        return CompelOutput(
+        return ConditioningOutput(
             conditioning=ConditioningField(
                 conditioning_name=conditioning_name,
             ),
@@ -176,7 +154,15 @@ class CompelInvocation(BaseInvocation):
 
 
 class SDXLPromptInvocationBase:
-    def run_clip_raw(self, context, clip_field, prompt, get_pooled, lora_prefix):
+    def run_clip_compel(
+        self,
+        context: InvocationContext,
+        clip_field: ClipField,
+        prompt: str,
+        get_pooled: bool,
+        lora_prefix: str,
+        zero_on_empty: bool,
+    ):
         tokenizer_info = context.services.model_manager.get_model(
             **clip_field.tokenizer.dict(),
             context=context,
@@ -186,82 +172,21 @@ class SDXLPromptInvocationBase:
             context=context,
         )
 
-        def _lora_loader():
-            for lora in clip_field.loras:
-                lora_info = context.services.model_manager.get_model(**lora.dict(exclude={"weight"}), context=context)
-                yield (lora_info.context.model, lora.weight)
-                del lora_info
-            return
-
-        # loras = [(context.services.model_manager.get_model(**lora.dict(exclude={"weight"})).context.model, lora.weight) for lora in self.clip.loras]
-
-        ti_list = []
-        for trigger in re.findall(r"<[a-zA-Z0-9., _-]+>", prompt):
-            name = trigger[1:-1]
-            try:
-                ti_list.append(
-                    (
-                        name,
-                        context.services.model_manager.get_model(
-                            model_name=name,
-                            base_model=clip_field.text_encoder.base_model,
-                            model_type=ModelType.TextualInversion,
-                            context=context,
-                        ).context.model,
-                    )
-                )
-            except ModelNotFoundException:
-                # print(e)
-                # import traceback
-                # print(traceback.format_exc())
-                print(f'Warn: trigger: "{trigger}" not found')
-
-        with ModelPatcher.apply_lora(
-            text_encoder_info.context.model, _lora_loader(), lora_prefix
-        ), ModelPatcher.apply_ti(tokenizer_info.context.model, text_encoder_info.context.model, ti_list) as (
-            tokenizer,
-            ti_manager,
-        ), ModelPatcher.apply_clip_skip(
-            text_encoder_info.context.model, clip_field.skipped_layers
-        ), text_encoder_info as text_encoder:
-            text_inputs = tokenizer(
-                prompt,
-                padding="max_length",
-                max_length=tokenizer.model_max_length,
-                truncation=True,
-                return_tensors="pt",
-            )
-            text_input_ids = text_inputs.input_ids
-            prompt_embeds = text_encoder(
-                text_input_ids.to(text_encoder.device),
-                output_hidden_states=True,
+        # return zero on empty
+        if prompt == "" and zero_on_empty:
+            cpu_text_encoder = text_encoder_info.context.model
+            c = torch.zeros(
+                (1, cpu_text_encoder.config.max_position_embeddings, cpu_text_encoder.config.hidden_size),
+                dtype=text_encoder_info.context.cache.precision,
             )
             if get_pooled:
-                c_pooled = prompt_embeds[0]
+                c_pooled = torch.zeros(
+                    (1, cpu_text_encoder.config.hidden_size),
+                    dtype=c.dtype,
+                )
             else:
                 c_pooled = None
-            c = prompt_embeds.hidden_states[-2]
-
-        del tokenizer
-        del text_encoder
-        del tokenizer_info
-        del text_encoder_info
-
-        c = c.detach().to("cpu")
-        if c_pooled is not None:
-            c_pooled = c_pooled.detach().to("cpu")
-
-        return c, c_pooled, None
-
-    def run_clip_compel(self, context, clip_field, prompt, get_pooled, lora_prefix):
-        tokenizer_info = context.services.model_manager.get_model(
-            **clip_field.tokenizer.dict(),
-            context=context,
-        )
-        text_encoder_info = context.services.model_manager.get_model(
-            **clip_field.text_encoder.dict(),
-            context=context,
-        )
+            return c, c_pooled, None
 
         def _lora_loader():
             for lora in clip_field.loras:
@@ -342,35 +267,37 @@ class SDXLPromptInvocationBase:
         return c, c_pooled, ec
 
 
+@title("SDXL Compel Prompt")
+@tags("sdxl", "compel", "prompt")
 class SDXLCompelPromptInvocation(BaseInvocation, SDXLPromptInvocationBase):
     """Parse prompt using compel package to conditioning."""
 
     type: Literal["sdxl_compel_prompt"] = "sdxl_compel_prompt"
 
-    prompt: str = Field(default="", description="Prompt")
-    style: str = Field(default="", description="Style prompt")
-    original_width: int = Field(1024, description="")
-    original_height: int = Field(1024, description="")
-    crop_top: int = Field(0, description="")
-    crop_left: int = Field(0, description="")
-    target_width: int = Field(1024, description="")
-    target_height: int = Field(1024, description="")
-    clip: ClipField = Field(None, description="Clip to use")
-    clip2: ClipField = Field(None, description="Clip2 to use")
-
-    # Schema customisation
-    class Config(InvocationConfig):
-        schema_extra = {
-            "ui": {"title": "SDXL Prompt (Compel)", "tags": ["prompt", "compel"], "type_hints": {"model": "model"}},
-        }
+    prompt: str = InputField(default="", description=FieldDescriptions.compel_prompt, ui_component=UIComponent.Textarea)
+    style: str = InputField(default="", description=FieldDescriptions.compel_prompt, ui_component=UIComponent.Textarea)
+    original_width: int = InputField(default=1024, description="")
+    original_height: int = InputField(default=1024, description="")
+    crop_top: int = InputField(default=0, description="")
+    crop_left: int = InputField(default=0, description="")
+    target_width: int = InputField(default=1024, description="")
+    target_height: int = InputField(default=1024, description="")
+    clip: ClipField = InputField(description=FieldDescriptions.clip, input=Input.Connection)
+    clip2: ClipField = InputField(description=FieldDescriptions.clip, input=Input.Connection)
 
     @torch.no_grad()
-    def invoke(self, context: InvocationContext) -> CompelOutput:
-        c1, c1_pooled, ec1 = self.run_clip_compel(context, self.clip, self.prompt, False, "lora_te1_")
+    def invoke(self, context: InvocationContext) -> ConditioningOutput:
+        c1, c1_pooled, ec1 = self.run_clip_compel(
+            context, self.clip, self.prompt, False, "lora_te1_", zero_on_empty=True
+        )
         if self.style.strip() == "":
-            c2, c2_pooled, ec2 = self.run_clip_compel(context, self.clip2, self.prompt, True, "lora_te2_")
+            c2, c2_pooled, ec2 = self.run_clip_compel(
+                context, self.clip2, self.prompt, True, "lora_te2_", zero_on_empty=True
+            )
         else:
-            c2, c2_pooled, ec2 = self.run_clip_compel(context, self.clip2, self.style, True, "lora_te2_")
+            c2, c2_pooled, ec2 = self.run_clip_compel(
+                context, self.clip2, self.style, True, "lora_te2_", zero_on_empty=True
+            )
 
         original_size = (self.original_height, self.original_width)
         crop_coords = (self.crop_top, self.crop_left)
@@ -392,40 +319,34 @@ class SDXLCompelPromptInvocation(BaseInvocation, SDXLPromptInvocationBase):
         conditioning_name = f"{context.graph_execution_state_id}_{self.id}_conditioning"
         context.services.latents.save(conditioning_name, conditioning_data)
 
-        return CompelOutput(
+        return ConditioningOutput(
             conditioning=ConditioningField(
                 conditioning_name=conditioning_name,
             ),
         )
 
 
+@title("SDXL Refiner Compel Prompt")
+@tags("sdxl", "compel", "prompt")
 class SDXLRefinerCompelPromptInvocation(BaseInvocation, SDXLPromptInvocationBase):
     """Parse prompt using compel package to conditioning."""
 
     type: Literal["sdxl_refiner_compel_prompt"] = "sdxl_refiner_compel_prompt"
 
-    style: str = Field(default="", description="Style prompt")  # TODO: ?
-    original_width: int = Field(1024, description="")
-    original_height: int = Field(1024, description="")
-    crop_top: int = Field(0, description="")
-    crop_left: int = Field(0, description="")
-    aesthetic_score: float = Field(6.0, description="")
-    clip2: ClipField = Field(None, description="Clip to use")
-
-    # Schema customisation
-    class Config(InvocationConfig):
-        schema_extra = {
-            "ui": {
-                "title": "SDXL Refiner Prompt (Compel)",
-                "tags": ["prompt", "compel"],
-                "type_hints": {"model": "model"},
-            },
-        }
+    style: str = InputField(
+        default="", description=FieldDescriptions.compel_prompt, ui_component=UIComponent.Textarea
+    )  # TODO: ?
+    original_width: int = InputField(default=1024, description="")
+    original_height: int = InputField(default=1024, description="")
+    crop_top: int = InputField(default=0, description="")
+    crop_left: int = InputField(default=0, description="")
+    aesthetic_score: float = InputField(default=6.0, description=FieldDescriptions.sdxl_aesthetic)
+    clip2: ClipField = InputField(description=FieldDescriptions.clip, input=Input.Connection)
 
     @torch.no_grad()
-    def invoke(self, context: InvocationContext) -> CompelOutput:
+    def invoke(self, context: InvocationContext) -> ConditioningOutput:
         # TODO: if there will appear lora for refiner - write proper prefix
-        c2, c2_pooled, ec2 = self.run_clip_compel(context, self.clip2, self.style, True, "<NONE>")
+        c2, c2_pooled, ec2 = self.run_clip_compel(context, self.clip2, self.style, True, "<NONE>", zero_on_empty=False)
 
         original_size = (self.original_height, self.original_width)
         crop_coords = (self.crop_top, self.crop_left)
@@ -446,118 +367,7 @@ class SDXLRefinerCompelPromptInvocation(BaseInvocation, SDXLPromptInvocationBase
         conditioning_name = f"{context.graph_execution_state_id}_{self.id}_conditioning"
         context.services.latents.save(conditioning_name, conditioning_data)
 
-        return CompelOutput(
-            conditioning=ConditioningField(
-                conditioning_name=conditioning_name,
-            ),
-        )
-
-
-class SDXLRawPromptInvocation(BaseInvocation, SDXLPromptInvocationBase):
-    """Pass unmodified prompt to conditioning without compel processing."""
-
-    type: Literal["sdxl_raw_prompt"] = "sdxl_raw_prompt"
-
-    prompt: str = Field(default="", description="Prompt")
-    style: str = Field(default="", description="Style prompt")
-    original_width: int = Field(1024, description="")
-    original_height: int = Field(1024, description="")
-    crop_top: int = Field(0, description="")
-    crop_left: int = Field(0, description="")
-    target_width: int = Field(1024, description="")
-    target_height: int = Field(1024, description="")
-    clip: ClipField = Field(None, description="Clip to use")
-    clip2: ClipField = Field(None, description="Clip2 to use")
-
-    # Schema customisation
-    class Config(InvocationConfig):
-        schema_extra = {
-            "ui": {"title": "SDXL Prompt (Raw)", "tags": ["prompt", "compel"], "type_hints": {"model": "model"}},
-        }
-
-    @torch.no_grad()
-    def invoke(self, context: InvocationContext) -> CompelOutput:
-        c1, c1_pooled, ec1 = self.run_clip_raw(context, self.clip, self.prompt, False, "lora_te1_")
-        if self.style.strip() == "":
-            c2, c2_pooled, ec2 = self.run_clip_raw(context, self.clip2, self.prompt, True, "lora_te2_")
-        else:
-            c2, c2_pooled, ec2 = self.run_clip_raw(context, self.clip2, self.style, True, "lora_te2_")
-
-        original_size = (self.original_height, self.original_width)
-        crop_coords = (self.crop_top, self.crop_left)
-        target_size = (self.target_height, self.target_width)
-
-        add_time_ids = torch.tensor([original_size + crop_coords + target_size])
-
-        conditioning_data = ConditioningFieldData(
-            conditionings=[
-                SDXLConditioningInfo(
-                    embeds=torch.cat([c1, c2], dim=-1),
-                    pooled_embeds=c2_pooled,
-                    add_time_ids=add_time_ids,
-                    extra_conditioning=ec1,
-                )
-            ]
-        )
-
-        conditioning_name = f"{context.graph_execution_state_id}_{self.id}_conditioning"
-        context.services.latents.save(conditioning_name, conditioning_data)
-
-        return CompelOutput(
-            conditioning=ConditioningField(
-                conditioning_name=conditioning_name,
-            ),
-        )
-
-
-class SDXLRefinerRawPromptInvocation(BaseInvocation, SDXLPromptInvocationBase):
-    """Parse prompt using compel package to conditioning."""
-
-    type: Literal["sdxl_refiner_raw_prompt"] = "sdxl_refiner_raw_prompt"
-
-    style: str = Field(default="", description="Style prompt")  # TODO: ?
-    original_width: int = Field(1024, description="")
-    original_height: int = Field(1024, description="")
-    crop_top: int = Field(0, description="")
-    crop_left: int = Field(0, description="")
-    aesthetic_score: float = Field(6.0, description="")
-    clip2: ClipField = Field(None, description="Clip to use")
-
-    # Schema customisation
-    class Config(InvocationConfig):
-        schema_extra = {
-            "ui": {
-                "title": "SDXL Refiner Prompt (Raw)",
-                "tags": ["prompt", "compel"],
-                "type_hints": {"model": "model"},
-            },
-        }
-
-    @torch.no_grad()
-    def invoke(self, context: InvocationContext) -> CompelOutput:
-        # TODO: if there will appear lora for refiner - write proper prefix
-        c2, c2_pooled, ec2 = self.run_clip_raw(context, self.clip2, self.style, True, "<NONE>")
-
-        original_size = (self.original_height, self.original_width)
-        crop_coords = (self.crop_top, self.crop_left)
-
-        add_time_ids = torch.tensor([original_size + crop_coords + (self.aesthetic_score,)])
-
-        conditioning_data = ConditioningFieldData(
-            conditionings=[
-                SDXLConditioningInfo(
-                    embeds=c2,
-                    pooled_embeds=c2_pooled,
-                    add_time_ids=add_time_ids,
-                    extra_conditioning=ec2,  # or None
-                )
-            ]
-        )
-
-        conditioning_name = f"{context.graph_execution_state_id}_{self.id}_conditioning"
-        context.services.latents.save(conditioning_name, conditioning_data)
-
-        return CompelOutput(
+        return ConditioningOutput(
             conditioning=ConditioningField(
                 conditioning_name=conditioning_name,
             ),
@@ -568,21 +378,18 @@ class ClipSkipInvocationOutput(BaseInvocationOutput):
     """Clip skip node output"""
 
     type: Literal["clip_skip_output"] = "clip_skip_output"
-    clip: ClipField = Field(None, description="Clip with skipped layers")
+    clip: ClipField = OutputField(default=None, description=FieldDescriptions.clip, title="CLIP")
 
 
+@title("CLIP Skip")
+@tags("clipskip", "clip", "skip")
 class ClipSkipInvocation(BaseInvocation):
     """Skip layers in clip text_encoder model."""
 
     type: Literal["clip_skip"] = "clip_skip"
 
-    clip: ClipField = Field(None, description="Clip to use")
-    skipped_layers: int = Field(0, description="Number of layers to skip in text_encoder")
-
-    class Config(InvocationConfig):
-        schema_extra = {
-            "ui": {"title": "CLIP Skip", "tags": ["clip", "skip"]},
-        }
+    clip: ClipField = InputField(description=FieldDescriptions.clip, input=Input.Connection, title="CLIP")
+    skipped_layers: int = InputField(default=0, description=FieldDescriptions.skipped_layers)
 
     def invoke(self, context: InvocationContext) -> ClipSkipInvocationOutput:
         self.clip.skipped_layers += self.skipped_layers

invokeai/app/invocations/controlnet_image_processors.py

@@ -26,79 +26,31 @@ from controlnet_aux.util import HWC3, ade_palette
 from PIL import Image
 from pydantic import BaseModel, Field, validator
 
+from invokeai.app.invocations.primitives import ImageField, ImageOutput
+
+
 from ...backend.model_management import BaseModelType, ModelType
-from ..models.image import ImageCategory, ImageField, ResourceOrigin
-from .baseinvocation import BaseInvocation, BaseInvocationOutput, InvocationConfig, InvocationContext
-from ..models.image import ImageOutput, PILInvocationConfig
+from ..models.image import ImageCategory, ResourceOrigin
+from .baseinvocation import (
+    BaseInvocation,
+    BaseInvocationOutput,
+    FieldDescriptions,
+    InputField,
+    Input,
+    InvocationContext,
+    OutputField,
+    UIType,
+    tags,
+    title,
+)
 
-CONTROLNET_DEFAULT_MODELS = [
-    ###########################################
-    # lllyasviel sd v1.5, ControlNet v1.0 models
-    ##############################################
-    "lllyasviel/sd-controlnet-canny",
-    "lllyasviel/sd-controlnet-depth",
-    "lllyasviel/sd-controlnet-hed",
-    "lllyasviel/sd-controlnet-seg",
-    "lllyasviel/sd-controlnet-openpose",
-    "lllyasviel/sd-controlnet-scribble",
-    "lllyasviel/sd-controlnet-normal",
-    "lllyasviel/sd-controlnet-mlsd",
-    #############################################
-    # lllyasviel sd v1.5, ControlNet v1.1 models
-    #############################################
-    "lllyasviel/control_v11p_sd15_canny",
-    "lllyasviel/control_v11p_sd15_openpose",
-    "lllyasviel/control_v11p_sd15_seg",
-    # "lllyasviel/control_v11p_sd15_depth",  # broken
-    "lllyasviel/control_v11f1p_sd15_depth",
-    "lllyasviel/control_v11p_sd15_normalbae",
-    "lllyasviel/control_v11p_sd15_scribble",
-    "lllyasviel/control_v11p_sd15_mlsd",
-    "lllyasviel/control_v11p_sd15_softedge",
-    "lllyasviel/control_v11p_sd15s2_lineart_anime",
-    "lllyasviel/control_v11p_sd15_lineart",
-    "lllyasviel/control_v11p_sd15_inpaint",
-    # "lllyasviel/control_v11u_sd15_tile",
-    # problem (temporary?) with huffingface "lllyasviel/control_v11u_sd15_tile",
-    # so for now replace  "lllyasviel/control_v11f1e_sd15_tile",
-    "lllyasviel/control_v11e_sd15_shuffle",
-    "lllyasviel/control_v11e_sd15_ip2p",
-    "lllyasviel/control_v11f1e_sd15_tile",
-    #################################################
-    #  thibaud sd v2.1 models (ControlNet v1.0? or v1.1?
-    ##################################################
-    "thibaud/controlnet-sd21-openpose-diffusers",
-    "thibaud/controlnet-sd21-canny-diffusers",
-    "thibaud/controlnet-sd21-depth-diffusers",
-    "thibaud/controlnet-sd21-scribble-diffusers",
-    "thibaud/controlnet-sd21-hed-diffusers",
-    "thibaud/controlnet-sd21-zoedepth-diffusers",
-    "thibaud/controlnet-sd21-color-diffusers",
-    "thibaud/controlnet-sd21-openposev2-diffusers",
-    "thibaud/controlnet-sd21-lineart-diffusers",
-    "thibaud/controlnet-sd21-normalbae-diffusers",
-    "thibaud/controlnet-sd21-ade20k-diffusers",
-    ##############################################
-    #  ControlNetMediaPipeface, ControlNet v1.1
-    ##############################################
-    # ["CrucibleAI/ControlNetMediaPipeFace", "diffusion_sd15"],  # SD 1.5
-    #    diffusion_sd15 needs to be passed to from_pretrained() as subfolder arg
-    #    hacked t2l to split to model & subfolder if format is "model,subfolder"
-    "CrucibleAI/ControlNetMediaPipeFace,diffusion_sd15",  # SD 1.5
-    "CrucibleAI/ControlNetMediaPipeFace",  # SD 2.1?
-]
 
-CONTROLNET_NAME_VALUES = Literal[tuple(CONTROLNET_DEFAULT_MODELS)]
-CONTROLNET_MODE_VALUES = Literal[tuple(["balanced", "more_prompt", "more_control", "unbalanced"])]
+CONTROLNET_MODE_VALUES = Literal["balanced", "more_prompt", "more_control", "unbalanced"]
 CONTROLNET_RESIZE_VALUES = Literal[
-    tuple(
-        [
-            "just_resize",
-            "crop_resize",
-            "fill_resize",
-            "just_resize_simple",
-        ]
-    )
+    "just_resize",
+    "crop_resize",
+    "fill_resize",
+    "just_resize_simple",
 ]
 
 
@@ -110,9 +62,8 @@ class ControlNetModelField(BaseModel):
 
 
 class ControlField(BaseModel):
-    image: ImageField = Field(default=None, description="The control image")
-    control_model: Optional[ControlNetModelField] = Field(default=None, description="The ControlNet model to use")
-    # control_weight: Optional[float] = Field(default=1, description="weight given to controlnet")
+    image: ImageField = Field(description="The control image")
+    control_model: ControlNetModelField = Field(description="The ControlNet model to use")
     control_weight: Union[float, List[float]] = Field(default=1, description="The weight given to the ControlNet")
     begin_step_percent: float = Field(
         default=0, ge=0, le=1, description="When the ControlNet is first applied (% of total steps)"
@@ -135,60 +86,39 @@ class ControlField(BaseModel):
                 raise ValueError("Control weights must be within -1 to 2 range")
         return v
 
-    class Config:
-        schema_extra = {
-            "required": ["image", "control_model", "control_weight", "begin_step_percent", "end_step_percent"],
-            "ui": {
-                "type_hints": {
-                    "control_weight": "float",
-                    "control_model": "controlnet_model",
-                    # "control_weight": "number",
-                }
-            },
-        }
-
 
 class ControlOutput(BaseInvocationOutput):
     """node output for ControlNet info"""
 
-    # fmt: off
     type: Literal["control_output"] = "control_output"
-    control: ControlField = Field(default=None, description="The control info")
-    # fmt: on
+
+    # Outputs
+    control: ControlField = OutputField(description=FieldDescriptions.control)
 
 
+@title("ControlNet")
+@tags("controlnet")
 class ControlNetInvocation(BaseInvocation):
     """Collects ControlNet info to pass to other nodes"""
 
-    # fmt: off
     type: Literal["controlnet"] = "controlnet"
-    # Inputs
-    image: ImageField = Field(default=None, description="The control image")
-    control_model: ControlNetModelField = Field(default="lllyasviel/sd-controlnet-canny",
-                                                  description="control model used")
-    control_weight: Union[float, List[float]] = Field(default=1.0, description="The weight given to the ControlNet")
-    begin_step_percent: float = Field(default=0, ge=-1, le=2,
-                                      description="When the ControlNet is first applied (% of total steps)")
-    end_step_percent: float = Field(default=1, ge=0, le=1,
-                                    description="When the ControlNet is last applied (% of total steps)")
-    control_mode: CONTROLNET_MODE_VALUES = Field(default="balanced", description="The control mode used")
-    resize_mode: CONTROLNET_RESIZE_VALUES = Field(default="just_resize", description="The resize mode used")
-    # fmt: on
 
-    class Config(InvocationConfig):
-        schema_extra = {
-            "ui": {
-                "title": "ControlNet",
-                "tags": ["controlnet", "latents"],
-                "type_hints": {
-                    "model": "model",
-                    "control": "control",
-                    # "cfg_scale": "float",
-                    "cfg_scale": "number",
-                    "control_weight": "float",
-                },
-            },
-        }
+    # Inputs
+    image: ImageField = InputField(description="The control image")
+    control_model: ControlNetModelField = InputField(
+        default="lllyasviel/sd-controlnet-canny", description=FieldDescriptions.controlnet_model, input=Input.Direct
+    )
+    control_weight: Union[float, List[float]] = InputField(
+        default=1.0, description="The weight given to the ControlNet", ui_type=UIType.Float
+    )
+    begin_step_percent: float = InputField(
+        default=0, ge=-1, le=2, description="When the ControlNet is first applied (% of total steps)"
+    )
+    end_step_percent: float = InputField(
+        default=1, ge=0, le=1, description="When the ControlNet is last applied (% of total steps)"
+    )
+    control_mode: CONTROLNET_MODE_VALUES = InputField(default="balanced", description="The control mode used")
+    resize_mode: CONTROLNET_RESIZE_VALUES = InputField(default="just_resize", description="The resize mode used")
 
     def invoke(self, context: InvocationContext) -> ControlOutput:
         return ControlOutput(
@@ -204,19 +134,13 @@ class ControlNetInvocation(BaseInvocation):
         )
 
 
-class ImageProcessorInvocation(BaseInvocation, PILInvocationConfig):
+class ImageProcessorInvocation(BaseInvocation):
     """Base class for invocations that preprocess images for ControlNet"""
 
-    # fmt: off
     type: Literal["image_processor"] = "image_processor"
-    # Inputs
-    image: ImageField = Field(default=None, description="The image to process")
-    # fmt: on
 
-    class Config(InvocationConfig):
-        schema_extra = {
-            "ui": {"title": "Image Processor", "tags": ["image", "processor"]},
-        }
+    # Inputs
+    image: ImageField = InputField(description="The image to process")
 
     def run_processor(self, image):
         # superclass just passes through image without processing
@@ -255,20 +179,20 @@ class ImageProcessorInvocation(BaseInvocation, PILInvocationConfig):
         )
 
 
-class CannyImageProcessorInvocation(ImageProcessorInvocation, PILInvocationConfig):
+@title("Canny Processor")
+@tags("controlnet", "canny")
+class CannyImageProcessorInvocation(ImageProcessorInvocation):
     """Canny edge detection for ControlNet"""
 
-    # fmt: off
     type: Literal["canny_image_processor"] = "canny_image_processor"
-    # Input
-    low_threshold: int = Field(default=100, ge=0, le=255, description="The low threshold of the Canny pixel gradient (0-255)")
-    high_threshold: int = Field(default=200, ge=0, le=255, description="The high threshold of the Canny pixel gradient (0-255)")
-    # fmt: on
 
-    class Config(InvocationConfig):
-        schema_extra = {
-            "ui": {"title": "Canny Processor", "tags": ["controlnet", "canny", "image", "processor"]},
-        }
+    # Input
+    low_threshold: int = InputField(
+        default=100, ge=0, le=255, description="The low threshold of the Canny pixel gradient (0-255)"
+    )
+    high_threshold: int = InputField(
+        default=200, ge=0, le=255, description="The high threshold of the Canny pixel gradient (0-255)"
+    )
 
     def run_processor(self, image):
         canny_processor = CannyDetector()
@@ -276,23 +200,19 @@ class CannyImageProcessorInvocation(ImageProcessorInvocation, PILInvocationConfi
         return processed_image
 
 
-class HedImageProcessorInvocation(ImageProcessorInvocation, PILInvocationConfig):
+@title("HED (softedge) Processor")
+@tags("controlnet", "hed", "softedge")
+class HedImageProcessorInvocation(ImageProcessorInvocation):
     """Applies HED edge detection to image"""
 
-    # fmt: off
     type: Literal["hed_image_processor"] = "hed_image_processor"
-    # Inputs
-    detect_resolution: int = Field(default=512, ge=0, description="The pixel resolution for detection")
-    image_resolution: int = Field(default=512, ge=0, description="The pixel resolution for the output image")
-    # safe not supported in controlnet_aux v0.0.3
-    # safe: bool = Field(default=False, description="whether to use safe mode")
-    scribble: bool = Field(default=False, description="Whether to use scribble mode")
-    # fmt: on
 
-    class Config(InvocationConfig):
-        schema_extra = {
-            "ui": {"title": "Softedge(HED) Processor", "tags": ["controlnet", "softedge", "hed", "image", "processor"]},
-        }
+    # Inputs
+    detect_resolution: int = InputField(default=512, ge=0, description=FieldDescriptions.detect_res)
+    image_resolution: int = InputField(default=512, ge=0, description=FieldDescriptions.image_res)
+    # safe not supported in controlnet_aux v0.0.3
+    # safe: bool = InputField(default=False, description=FieldDescriptions.safe_mode)
+    scribble: bool = InputField(default=False, description=FieldDescriptions.scribble_mode)
 
     def run_processor(self, image):
         hed_processor = HEDdetector.from_pretrained("lllyasviel/Annotators")
@@ -307,21 +227,17 @@ class HedImageProcessorInvocation(ImageProcessorInvocation, PILInvocationConfig)
         return processed_image
 
 
-class LineartImageProcessorInvocation(ImageProcessorInvocation, PILInvocationConfig):
+@title("Lineart Processor")
+@tags("controlnet", "lineart")
+class LineartImageProcessorInvocation(ImageProcessorInvocation):
     """Applies line art processing to image"""
 
-    # fmt: off
     type: Literal["lineart_image_processor"] = "lineart_image_processor"
-    # Inputs
-    detect_resolution: int = Field(default=512, ge=0, description="The pixel resolution for detection")
-    image_resolution: int = Field(default=512, ge=0, description="The pixel resolution for the output image")
-    coarse: bool = Field(default=False, description="Whether to use coarse mode")
-    # fmt: on
 
-    class Config(InvocationConfig):
-        schema_extra = {
-            "ui": {"title": "Lineart Processor", "tags": ["controlnet", "lineart", "image", "processor"]},
-        }
+    # Inputs
+    detect_resolution: int = InputField(default=512, ge=0, description=FieldDescriptions.detect_res)
+    image_resolution: int = InputField(default=512, ge=0, description=FieldDescriptions.image_res)
+    coarse: bool = InputField(default=False, description="Whether to use coarse mode")
 
     def run_processor(self, image):
         lineart_processor = LineartDetector.from_pretrained("lllyasviel/Annotators")
@@ -331,23 +247,16 @@ class LineartImageProcessorInvocation(ImageProcessorInvocation, PILInvocationCon
         return processed_image
 
 
-class LineartAnimeImageProcessorInvocation(ImageProcessorInvocation, PILInvocationConfig):
+@title("Lineart Anime Processor")
+@tags("controlnet", "lineart", "anime")
+class LineartAnimeImageProcessorInvocation(ImageProcessorInvocation):
     """Applies line art anime processing to image"""
 
-    # fmt: off
     type: Literal["lineart_anime_image_processor"] = "lineart_anime_image_processor"
-    # Inputs
-    detect_resolution: int = Field(default=512, ge=0, description="The pixel resolution for detection")
-    image_resolution: int = Field(default=512, ge=0, description="The pixel resolution for the output image")
-    # fmt: on
 
-    class Config(InvocationConfig):
-        schema_extra = {
-            "ui": {
-                "title": "Lineart Anime Processor",
-                "tags": ["controlnet", "lineart", "anime", "image", "processor"],
-            },
-        }
+    # Inputs
+    detect_resolution: int = InputField(default=512, ge=0, description=FieldDescriptions.detect_res)
+    image_resolution: int = InputField(default=512, ge=0, description=FieldDescriptions.image_res)
 
     def run_processor(self, image):
         processor = LineartAnimeDetector.from_pretrained("lllyasviel/Annotators")
@@ -359,21 +268,17 @@ class LineartAnimeImageProcessorInvocation(ImageProcessorInvocation, PILInvocati
         return processed_image
 
 
-class OpenposeImageProcessorInvocation(ImageProcessorInvocation, PILInvocationConfig):
+@title("Openpose Processor")
+@tags("controlnet", "openpose", "pose")
+class OpenposeImageProcessorInvocation(ImageProcessorInvocation):
     """Applies Openpose processing to image"""
 
-    # fmt: off
     type: Literal["openpose_image_processor"] = "openpose_image_processor"
-    # Inputs
-    hand_and_face: bool = Field(default=False, description="Whether to use hands and face mode")
-    detect_resolution: int = Field(default=512, ge=0, description="The pixel resolution for detection")
-    image_resolution: int = Field(default=512, ge=0, description="The pixel resolution for the output image")
-    # fmt: on
 
-    class Config(InvocationConfig):
-        schema_extra = {
-            "ui": {"title": "Openpose Processor", "tags": ["controlnet", "openpose", "image", "processor"]},
-        }
+    # Inputs
+    hand_and_face: bool = InputField(default=False, description="Whether to use hands and face mode")
+    detect_resolution: int = InputField(default=512, ge=0, description=FieldDescriptions.detect_res)
+    image_resolution: int = InputField(default=512, ge=0, description=FieldDescriptions.image_res)
 
     def run_processor(self, image):
         openpose_processor = OpenposeDetector.from_pretrained("lllyasviel/Annotators")
@@ -386,22 +291,18 @@ class OpenposeImageProcessorInvocation(ImageProcessorInvocation, PILInvocationCo
         return processed_image
 
 
-class MidasDepthImageProcessorInvocation(ImageProcessorInvocation, PILInvocationConfig):
+@title("Midas (Depth) Processor")
+@tags("controlnet", "midas", "depth")
+class MidasDepthImageProcessorInvocation(ImageProcessorInvocation):
     """Applies Midas depth processing to image"""
 
-    # fmt: off
     type: Literal["midas_depth_image_processor"] = "midas_depth_image_processor"
-    # Inputs
-    a_mult: float = Field(default=2.0, ge=0, description="Midas parameter `a_mult` (a = a_mult * PI)")
-    bg_th: float = Field(default=0.1, ge=0, description="Midas parameter `bg_th`")
-    # depth_and_normal not supported in controlnet_aux v0.0.3
-    # depth_and_normal: bool = Field(default=False, description="whether to use depth and normal mode")
-    # fmt: on
 
-    class Config(InvocationConfig):
-        schema_extra = {
-            "ui": {"title": "Midas (Depth) Processor", "tags": ["controlnet", "midas", "depth", "image", "processor"]},
-        }
+    # Inputs
+    a_mult: float = InputField(default=2.0, ge=0, description="Midas parameter `a_mult` (a = a_mult * PI)")
+    bg_th: float = InputField(default=0.1, ge=0, description="Midas parameter `bg_th`")
+    # depth_and_normal not supported in controlnet_aux v0.0.3
+    # depth_and_normal: bool = InputField(default=False, description="whether to use depth and normal mode")
 
     def run_processor(self, image):
         midas_processor = MidasDetector.from_pretrained("lllyasviel/Annotators")
@@ -415,20 +316,16 @@ class MidasDepthImageProcessorInvocation(ImageProcessorInvocation, PILInvocation
         return processed_image
 
 
-class NormalbaeImageProcessorInvocation(ImageProcessorInvocation, PILInvocationConfig):
+@title("Normal BAE Processor")
+@tags("controlnet", "normal", "bae")
+class NormalbaeImageProcessorInvocation(ImageProcessorInvocation):
     """Applies NormalBae processing to image"""
 
-    # fmt: off
     type: Literal["normalbae_image_processor"] = "normalbae_image_processor"
-    # Inputs
-    detect_resolution: int = Field(default=512, ge=0, description="The pixel resolution for detection")
-    image_resolution: int = Field(default=512, ge=0, description="The pixel resolution for the output image")
-    # fmt: on
 
-    class Config(InvocationConfig):
-        schema_extra = {
-            "ui": {"title": "Normal BAE Processor", "tags": ["controlnet", "normal", "bae", "image", "processor"]},
-        }
+    # Inputs
+    detect_resolution: int = InputField(default=512, ge=0, description=FieldDescriptions.detect_res)
+    image_resolution: int = InputField(default=512, ge=0, description=FieldDescriptions.image_res)
 
     def run_processor(self, image):
         normalbae_processor = NormalBaeDetector.from_pretrained("lllyasviel/Annotators")
@@ -438,22 +335,18 @@ class NormalbaeImageProcessorInvocation(ImageProcessorInvocation, PILInvocationC
         return processed_image
 
 
-class MlsdImageProcessorInvocation(ImageProcessorInvocation, PILInvocationConfig):
+@title("MLSD Processor")
+@tags("controlnet", "mlsd")
+class MlsdImageProcessorInvocation(ImageProcessorInvocation):
     """Applies MLSD processing to image"""
 
-    # fmt: off
     type: Literal["mlsd_image_processor"] = "mlsd_image_processor"
-    # Inputs
-    detect_resolution: int = Field(default=512, ge=0, description="The pixel resolution for detection")
-    image_resolution: int = Field(default=512, ge=0, description="The pixel resolution for the output image")
-    thr_v: float = Field(default=0.1, ge=0, description="MLSD parameter `thr_v`")
-    thr_d: float = Field(default=0.1, ge=0, description="MLSD parameter `thr_d`")
-    # fmt: on
 
-    class Config(InvocationConfig):
-        schema_extra = {
-            "ui": {"title": "MLSD Processor", "tags": ["controlnet", "mlsd", "image", "processor"]},
-        }
+    # Inputs
+    detect_resolution: int = InputField(default=512, ge=0, description=FieldDescriptions.detect_res)
+    image_resolution: int = InputField(default=512, ge=0, description=FieldDescriptions.image_res)
+    thr_v: float = InputField(default=0.1, ge=0, description="MLSD parameter `thr_v`")
+    thr_d: float = InputField(default=0.1, ge=0, description="MLSD parameter `thr_d`")
 
     def run_processor(self, image):
         mlsd_processor = MLSDdetector.from_pretrained("lllyasviel/Annotators")
@@ -467,22 +360,18 @@ class MlsdImageProcessorInvocation(ImageProcessorInvocation, PILInvocationConfig
         return processed_image
 
 
-class PidiImageProcessorInvocation(ImageProcessorInvocation, PILInvocationConfig):
+@title("PIDI Processor")
+@tags("controlnet", "pidi")
+class PidiImageProcessorInvocation(ImageProcessorInvocation):
     """Applies PIDI processing to image"""
 
-    # fmt: off
     type: Literal["pidi_image_processor"] = "pidi_image_processor"
-    # Inputs
-    detect_resolution: int = Field(default=512, ge=0, description="The pixel resolution for detection")
-    image_resolution: int = Field(default=512, ge=0, description="The pixel resolution for the output image")
-    safe: bool = Field(default=False, description="Whether to use safe mode")
-    scribble: bool = Field(default=False, description="Whether to use scribble mode")
-    # fmt: on
 
-    class Config(InvocationConfig):
-        schema_extra = {
-            "ui": {"title": "PIDI Processor", "tags": ["controlnet", "pidi", "image", "processor"]},
-        }
+    # Inputs
+    detect_resolution: int = InputField(default=512, ge=0, description=FieldDescriptions.detect_res)
+    image_resolution: int = InputField(default=512, ge=0, description=FieldDescriptions.image_res)
+    safe: bool = InputField(default=False, description=FieldDescriptions.safe_mode)
+    scribble: bool = InputField(default=False, description=FieldDescriptions.scribble_mode)
 
     def run_processor(self, image):
         pidi_processor = PidiNetDetector.from_pretrained("lllyasviel/Annotators")
@@ -496,26 +385,19 @@ class PidiImageProcessorInvocation(ImageProcessorInvocation, PILInvocationConfig
         return processed_image
 
 
-class ContentShuffleImageProcessorInvocation(ImageProcessorInvocation, PILInvocationConfig):
+@title("Content Shuffle Processor")
+@tags("controlnet", "contentshuffle")
+class ContentShuffleImageProcessorInvocation(ImageProcessorInvocation):
     """Applies content shuffle processing to image"""
 
-    # fmt: off
     type: Literal["content_shuffle_image_processor"] = "content_shuffle_image_processor"
-    # Inputs
-    detect_resolution: int = Field(default=512, ge=0, description="The pixel resolution for detection")
-    image_resolution: int = Field(default=512, ge=0, description="The pixel resolution for the output image")
-    h: Optional[int] = Field(default=512, ge=0, description="Content shuffle `h` parameter")
-    w: Optional[int] = Field(default=512, ge=0, description="Content shuffle `w` parameter")
-    f: Optional[int] = Field(default=256, ge=0, description="Content shuffle `f` parameter")
-    # fmt: on
 
-    class Config(InvocationConfig):
-        schema_extra = {
-            "ui": {
-                "title": "Content Shuffle Processor",
-                "tags": ["controlnet", "contentshuffle", "image", "processor"],
-            },
-        }
+    # Inputs
+    detect_resolution: int = InputField(default=512, ge=0, description=FieldDescriptions.detect_res)
+    image_resolution: int = InputField(default=512, ge=0, description=FieldDescriptions.image_res)
+    h: Optional[int] = InputField(default=512, ge=0, description="Content shuffle `h` parameter")
+    w: Optional[int] = InputField(default=512, ge=0, description="Content shuffle `w` parameter")
+    f: Optional[int] = InputField(default=256, ge=0, description="Content shuffle `f` parameter")
 
     def run_processor(self, image):
         content_shuffle_processor = ContentShuffleDetector()
@@ -531,17 +413,12 @@ class ContentShuffleImageProcessorInvocation(ImageProcessorInvocation, PILInvoca
 
 
 # should work with controlnet_aux >= 0.0.4 and timm <= 0.6.13
-class ZoeDepthImageProcessorInvocation(ImageProcessorInvocation, PILInvocationConfig):
+@title("Zoe (Depth) Processor")
+@tags("controlnet", "zoe", "depth")
+class ZoeDepthImageProcessorInvocation(ImageProcessorInvocation):
     """Applies Zoe depth processing to image"""
 
-    # fmt: off
     type: Literal["zoe_depth_image_processor"] = "zoe_depth_image_processor"
-    # fmt: on
-
-    class Config(InvocationConfig):
-        schema_extra = {
-            "ui": {"title": "Zoe (Depth) Processor", "tags": ["controlnet", "zoe", "depth", "image", "processor"]},
-        }
 
     def run_processor(self, image):
         zoe_depth_processor = ZoeDetector.from_pretrained("lllyasviel/Annotators")
@@ -549,20 +426,16 @@ class ZoeDepthImageProcessorInvocation(ImageProcessorInvocation, PILInvocationCo
         return processed_image
 
 
-class MediapipeFaceProcessorInvocation(ImageProcessorInvocation, PILInvocationConfig):
+@title("Mediapipe Face Processor")
+@tags("controlnet", "mediapipe", "face")
+class MediapipeFaceProcessorInvocation(ImageProcessorInvocation):
     """Applies mediapipe face processing to image"""
 
-    # fmt: off
     type: Literal["mediapipe_face_processor"] = "mediapipe_face_processor"
-    # Inputs
-    max_faces: int = Field(default=1, ge=1, description="Maximum number of faces to detect")
-    min_confidence: float = Field(default=0.5, ge=0, le=1, description="Minimum confidence for face detection")
-    # fmt: on
 
-    class Config(InvocationConfig):
-        schema_extra = {
-            "ui": {"title": "Mediapipe Processor", "tags": ["controlnet", "mediapipe", "image", "processor"]},
-        }
+    # Inputs
+    max_faces: int = InputField(default=1, ge=1, description="Maximum number of faces to detect")
+    min_confidence: float = InputField(default=0.5, ge=0, le=1, description="Minimum confidence for face detection")
 
     def run_processor(self, image):
         # MediaPipeFaceDetector throws an error if image has alpha channel
@@ -574,23 +447,19 @@ class MediapipeFaceProcessorInvocation(ImageProcessorInvocation, PILInvocationCo
         return processed_image
 
 
-class LeresImageProcessorInvocation(ImageProcessorInvocation, PILInvocationConfig):
+@title("Leres (Depth) Processor")
+@tags("controlnet", "leres", "depth")
+class LeresImageProcessorInvocation(ImageProcessorInvocation):
     """Applies leres processing to image"""
 
-    # fmt: off
     type: Literal["leres_image_processor"] = "leres_image_processor"
-    # Inputs
-    thr_a: float = Field(default=0, description="Leres parameter `thr_a`")
-    thr_b: float = Field(default=0, description="Leres parameter `thr_b`")
-    boost: bool = Field(default=False, description="Whether to use boost mode")
-    detect_resolution: int = Field(default=512, ge=0, description="The pixel resolution for detection")
-    image_resolution: int = Field(default=512, ge=0, description="The pixel resolution for the output image")
-    # fmt: on
 
-    class Config(InvocationConfig):
-        schema_extra = {
-            "ui": {"title": "Leres (Depth) Processor", "tags": ["controlnet", "leres", "depth", "image", "processor"]},
-        }
+    # Inputs
+    thr_a: float = InputField(default=0, description="Leres parameter `thr_a`")
+    thr_b: float = InputField(default=0, description="Leres parameter `thr_b`")
+    boost: bool = InputField(default=False, description="Whether to use boost mode")
+    detect_resolution: int = InputField(default=512, ge=0, description=FieldDescriptions.detect_res)
+    image_resolution: int = InputField(default=512, ge=0, description=FieldDescriptions.image_res)
 
     def run_processor(self, image):
         leres_processor = LeresDetector.from_pretrained("lllyasviel/Annotators")
@@ -605,21 +474,16 @@ class LeresImageProcessorInvocation(ImageProcessorInvocation, PILInvocationConfi
         return processed_image
 
 
-class TileResamplerProcessorInvocation(ImageProcessorInvocation, PILInvocationConfig):
-    # fmt: off
-    type: Literal["tile_image_processor"] = "tile_image_processor"
-    # Inputs
-    #res: int = Field(default=512, ge=0, le=1024, description="The pixel resolution for each tile")
-    down_sampling_rate: float = Field(default=1.0, ge=1.0, le=8.0, description="Down sampling rate")
-    # fmt: on
+@title("Tile Resample Processor")
+@tags("controlnet", "tile")
+class TileResamplerProcessorInvocation(ImageProcessorInvocation):
+    """Tile resampler processor"""
 
-    class Config(InvocationConfig):
-        schema_extra = {
-            "ui": {
-                "title": "Tile Resample Processor",
-                "tags": ["controlnet", "tile", "resample", "image", "processor"],
-            },
-        }
+    type: Literal["tile_image_processor"] = "tile_image_processor"
+
+    # Inputs
+    # res: int = InputField(default=512, ge=0, le=1024, description="The pixel resolution for each tile")
+    down_sampling_rate: float = InputField(default=1.0, ge=1.0, le=8.0, description="Down sampling rate")
 
     # tile_resample copied from sd-webui-controlnet/scripts/processor.py
     def tile_resample(
@@ -648,20 +512,12 @@ class TileResamplerProcessorInvocation(ImageProcessorInvocation, PILInvocationCo
         return processed_image
 
 
-class SegmentAnythingProcessorInvocation(ImageProcessorInvocation, PILInvocationConfig):
+@title("Segment Anything Processor")
+@tags("controlnet", "segmentanything")
+class SegmentAnythingProcessorInvocation(ImageProcessorInvocation):
     """Applies segment anything processing to image"""
 
-    # fmt: off
     type: Literal["segment_anything_processor"] = "segment_anything_processor"
-    # fmt: on
-
-    class Config(InvocationConfig):
-        schema_extra = {
-            "ui": {
-                "title": "Segment Anything Processor",
-                "tags": ["controlnet", "segment", "anything", "sam", "image", "processor"],
-            },
-        }
 
     def run_processor(self, image):
         # segment_anything_processor = SamDetector.from_pretrained("ybelkada/segment-anything", subfolder="checkpoints")

invokeai/app/invocations/cv.py

@@ -5,40 +5,22 @@ from typing import Literal
 import cv2 as cv
 import numpy
 from PIL import Image, ImageOps
-from pydantic import BaseModel, Field
+from invokeai.app.invocations.primitives import ImageField, ImageOutput
 
-from invokeai.app.models.image import ImageCategory, ImageField, ResourceOrigin
-from .baseinvocation import BaseInvocation, InvocationContext, InvocationConfig
-from .image import ImageOutput
+from invokeai.app.models.image import ImageCategory, ResourceOrigin
+from .baseinvocation import BaseInvocation, InputField, InvocationContext, tags, title
 
 
-class CvInvocationConfig(BaseModel):
-    """Helper class to provide all OpenCV invocations with additional config"""
-
-    # Schema customisation
-    class Config(InvocationConfig):
-        schema_extra = {
-            "ui": {
-                "tags": ["cv", "image"],
-            },
-        }
-
-
-class CvInpaintInvocation(BaseInvocation, CvInvocationConfig):
+@title("OpenCV Inpaint")
+@tags("opencv", "inpaint")
+class CvInpaintInvocation(BaseInvocation):
     """Simple inpaint using opencv."""
 
-    # fmt: off
     type: Literal["cv_inpaint"] = "cv_inpaint"
 
     # Inputs
-    image: ImageField = Field(default=None, description="The image to inpaint")
-    mask: ImageField = Field(default=None, description="The mask to use when inpainting")
-    # fmt: on
-
-    class Config(InvocationConfig):
-        schema_extra = {
-            "ui": {"title": "OpenCV Inpaint", "tags": ["opencv", "inpaint"]},
-        }
+    image: ImageField = InputField(description="The image to inpaint")
+    mask: ImageField = InputField(description="The mask to use when inpainting")
 
     def invoke(self, context: InvocationContext) -> ImageOutput:
         image = context.services.images.get_pil_image(self.image.image_name)

invokeai/app/invocations/generate.py

@@ -1,248 +0,0 @@
-# Copyright (c) 2022 Kyle Schouviller (https://github.com/kyle0654)
-
-from contextlib import contextmanager, ContextDecorator
-from functools import partial
-from typing import Literal, Optional, get_args
-
-from pydantic import Field
-
-from invokeai.app.models.image import ColorField, ImageCategory, ImageField, ResourceOrigin
-from invokeai.app.util.misc import SEED_MAX, get_random_seed
-from invokeai.backend.generator.inpaint import infill_methods
-from .baseinvocation import BaseInvocation, InvocationConfig, InvocationContext
-from .compel import ConditioningField
-from .image import ImageOutput
-from .model import UNetField, VaeField
-from ..util.step_callback import stable_diffusion_step_callback
-from ...backend.generator import Inpaint, InvokeAIGenerator
-from ...backend.model_management.lora import ModelPatcher
-from ...backend.stable_diffusion import PipelineIntermediateState
-from ...backend.stable_diffusion.diffusers_pipeline import StableDiffusionGeneratorPipeline
-
-SAMPLER_NAME_VALUES = Literal[tuple(InvokeAIGenerator.schedulers())]
-INFILL_METHODS = Literal[tuple(infill_methods())]
-DEFAULT_INFILL_METHOD = "patchmatch" if "patchmatch" in get_args(INFILL_METHODS) else "tile"
-
-
-from .latent import get_scheduler
-
-
-class OldModelContext(ContextDecorator):
-    model: StableDiffusionGeneratorPipeline
-
-    def __init__(self, model):
-        self.model = model
-
-    def __enter__(self):
-        return self.model
-
-    def __exit__(self, *exc):
-        return False
-
-
-class OldModelInfo:
-    name: str
-    hash: str
-    context: OldModelContext
-
-    def __init__(self, name: str, hash: str, model: StableDiffusionGeneratorPipeline):
-        self.name = name
-        self.hash = hash
-        self.context = OldModelContext(
-            model=model,
-        )
-
-
-class InpaintInvocation(BaseInvocation):
-    """Generates an image using inpaint."""
-
-    type: Literal["inpaint"] = "inpaint"
-
-    positive_conditioning: Optional[ConditioningField] = Field(description="Positive conditioning for generation")
-    negative_conditioning: Optional[ConditioningField] = Field(description="Negative conditioning for generation")
-    seed: int = Field(
-        ge=0, le=SEED_MAX, description="The seed to use (omit for random)", default_factory=get_random_seed
-    )
-    steps: int = Field(default=30, gt=0, description="The number of steps to use to generate the image")
-    width: int = Field(
-        default=512,
-        multiple_of=8,
-        gt=0,
-        description="The width of the resulting image",
-    )
-    height: int = Field(
-        default=512,
-        multiple_of=8,
-        gt=0,
-        description="The height of the resulting image",
-    )
-    cfg_scale: float = Field(
-        default=7.5,
-        ge=1,
-        description="The Classifier-Free Guidance, higher values may result in a result closer to the prompt",
-    )
-    scheduler: SAMPLER_NAME_VALUES = Field(default="euler", description="The scheduler to use")
-    unet: UNetField = Field(default=None, description="UNet model")
-    vae: VaeField = Field(default=None, description="Vae model")
-
-    # Inputs
-    image: Optional[ImageField] = Field(description="The input image")
-    strength: float = Field(default=0.75, gt=0, le=1, description="The strength of the original image")
-    fit: bool = Field(
-        default=True,
-        description="Whether or not the result should be fit to the aspect ratio of the input image",
-    )
-
-    # Inputs
-    mask: Optional[ImageField] = Field(description="The mask")
-    seam_size: int = Field(default=96, ge=1, description="The seam inpaint size (px)")
-    seam_blur: int = Field(default=16, ge=0, description="The seam inpaint blur radius (px)")
-    seam_strength: float = Field(default=0.75, gt=0, le=1, description="The seam inpaint strength")
-    seam_steps: int = Field(default=30, ge=1, description="The number of steps to use for seam inpaint")
-    tile_size: int = Field(default=32, ge=1, description="The tile infill method size (px)")
-    infill_method: INFILL_METHODS = Field(
-        default=DEFAULT_INFILL_METHOD,
-        description="The method used to infill empty regions (px)",
-    )
-    inpaint_width: Optional[int] = Field(
-        default=None,
-        multiple_of=8,
-        gt=0,
-        description="The width of the inpaint region (px)",
-    )
-    inpaint_height: Optional[int] = Field(
-        default=None,
-        multiple_of=8,
-        gt=0,
-        description="The height of the inpaint region (px)",
-    )
-    inpaint_fill: Optional[ColorField] = Field(
-        default=ColorField(r=127, g=127, b=127, a=255),
-        description="The solid infill method color",
-    )
-    inpaint_replace: float = Field(
-        default=0.0,
-        ge=0.0,
-        le=1.0,
-        description="The amount by which to replace masked areas with latent noise",
-    )
-
-    # Schema customisation
-    class Config(InvocationConfig):
-        schema_extra = {
-            "ui": {"tags": ["stable-diffusion", "image"], "title": "Inpaint"},
-        }
-
-    def dispatch_progress(
-        self,
-        context: InvocationContext,
-        source_node_id: str,
-        intermediate_state: PipelineIntermediateState,
-    ) -> None:
-        stable_diffusion_step_callback(
-            context=context,
-            intermediate_state=intermediate_state,
-            node=self.dict(),
-            source_node_id=source_node_id,
-        )
-
-    def get_conditioning(self, context, unet):
-        positive_cond_data = context.services.latents.get(self.positive_conditioning.conditioning_name)
-        c = positive_cond_data.conditionings[0].embeds.to(device=unet.device, dtype=unet.dtype)
-        extra_conditioning_info = positive_cond_data.conditionings[0].extra_conditioning
-
-        negative_cond_data = context.services.latents.get(self.negative_conditioning.conditioning_name)
-        uc = negative_cond_data.conditionings[0].embeds.to(device=unet.device, dtype=unet.dtype)
-
-        return (uc, c, extra_conditioning_info)
-
-    @contextmanager
-    def load_model_old_way(self, context, scheduler):
-        def _lora_loader():
-            for lora in self.unet.loras:
-                lora_info = context.services.model_manager.get_model(
-                    **lora.dict(exclude={"weight"}),
-                    context=context,
-                )
-                yield (lora_info.context.model, lora.weight)
-                del lora_info
-            return
-
-        unet_info = context.services.model_manager.get_model(
-            **self.unet.unet.dict(),
-            context=context,
-        )
-        vae_info = context.services.model_manager.get_model(
-            **self.vae.vae.dict(),
-            context=context,
-        )
-
-        with vae_info as vae, ModelPatcher.apply_lora_unet(unet_info.context.model, _lora_loader()), unet_info as unet:
-            device = context.services.model_manager.mgr.cache.execution_device
-            dtype = context.services.model_manager.mgr.cache.precision
-
-            vae.to(dtype=unet.dtype)
-
-            pipeline = StableDiffusionGeneratorPipeline(
-                vae=vae,
-                text_encoder=None,
-                tokenizer=None,
-                unet=unet,
-                scheduler=scheduler,
-                safety_checker=None,
-                feature_extractor=None,
-                requires_safety_checker=False,
-            )
-
-            yield OldModelInfo(
-                name=self.unet.unet.model_name,
-                hash="<NO-HASH>",
-                model=pipeline,
-            )
-
-    def invoke(self, context: InvocationContext) -> ImageOutput:
-        image = None if self.image is None else context.services.images.get_pil_image(self.image.image_name)
-        mask = None if self.mask is None else context.services.images.get_pil_image(self.mask.image_name)
-
-        # Get the source node id (we are invoking the prepared node)
-        graph_execution_state = context.services.graph_execution_manager.get(context.graph_execution_state_id)
-        source_node_id = graph_execution_state.prepared_source_mapping[self.id]
-
-        scheduler = get_scheduler(
-            context=context,
-            scheduler_info=self.unet.scheduler,
-            scheduler_name=self.scheduler,
-        )
-
-        with self.load_model_old_way(context, scheduler) as model:
-            conditioning = self.get_conditioning(context, model.context.model.unet)
-
-            outputs = Inpaint(model).generate(
-                conditioning=conditioning,
-                scheduler=scheduler,
-                init_image=image,
-                mask_image=mask,
-                step_callback=partial(self.dispatch_progress, context, source_node_id),
-                **self.dict(
-                    exclude={"positive_conditioning", "negative_conditioning", "scheduler", "image", "mask"}
-                ),  # Shorthand for passing all of the parameters above manually
-            )
-
-            # Outputs is an infinite iterator that will return a new InvokeAIGeneratorOutput object
-            # each time it is called. We only need the first one.
-            generator_output = next(outputs)
-
-            image_dto = context.services.images.create(
-                image=generator_output.image,
-                image_origin=ResourceOrigin.INTERNAL,
-                image_category=ImageCategory.GENERAL,
-                session_id=context.graph_execution_state_id,
-                node_id=self.id,
-                is_intermediate=self.is_intermediate,
-            )
-
-            return ImageOutput(
-                image=ImageField(image_name=image_dto.image_name),
-                width=image_dto.width,
-                height=image_dto.height,
-            )

invokeai/app/invocations/image.py

@@ -1,70 +1,31 @@
 # Copyright (c) 2022 Kyle Schouviller (https://github.com/kyle0654)
 
+from pathlib import Path
 from typing import Literal, Optional
 
-import numpy
 import cv2
-from PIL import Image, ImageFilter, ImageOps, ImageChops
-from pydantic import Field
-from pathlib import Path
-from typing import Union
+import numpy
+from PIL import Image, ImageChops, ImageFilter, ImageOps
+
 from invokeai.app.invocations.metadata import CoreMetadata
-from ..models.image import (
-    ImageCategory,
-    ImageField,
-    ResourceOrigin,
-    PILInvocationConfig,
-    ImageOutput,
-    MaskOutput,
-)
-from .baseinvocation import (
-    BaseInvocation,
-    InvocationContext,
-    InvocationConfig,
-)
-from invokeai.backend.image_util.safety_checker import SafetyChecker
+from invokeai.app.invocations.primitives import ImageField, ImageOutput
 from invokeai.backend.image_util.invisible_watermark import InvisibleWatermark
+from invokeai.backend.image_util.safety_checker import SafetyChecker
+
+from ..models.image import ImageCategory, ResourceOrigin
+from .baseinvocation import BaseInvocation, FieldDescriptions, InputField, InvocationContext, tags, title
 
 
-class LoadImageInvocation(BaseInvocation):
-    """Load an image and provide it as output."""
-
-    # fmt: off
-    type: Literal["load_image"] = "load_image"
-
-    # Inputs
-    image: Optional[ImageField] = Field(
-        default=None, description="The image to load"
-    )
-    # fmt: on
-
-    class Config(InvocationConfig):
-        schema_extra = {
-            "ui": {"title": "Load Image", "tags": ["image", "load"]},
-        }
-
-    def invoke(self, context: InvocationContext) -> ImageOutput:
-        image = context.services.images.get_pil_image(self.image.image_name)
-
-        return ImageOutput(
-            image=ImageField(image_name=self.image.image_name),
-            width=image.width,
-            height=image.height,
-        )
-
-
+@title("Show Image")
+@tags("image")
 class ShowImageInvocation(BaseInvocation):
     """Displays a provided image, and passes it forward in the pipeline."""
 
+    # Metadata
     type: Literal["show_image"] = "show_image"
 
     # Inputs
-    image: Optional[ImageField] = Field(default=None, description="The image to show")
-
-    class Config(InvocationConfig):
-        schema_extra = {
-            "ui": {"title": "Show Image", "tags": ["image", "show"]},
-        }
+    image: ImageField = InputField(description="The image to show")
 
     def invoke(self, context: InvocationContext) -> ImageOutput:
         image = context.services.images.get_pil_image(self.image.image_name)
@@ -80,24 +41,20 @@ class ShowImageInvocation(BaseInvocation):
         )
 
 
-class ImageCropInvocation(BaseInvocation, PILInvocationConfig):
+@title("Crop Image")
+@tags("image", "crop")
+class ImageCropInvocation(BaseInvocation):
     """Crops an image to a specified box. The box can be outside of the image."""
 
-    # fmt: off
+    # Metadata
     type: Literal["img_crop"] = "img_crop"
 
     # Inputs
-    image: Optional[ImageField]  = Field(default=None, description="The image to crop")
-    x:      int = Field(default=0, description="The left x coordinate of the crop rectangle")
-    y:      int = Field(default=0, description="The top y coordinate of the crop rectangle")
-    width:  int = Field(default=512, gt=0, description="The width of the crop rectangle")
-    height: int = Field(default=512, gt=0, description="The height of the crop rectangle")
-    # fmt: on
-
-    class Config(InvocationConfig):
-        schema_extra = {
-            "ui": {"title": "Crop Image", "tags": ["image", "crop"]},
-        }
+    image: ImageField = InputField(description="The image to crop")
+    x: int = InputField(default=0, description="The left x coordinate of the crop rectangle")
+    y: int = InputField(default=0, description="The top y coordinate of the crop rectangle")
+    width: int = InputField(default=512, gt=0, description="The width of the crop rectangle")
+    height: int = InputField(default=512, gt=0, description="The height of the crop rectangle")
 
     def invoke(self, context: InvocationContext) -> ImageOutput:
         image = context.services.images.get_pil_image(self.image.image_name)
@@ -121,31 +78,31 @@ class ImageCropInvocation(BaseInvocation, PILInvocationConfig):
         )
 
 
-class ImagePasteInvocation(BaseInvocation, PILInvocationConfig):
+@title("Paste Image")
+@tags("image", "paste")
+class ImagePasteInvocation(BaseInvocation):
     """Pastes an image into another image."""
 
-    # fmt: off
+    # Metadata
     type: Literal["img_paste"] = "img_paste"
 
     # Inputs
-    base_image:     Optional[ImageField]  = Field(default=None, description="The base image")
-    image:          Optional[ImageField]  = Field(default=None, description="The image to paste")
-    mask: Optional[ImageField] = Field(default=None, description="The mask to use when pasting")
-    x:                     int = Field(default=0, description="The left x coordinate at which to paste the image")
-    y:                     int = Field(default=0, description="The top y coordinate at which to paste the image")
-    # fmt: on
-
-    class Config(InvocationConfig):
-        schema_extra = {
-            "ui": {"title": "Paste Image", "tags": ["image", "paste"]},
-        }
+    base_image: ImageField = InputField(description="The base image")
+    image: ImageField = InputField(description="The image to paste")
+    mask: Optional[ImageField] = InputField(
+        default=None,
+        description="The mask to use when pasting",
+    )
+    x: int = InputField(default=0, description="The left x coordinate at which to paste the image")
+    y: int = InputField(default=0, description="The top y coordinate at which to paste the image")
 
     def invoke(self, context: InvocationContext) -> ImageOutput:
         base_image = context.services.images.get_pil_image(self.base_image.image_name)
         image = context.services.images.get_pil_image(self.image.image_name)
-        mask = (
-            None if self.mask is None else ImageOps.invert(context.services.images.get_pil_image(self.mask.image_name))
-        )
+        mask = None
+        if self.mask is not None:
+            mask = context.services.images.get_pil_image(self.mask.image_name)
+            mask = ImageOps.invert(mask.convert("L"))
         # TODO: probably shouldn't invert mask here... should user be required to do it?
 
         min_x = min(0, self.x)
@@ -173,23 +130,19 @@ class ImagePasteInvocation(BaseInvocation, PILInvocationConfig):
         )
 
 
-class MaskFromAlphaInvocation(BaseInvocation, PILInvocationConfig):
+@title("Mask from Alpha")
+@tags("image", "mask")
+class MaskFromAlphaInvocation(BaseInvocation):
     """Extracts the alpha channel of an image as a mask."""
 
-    # fmt: off
+    # Metadata
     type: Literal["tomask"] = "tomask"
 
     # Inputs
-    image: Optional[ImageField]  = Field(default=None, description="The image to create the mask from")
-    invert:      bool = Field(default=False, description="Whether or not to invert the mask")
-    # fmt: on
+    image: ImageField = InputField(description="The image to create the mask from")
+    invert: bool = InputField(default=False, description="Whether or not to invert the mask")
 
-    class Config(InvocationConfig):
-        schema_extra = {
-            "ui": {"title": "Mask From Alpha", "tags": ["image", "mask", "alpha"]},
-        }
-
-    def invoke(self, context: InvocationContext) -> MaskOutput:
+    def invoke(self, context: InvocationContext) -> ImageOutput:
         image = context.services.images.get_pil_image(self.image.image_name)
 
         image_mask = image.split()[-1]
@@ -205,28 +158,24 @@ class MaskFromAlphaInvocation(BaseInvocation, PILInvocationConfig):
             is_intermediate=self.is_intermediate,
         )
 
-        return MaskOutput(
-            mask=ImageField(image_name=image_dto.image_name),
+        return ImageOutput(
+            image=ImageField(image_name=image_dto.image_name),
             width=image_dto.width,
             height=image_dto.height,
         )
 
 
-class ImageMultiplyInvocation(BaseInvocation, PILInvocationConfig):
+@title("Multiply Images")
+@tags("image", "multiply")
+class ImageMultiplyInvocation(BaseInvocation):
     """Multiplies two images together using `PIL.ImageChops.multiply()`."""
 
-    # fmt: off
+    # Metadata
     type: Literal["img_mul"] = "img_mul"
 
     # Inputs
-    image1: Optional[ImageField]  = Field(default=None, description="The first image to multiply")
-    image2: Optional[ImageField]  = Field(default=None, description="The second image to multiply")
-    # fmt: on
-
-    class Config(InvocationConfig):
-        schema_extra = {
-            "ui": {"title": "Multiply Images", "tags": ["image", "multiply"]},
-        }
+    image1: ImageField = InputField(description="The first image to multiply")
+    image2: ImageField = InputField(description="The second image to multiply")
 
     def invoke(self, context: InvocationContext) -> ImageOutput:
         image1 = context.services.images.get_pil_image(self.image1.image_name)
@@ -253,21 +202,17 @@ class ImageMultiplyInvocation(BaseInvocation, PILInvocationConfig):
 IMAGE_CHANNELS = Literal["A", "R", "G", "B"]
 
 
-class ImageChannelInvocation(BaseInvocation, PILInvocationConfig):
+@title("Extract Image Channel")
+@tags("image", "channel")
+class ImageChannelInvocation(BaseInvocation):
     """Gets a channel from an image."""
 
-    # fmt: off
+    # Metadata
     type: Literal["img_chan"] = "img_chan"
 
     # Inputs
-    image: Optional[ImageField]  = Field(default=None, description="The image to get the channel from")
-    channel: IMAGE_CHANNELS  = Field(default="A", description="The channel to get")
-    # fmt: on
-
-    class Config(InvocationConfig):
-        schema_extra = {
-            "ui": {"title": "Image Channel", "tags": ["image", "channel"]},
-        }
+    image: ImageField = InputField(description="The image to get the channel from")
+    channel: IMAGE_CHANNELS = InputField(default="A", description="The channel to get")
 
     def invoke(self, context: InvocationContext) -> ImageOutput:
         image = context.services.images.get_pil_image(self.image.image_name)
@@ -293,21 +238,17 @@ class ImageChannelInvocation(BaseInvocation, PILInvocationConfig):
 IMAGE_MODES = Literal["L", "RGB", "RGBA", "CMYK", "YCbCr", "LAB", "HSV", "I", "F"]
 
 
-class ImageConvertInvocation(BaseInvocation, PILInvocationConfig):
+@title("Convert Image Mode")
+@tags("image", "convert")
+class ImageConvertInvocation(BaseInvocation):
     """Converts an image to a different mode."""
 
-    # fmt: off
+    # Metadata
     type: Literal["img_conv"] = "img_conv"
 
     # Inputs
-    image: Optional[ImageField]  = Field(default=None, description="The image to convert")
-    mode: IMAGE_MODES  = Field(default="L", description="The mode to convert to")
-    # fmt: on
-
-    class Config(InvocationConfig):
-        schema_extra = {
-            "ui": {"title": "Convert Image", "tags": ["image", "convert"]},
-        }
+    image: ImageField = InputField(description="The image to convert")
+    mode: IMAGE_MODES = InputField(default="L", description="The mode to convert to")
 
     def invoke(self, context: InvocationContext) -> ImageOutput:
         image = context.services.images.get_pil_image(self.image.image_name)
@@ -330,22 +271,19 @@ class ImageConvertInvocation(BaseInvocation, PILInvocationConfig):
         )
 
 
-class ImageBlurInvocation(BaseInvocation, PILInvocationConfig):
+@title("Blur Image")
+@tags("image", "blur")
+class ImageBlurInvocation(BaseInvocation):
     """Blurs an image"""
 
-    # fmt: off
+    # Metadata
     type: Literal["img_blur"] = "img_blur"
 
     # Inputs
-    image: Optional[ImageField]  = Field(default=None, description="The image to blur")
-    radius:     float = Field(default=8.0, ge=0, description="The blur radius")
-    blur_type: Literal["gaussian", "box"] = Field(default="gaussian", description="The type of blur")
-    # fmt: on
-
-    class Config(InvocationConfig):
-        schema_extra = {
-            "ui": {"title": "Blur Image", "tags": ["image", "blur"]},
-        }
+    image: ImageField = InputField(description="The image to blur")
+    radius: float = InputField(default=8.0, ge=0, description="The blur radius")
+    # Metadata
+    blur_type: Literal["gaussian", "box"] = InputField(default="gaussian", description="The type of blur")
 
     def invoke(self, context: InvocationContext) -> ImageOutput:
         image = context.services.images.get_pil_image(self.image.image_name)
@@ -391,23 +329,19 @@ PIL_RESAMPLING_MAP = {
 }
 
 
-class ImageResizeInvocation(BaseInvocation, PILInvocationConfig):
+@title("Resize Image")
+@tags("image", "resize")
+class ImageResizeInvocation(BaseInvocation):
     """Resizes an image to specific dimensions"""
 
-    # fmt: off
+    # Metadata
     type: Literal["img_resize"] = "img_resize"
 
     # Inputs
-    image: Optional[ImageField]  = Field(default=None, description="The image to resize")
-    width:                         Union[int, None] = Field(ge=64, multiple_of=8, description="The width to resize to (px)")
-    height:                        Union[int, None] = Field(ge=64, multiple_of=8, description="The height to resize to (px)")
-    resample_mode:  PIL_RESAMPLING_MODES = Field(default="bicubic", description="The resampling mode")
-    # fmt: on
-
-    class Config(InvocationConfig):
-        schema_extra = {
-            "ui": {"title": "Resize Image", "tags": ["image", "resize"]},
-        }
+    image: ImageField = InputField(description="The image to resize")
+    width: int = InputField(default=512, ge=64, multiple_of=8, description="The width to resize to (px)")
+    height: int = InputField(default=512, ge=64, multiple_of=8, description="The height to resize to (px)")
+    resample_mode: PIL_RESAMPLING_MODES = InputField(default="bicubic", description="The resampling mode")
 
     def invoke(self, context: InvocationContext) -> ImageOutput:
         image = context.services.images.get_pil_image(self.image.image_name)
@@ -435,22 +369,22 @@ class ImageResizeInvocation(BaseInvocation, PILInvocationConfig):
         )
 
 
-class ImageScaleInvocation(BaseInvocation, PILInvocationConfig):
+@title("Scale Image")
+@tags("image", "scale")
+class ImageScaleInvocation(BaseInvocation):
     """Scales an image by a factor"""
 
-    # fmt: off
+    # Metadata
     type: Literal["img_scale"] = "img_scale"
 
     # Inputs
-    image:          Optional[ImageField] = Field(default=None, description="The image to scale")
-    scale_factor:        Optional[float] = Field(default=2.0, gt=0, description="The factor by which to scale the image")
-    resample_mode:  PIL_RESAMPLING_MODES = Field(default="bicubic", description="The resampling mode")
-    # fmt: on
-
-    class Config(InvocationConfig):
-        schema_extra = {
-            "ui": {"title": "Scale Image", "tags": ["image", "scale"]},
-        }
+    image: ImageField = InputField(description="The image to scale")
+    scale_factor: float = InputField(
+        default=2.0,
+        gt=0,
+        description="The factor by which to scale the image",
+    )
+    resample_mode: PIL_RESAMPLING_MODES = InputField(default="bicubic", description="The resampling mode")
 
     def invoke(self, context: InvocationContext) -> ImageOutput:
         image = context.services.images.get_pil_image(self.image.image_name)
@@ -480,22 +414,18 @@ class ImageScaleInvocation(BaseInvocation, PILInvocationConfig):
         )
 
 
-class ImageLerpInvocation(BaseInvocation, PILInvocationConfig):
+@title("Lerp Image")
+@tags("image", "lerp")
+class ImageLerpInvocation(BaseInvocation):
     """Linear interpolation of all pixels of an image"""
 
-    # fmt: off
+    # Metadata
     type: Literal["img_lerp"] = "img_lerp"
 
     # Inputs
-    image: Optional[ImageField]  = Field(default=None, description="The image to lerp")
-    min: int = Field(default=0, ge=0, le=255, description="The minimum output value")
-    max: int = Field(default=255, ge=0, le=255, description="The maximum output value")
-    # fmt: on
-
-    class Config(InvocationConfig):
-        schema_extra = {
-            "ui": {"title": "Image Linear Interpolation", "tags": ["image", "linear", "interpolation", "lerp"]},
-        }
+    image: ImageField = InputField(description="The image to lerp")
+    min: int = InputField(default=0, ge=0, le=255, description="The minimum output value")
+    max: int = InputField(default=255, ge=0, le=255, description="The maximum output value")
 
     def invoke(self, context: InvocationContext) -> ImageOutput:
         image = context.services.images.get_pil_image(self.image.image_name)
@@ -521,25 +451,18 @@ class ImageLerpInvocation(BaseInvocation, PILInvocationConfig):
         )
 
 
-class ImageInverseLerpInvocation(BaseInvocation, PILInvocationConfig):
+@title("Inverse Lerp Image")
+@tags("image", "ilerp")
+class ImageInverseLerpInvocation(BaseInvocation):
     """Inverse linear interpolation of all pixels of an image"""
 
-    # fmt: off
+    # Metadata
     type: Literal["img_ilerp"] = "img_ilerp"
 
     # Inputs
-    image: Optional[ImageField]  = Field(default=None, description="The image to lerp")
-    min: int = Field(default=0, ge=0, le=255, description="The minimum input value")
-    max: int = Field(default=255, ge=0, le=255, description="The maximum input value")
-    # fmt: on
-
-    class Config(InvocationConfig):
-        schema_extra = {
-            "ui": {
-                "title": "Image Inverse Linear Interpolation",
-                "tags": ["image", "linear", "interpolation", "inverse"],
-            },
-        }
+    image: ImageField = InputField(description="The image to lerp")
+    min: int = InputField(default=0, ge=0, le=255, description="The minimum input value")
+    max: int = InputField(default=255, ge=0, le=255, description="The maximum input value")
 
     def invoke(self, context: InvocationContext) -> ImageOutput:
         image = context.services.images.get_pil_image(self.image.image_name)
@@ -565,21 +488,19 @@ class ImageInverseLerpInvocation(BaseInvocation, PILInvocationConfig):
         )
 
 
-class ImageNSFWBlurInvocation(BaseInvocation, PILInvocationConfig):
+@title("Blur NSFW Image")
+@tags("image", "nsfw")
+class ImageNSFWBlurInvocation(BaseInvocation):
     """Add blur to NSFW-flagged images"""
 
-    # fmt: off
+    # Metadata
     type: Literal["img_nsfw"] = "img_nsfw"
 
     # Inputs
-    image: Optional[ImageField]  = Field(default=None, description="The image to check")
-    metadata: Optional[CoreMetadata] = Field(default=None, description="Optional core metadata to be written to the image")    
-    # fmt: on
-
-    class Config(InvocationConfig):
-        schema_extra = {
-            "ui": {"title": "Blur NSFW Images", "tags": ["image", "nsfw", "checker"]},
-        }
+    image: ImageField = InputField(description="The image to check")
+    metadata: Optional[CoreMetadata] = InputField(
+        default=None, description=FieldDescriptions.core_metadata, ui_hidden=True
+    )
 
     def invoke(self, context: InvocationContext) -> ImageOutput:
         image = context.services.images.get_pil_image(self.image.image_name)
@@ -616,22 +537,20 @@ class ImageNSFWBlurInvocation(BaseInvocation, PILInvocationConfig):
         return caution.resize((caution.width // 2, caution.height // 2))
 
 
-class ImageWatermarkInvocation(BaseInvocation, PILInvocationConfig):
+@title("Add Invisible Watermark")
+@tags("image", "watermark")
+class ImageWatermarkInvocation(BaseInvocation):
     """Add an invisible watermark to an image"""
 
-    # fmt: off
+    # Metadata
     type: Literal["img_watermark"] = "img_watermark"
 
     # Inputs
-    image: Optional[ImageField]  = Field(default=None, description="The image to check")
-    text: str = Field(default='InvokeAI', description="Watermark text")
-    metadata: Optional[CoreMetadata] = Field(default=None, description="Optional core metadata to be written to the image")    
-    # fmt: on
-
-    class Config(InvocationConfig):
-        schema_extra = {
-            "ui": {"title": "Add Invisible Watermark", "tags": ["image", "watermark", "invisible"]},
-        }
+    image: ImageField = InputField(description="The image to check")
+    text: str = InputField(default="InvokeAI", description="Watermark text")
+    metadata: Optional[CoreMetadata] = InputField(
+        default=None, description=FieldDescriptions.core_metadata, ui_hidden=True
+    )
 
     def invoke(self, context: InvocationContext) -> ImageOutput:
         image = context.services.images.get_pil_image(self.image.image_name)
@@ -653,16 +572,205 @@ class ImageWatermarkInvocation(BaseInvocation, PILInvocationConfig):
         )
 
 
+@title("Mask Edge")
+@tags("image", "mask", "inpaint")
+class MaskEdgeInvocation(BaseInvocation):
+    """Applies an edge mask to an image"""
+
+    type: Literal["mask_edge"] = "mask_edge"
+
+    # Inputs
+    image: ImageField = InputField(description="The image to apply the mask to")
+    edge_size: int = InputField(description="The size of the edge")
+    edge_blur: int = InputField(description="The amount of blur on the edge")
+    low_threshold: int = InputField(description="First threshold for the hysteresis procedure in Canny edge detection")
+    high_threshold: int = InputField(
+        description="Second threshold for the hysteresis procedure in Canny edge detection"
+    )
+
+    def invoke(self, context: InvocationContext) -> ImageOutput:
+        mask = context.services.images.get_pil_image(self.image.image_name)
+
+        npimg = numpy.asarray(mask, dtype=numpy.uint8)
+        npgradient = numpy.uint8(255 * (1.0 - numpy.floor(numpy.abs(0.5 - numpy.float32(npimg) / 255.0) * 2.0)))
+        npedge = cv2.Canny(npimg, threshold1=self.low_threshold, threshold2=self.high_threshold)
+        npmask = npgradient + npedge
+        npmask = cv2.dilate(npmask, numpy.ones((3, 3), numpy.uint8), iterations=int(self.edge_size / 2))
+
+        new_mask = Image.fromarray(npmask)
+
+        if self.edge_blur > 0:
+            new_mask = new_mask.filter(ImageFilter.BoxBlur(self.edge_blur))
+
+        new_mask = ImageOps.invert(new_mask)
+
+        image_dto = context.services.images.create(
+            image=new_mask,
+            image_origin=ResourceOrigin.INTERNAL,
+            image_category=ImageCategory.MASK,
+            node_id=self.id,
+            session_id=context.graph_execution_state_id,
+            is_intermediate=self.is_intermediate,
+        )
+
+        return ImageOutput(
+            image=ImageField(image_name=image_dto.image_name),
+            width=image_dto.width,
+            height=image_dto.height,
+        )
+
+
+@title("Combine Mask")
+@tags("image", "mask", "multiply")
+class MaskCombineInvocation(BaseInvocation):
+    """Combine two masks together by multiplying them using `PIL.ImageChops.multiply()`."""
+
+    type: Literal["mask_combine"] = "mask_combine"
+
+    # Inputs
+    mask1: ImageField = InputField(description="The first mask to combine")
+    mask2: ImageField = InputField(description="The second image to combine")
+
+    def invoke(self, context: InvocationContext) -> ImageOutput:
+        mask1 = context.services.images.get_pil_image(self.mask1.image_name).convert("L")
+        mask2 = context.services.images.get_pil_image(self.mask2.image_name).convert("L")
+
+        combined_mask = ImageChops.multiply(mask1, mask2)
+
+        image_dto = context.services.images.create(
+            image=combined_mask,
+            image_origin=ResourceOrigin.INTERNAL,
+            image_category=ImageCategory.GENERAL,
+            node_id=self.id,
+            session_id=context.graph_execution_state_id,
+            is_intermediate=self.is_intermediate,
+        )
+
+        return ImageOutput(
+            image=ImageField(image_name=image_dto.image_name),
+            width=image_dto.width,
+            height=image_dto.height,
+        )
+
+
+@title("Color Correct")
+@tags("image", "color")
+class ColorCorrectInvocation(BaseInvocation):
+    """
+    Shifts the colors of a target image to match the reference image, optionally
+    using a mask to only color-correct certain regions of the target image.
+    """
+
+    type: Literal["color_correct"] = "color_correct"
+
+    # Inputs
+    image: ImageField = InputField(description="The image to color-correct")
+    reference: ImageField = InputField(description="Reference image for color-correction")
+    mask: Optional[ImageField] = InputField(default=None, description="Mask to use when applying color-correction")
+    mask_blur_radius: float = InputField(default=8, description="Mask blur radius")
+
+    def invoke(self, context: InvocationContext) -> ImageOutput:
+        pil_init_mask = None
+        if self.mask is not None:
+            pil_init_mask = context.services.images.get_pil_image(self.mask.image_name).convert("L")
+
+        init_image = context.services.images.get_pil_image(self.reference.image_name)
+
+        result = context.services.images.get_pil_image(self.image.image_name).convert("RGBA")
+
+        # if init_image is None or init_mask is None:
+        #    return result
+
+        # Get the original alpha channel of the mask if there is one.
+        # Otherwise it is some other black/white image format ('1', 'L' or 'RGB')
+        # pil_init_mask = (
+        #    init_mask.getchannel("A")
+        #    if init_mask.mode == "RGBA"
+        #    else init_mask.convert("L")
+        # )
+        pil_init_image = init_image.convert("RGBA")  # Add an alpha channel if one doesn't exist
+
+        # Build an image with only visible pixels from source to use as reference for color-matching.
+        init_rgb_pixels = numpy.asarray(init_image.convert("RGB"), dtype=numpy.uint8)
+        init_a_pixels = numpy.asarray(pil_init_image.getchannel("A"), dtype=numpy.uint8)
+        init_mask_pixels = numpy.asarray(pil_init_mask, dtype=numpy.uint8)
+
+        # Get numpy version of result
+        np_image = numpy.asarray(result.convert("RGB"), dtype=numpy.uint8)
+
+        # Mask and calculate mean and standard deviation
+        mask_pixels = init_a_pixels * init_mask_pixels > 0
+        np_init_rgb_pixels_masked = init_rgb_pixels[mask_pixels, :]
+        np_image_masked = np_image[mask_pixels, :]
+
+        if np_init_rgb_pixels_masked.size > 0:
+            init_means = np_init_rgb_pixels_masked.mean(axis=0)
+            init_std = np_init_rgb_pixels_masked.std(axis=0)
+            gen_means = np_image_masked.mean(axis=0)
+            gen_std = np_image_masked.std(axis=0)
+
+            # Color correct
+            np_matched_result = np_image.copy()
+            np_matched_result[:, :, :] = (
+                (
+                    (
+                        (np_matched_result[:, :, :].astype(numpy.float32) - gen_means[None, None, :])
+                        / gen_std[None, None, :]
+                    )
+                    * init_std[None, None, :]
+                    + init_means[None, None, :]
+                )
+                .clip(0, 255)
+                .astype(numpy.uint8)
+            )
+            matched_result = Image.fromarray(np_matched_result, mode="RGB")
+        else:
+            matched_result = Image.fromarray(np_image, mode="RGB")
+
+        # Blur the mask out (into init image) by specified amount
+        if self.mask_blur_radius > 0:
+            nm = numpy.asarray(pil_init_mask, dtype=numpy.uint8)
+            nmd = cv2.erode(
+                nm,
+                kernel=numpy.ones((3, 3), dtype=numpy.uint8),
+                iterations=int(self.mask_blur_radius / 2),
+            )
+            pmd = Image.fromarray(nmd, mode="L")
+            blurred_init_mask = pmd.filter(ImageFilter.BoxBlur(self.mask_blur_radius))
+        else:
+            blurred_init_mask = pil_init_mask
+
+        multiplied_blurred_init_mask = ImageChops.multiply(blurred_init_mask, result.split()[-1])
+
+        # Paste original on color-corrected generation (using blurred mask)
+        matched_result.paste(init_image, (0, 0), mask=multiplied_blurred_init_mask)
+
+        image_dto = context.services.images.create(
+            image=matched_result,
+            image_origin=ResourceOrigin.INTERNAL,
+            image_category=ImageCategory.GENERAL,
+            node_id=self.id,
+            session_id=context.graph_execution_state_id,
+            is_intermediate=self.is_intermediate,
+        )
+
+        return ImageOutput(
+            image=ImageField(image_name=image_dto.image_name),
+            width=image_dto.width,
+            height=image_dto.height,
+        )
+
+
+@title("Image Hue Adjustment")
+@tags("image", "hue", "hsl")
 class ImageHueAdjustmentInvocation(BaseInvocation):
     """Adjusts the Hue of an image."""
 
-    # fmt: off
     type: Literal["img_hue_adjust"] = "img_hue_adjust"
 
     # Inputs
-    image: ImageField = Field(default=None, description="The image to adjust")
-    hue: int = Field(default=0, description="The degrees by which to rotate the hue, 0-360")
-    # fmt: on
+    image: ImageField = InputField(description="The image to adjust")
+    hue: int = InputField(default=0, description="The degrees by which to rotate the hue, 0-360")
 
     def invoke(self, context: InvocationContext) -> ImageOutput:
         pil_image = context.services.images.get_pil_image(self.image.image_name)
@@ -697,16 +805,18 @@ class ImageHueAdjustmentInvocation(BaseInvocation):
         )
 
 
+@title("Image Luminosity Adjustment")
+@tags("image", "luminosity", "hsl")
 class ImageLuminosityAdjustmentInvocation(BaseInvocation):
     """Adjusts the Luminosity (Value) of an image."""
 
-    # fmt: off
     type: Literal["img_luminosity_adjust"] = "img_luminosity_adjust"
 
     # Inputs
-    image: ImageField = Field(default=None, description="The image to adjust")
-    luminosity: float = Field(default=1.0, ge=0, le=1, description="The factor by which to adjust the luminosity (value)")
-    # fmt: on
+    image: ImageField = InputField(description="The image to adjust")
+    luminosity: float = InputField(
+        default=1.0, ge=0, le=1, description="The factor by which to adjust the luminosity (value)"
+    )
 
     def invoke(self, context: InvocationContext) -> ImageOutput:
         pil_image = context.services.images.get_pil_image(self.image.image_name)
@@ -745,16 +855,16 @@ class ImageLuminosityAdjustmentInvocation(BaseInvocation):
         )
 
 
+@title("Image Saturation Adjustment")
+@tags("image", "saturation", "hsl")
 class ImageSaturationAdjustmentInvocation(BaseInvocation):
     """Adjusts the Saturation of an image."""
 
-    # fmt: off
     type: Literal["img_saturation_adjust"] = "img_saturation_adjust"
 
     # Inputs
-    image: ImageField = Field(default=None, description="The image to adjust")
-    saturation: float = Field(default=1.0, ge=0, le=1, description="The factor by which to adjust the saturation")
-    # fmt: on
+    image: ImageField = InputField(description="The image to adjust")
+    saturation: float = InputField(default=1.0, ge=0, le=1, description="The factor by which to adjust the saturation")
 
     def invoke(self, context: InvocationContext) -> ImageOutput:
         pil_image = context.services.images.get_pil_image(self.image.image_name)

invokeai/app/invocations/infill.py

@@ -5,18 +5,13 @@ from typing import Literal, Optional, get_args
 import numpy as np
 import math
 from PIL import Image, ImageOps
-from pydantic import Field
+from invokeai.app.invocations.primitives import ImageField, ImageOutput, ColorField
 
-from invokeai.app.invocations.image import ImageOutput
 from invokeai.app.util.misc import SEED_MAX, get_random_seed
 from invokeai.backend.image_util.patchmatch import PatchMatch
 
-from ..models.image import ColorField, ImageCategory, ImageField, ResourceOrigin
-from .baseinvocation import (
-    BaseInvocation,
-    InvocationConfig,
-    InvocationContext,
-)
+from ..models.image import ImageCategory, ResourceOrigin
+from .baseinvocation import BaseInvocation, InputField, InvocationContext, title, tags
 
 
 def infill_methods() -> list[str]:
@@ -114,21 +109,20 @@ def tile_fill_missing(im: Image.Image, tile_size: int = 16, seed: Optional[int]
     return si
 
 
+@title("Solid Color Infill")
+@tags("image", "inpaint")
 class InfillColorInvocation(BaseInvocation):
     """Infills transparent areas of an image with a solid color"""
 
     type: Literal["infill_rgba"] = "infill_rgba"
-    image: Optional[ImageField] = Field(default=None, description="The image to infill")
-    color: ColorField = Field(
+
+    # Inputs
+    image: ImageField = InputField(description="The image to infill")
+    color: ColorField = InputField(
         default=ColorField(r=127, g=127, b=127, a=255),
         description="The color to use to infill",
     )
 
-    class Config(InvocationConfig):
-        schema_extra = {
-            "ui": {"title": "Color Infill", "tags": ["image", "inpaint", "color", "infill"]},
-        }
-
     def invoke(self, context: InvocationContext) -> ImageOutput:
         image = context.services.images.get_pil_image(self.image.image_name)
 
@@ -153,25 +147,23 @@ class InfillColorInvocation(BaseInvocation):
         )
 
 
+@title("Tile Infill")
+@tags("image", "inpaint")
 class InfillTileInvocation(BaseInvocation):
     """Infills transparent areas of an image with tiles of the image"""
 
     type: Literal["infill_tile"] = "infill_tile"
 
-    image: Optional[ImageField] = Field(default=None, description="The image to infill")
-    tile_size: int = Field(default=32, ge=1, description="The tile size (px)")
-    seed: int = Field(
+    # Input
+    image: ImageField = InputField(description="The image to infill")
+    tile_size: int = InputField(default=32, ge=1, description="The tile size (px)")
+    seed: int = InputField(
         ge=0,
         le=SEED_MAX,
         description="The seed to use for tile generation (omit for random)",
         default_factory=get_random_seed,
     )
 
-    class Config(InvocationConfig):
-        schema_extra = {
-            "ui": {"title": "Tile Infill", "tags": ["image", "inpaint", "tile", "infill"]},
-        }
-
     def invoke(self, context: InvocationContext) -> ImageOutput:
         image = context.services.images.get_pil_image(self.image.image_name)
 
@@ -194,17 +186,15 @@ class InfillTileInvocation(BaseInvocation):
         )
 
 
+@title("PatchMatch Infill")
+@tags("image", "inpaint")
 class InfillPatchMatchInvocation(BaseInvocation):
     """Infills transparent areas of an image using the PatchMatch algorithm"""
 
     type: Literal["infill_patchmatch"] = "infill_patchmatch"
 
-    image: Optional[ImageField] = Field(default=None, description="The image to infill")
-
-    class Config(InvocationConfig):
-        schema_extra = {
-            "ui": {"title": "Patch Match Infill", "tags": ["image", "inpaint", "patchmatch", "infill"]},
-        }
+    # Inputs
+    image: ImageField = InputField(description="The image to infill")
 
     def invoke(self, context: InvocationContext) -> ImageOutput:
         image = context.services.images.get_pil_image(self.image.image_name)

invokeai/app/invocations/latent.py

@@ -5,6 +5,7 @@ from typing import List, Literal, Optional, Union
 
 import einops
 import torch
+import torchvision.transforms as T
 from diffusers.image_processor import VaeImageProcessor
 from diffusers.models.attention_processor import (
     AttnProcessor2_0,
@@ -12,20 +13,25 @@ from diffusers.models.attention_processor import (
     LoRAXFormersAttnProcessor,
     XFormersAttnProcessor,
 )
+from diffusers.schedulers import DPMSolverSDEScheduler
 from diffusers.schedulers import SchedulerMixin as Scheduler
 from pydantic import BaseModel, Field, validator
+from torchvision.transforms.functional import resize as tv_resize
 
 from invokeai.app.invocations.metadata import CoreMetadata
+from invokeai.app.invocations.primitives import (
+    ImageField,
+    ImageOutput,
+    LatentsField,
+    LatentsOutput,
+    build_latents_output,
+)
 from invokeai.app.util.controlnet_utils import prepare_control_image
 from invokeai.app.util.step_callback import stable_diffusion_step_callback
 from invokeai.backend.model_management.models import ModelType, SilenceWarnings
-from .baseinvocation import BaseInvocation, BaseInvocationOutput, InvocationConfig, InvocationContext
-from .compel import ConditioningField
-from .controlnet_image_processors import ControlField
-from .image import ImageOutput
-from .model import ModelInfo, UNetField, VaeField
-from ..models.image import ImageCategory, ImageField, ResourceOrigin
-from ...backend.model_management import ModelPatcher
+
+from ...backend.model_management import BaseModelType, ModelPatcher
+from ...backend.model_management.lora import ModelPatcher
 from ...backend.stable_diffusion import PipelineIntermediateState
 from ...backend.stable_diffusion.diffusers_pipeline import (
     ConditioningData,
@@ -35,41 +41,27 @@ from ...backend.stable_diffusion.diffusers_pipeline import (
 )
 from ...backend.stable_diffusion.diffusion.shared_invokeai_diffusion import PostprocessingSettings
 from ...backend.stable_diffusion.schedulers import SCHEDULER_MAP
-from ...backend.util.devices import choose_torch_device, torch_dtype, choose_precision
+from ...backend.util.devices import choose_precision, choose_torch_device
+from ..models.image import ImageCategory, ResourceOrigin
+from .baseinvocation import (
+    BaseInvocation,
+    BaseInvocationOutput,
+    FieldDescriptions,
+    Input,
+    InputField,
+    InvocationContext,
+    OutputField,
+    UIType,
+    tags,
+    title,
+)
+from .compel import ConditioningField
+from .controlnet_image_processors import ControlField
+from .model import ModelInfo, UNetField, VaeField
 
 DEFAULT_PRECISION = choose_precision(choose_torch_device())
 
 
-class LatentsField(BaseModel):
-    """A latents field used for passing latents between invocations"""
-
-    latents_name: Optional[str] = Field(default=None, description="The name of the latents")
-
-    class Config:
-        schema_extra = {"required": ["latents_name"]}
-
-
-class LatentsOutput(BaseInvocationOutput):
-    """Base class for invocations that output latents"""
-
-    # fmt: off
-    type: Literal["latents_output"] = "latents_output"
-
-    # Inputs
-    latents: LatentsField          = Field(default=None, description="The output latents")
-    width:                     int = Field(description="The width of the latents in pixels")
-    height:                    int = Field(description="The height of the latents in pixels")
-    # fmt: on
-
-
-def build_latents_output(latents_name: str, latents: torch.Tensor):
-    return LatentsOutput(
-        latents=LatentsField(latents_name=latents_name),
-        width=latents.size()[3] * 8,
-        height=latents.size()[2] * 8,
-    )
-
-
 SAMPLER_NAME_VALUES = Literal[tuple(list(SCHEDULER_MAP.keys()))]
 
 
@@ -77,6 +69,7 @@ def get_scheduler(
     context: InvocationContext,
     scheduler_info: ModelInfo,
     scheduler_name: str,
+    seed: int,
 ) -> Scheduler:
     scheduler_class, scheduler_extra_config = SCHEDULER_MAP.get(scheduler_name, SCHEDULER_MAP["ddim"])
     orig_scheduler_info = context.services.model_manager.get_model(
@@ -93,6 +86,11 @@ def get_scheduler(
         **scheduler_extra_config,
         "_backup": scheduler_config,
     }
+
+    # make dpmpp_sde reproducable(seed can be passed only in initializer)
+    if scheduler_class is DPMSolverSDEScheduler:
+        scheduler_config["noise_sampler_seed"] = seed
+
     scheduler = scheduler_class.from_config(scheduler_config)
 
     # hack copied over from generate.py
@@ -101,25 +99,37 @@ def get_scheduler(
     return scheduler
 
 
-# Text to image
-class TextToLatentsInvocation(BaseInvocation):
-    """Generates latents from conditionings."""
+@title("Denoise Latents")
+@tags("latents", "denoise", "txt2img", "t2i", "t2l", "img2img", "i2i", "l2l")
+class DenoiseLatentsInvocation(BaseInvocation):
+    """Denoises noisy latents to decodable images"""
 
-    type: Literal["t2l"] = "t2l"
+    type: Literal["denoise_latents"] = "denoise_latents"
 
     # Inputs
-    # fmt: off
-    positive_conditioning: Optional[ConditioningField] = Field(description="Positive conditioning for generation")
-    negative_conditioning: Optional[ConditioningField] = Field(description="Negative conditioning for generation")
-    noise: Optional[LatentsField] = Field(description="The noise to use")
-    steps:       int = Field(default=10, gt=0, description="The number of steps to use to generate the image")
-    cfg_scale: Union[float, List[float]] = Field(default=7.5, ge=1, description="The Classifier-Free Guidance, higher values may result in a result closer to the prompt", )
-    scheduler: SAMPLER_NAME_VALUES = Field(default="euler", description="The scheduler to use" )
-    unet: UNetField = Field(default=None, description="UNet submodel")
-    control: Union[ControlField, list[ControlField]] = Field(default=None, description="The control to use")
-    # seamless:   bool = Field(default=False, description="Whether or not to generate an image that can tile without seams", )
-    # seamless_axes: str = Field(default="", description="The axes to tile the image on, 'x' and/or 'y'")
-    # fmt: on
+    positive_conditioning: ConditioningField = InputField(
+        description=FieldDescriptions.positive_cond, input=Input.Connection
+    )
+    negative_conditioning: ConditioningField = InputField(
+        description=FieldDescriptions.negative_cond, input=Input.Connection
+    )
+    noise: Optional[LatentsField] = InputField(description=FieldDescriptions.noise, input=Input.Connection)
+    steps: int = InputField(default=10, gt=0, description=FieldDescriptions.steps)
+    cfg_scale: Union[float, List[float]] = InputField(
+        default=7.5, ge=1, description=FieldDescriptions.cfg_scale, ui_type=UIType.Float
+    )
+    denoising_start: float = InputField(default=0.0, ge=0, le=1, description=FieldDescriptions.denoising_start)
+    denoising_end: float = InputField(default=1.0, ge=0, le=1, description=FieldDescriptions.denoising_end)
+    scheduler: SAMPLER_NAME_VALUES = InputField(default="euler", description=FieldDescriptions.scheduler)
+    unet: UNetField = InputField(description=FieldDescriptions.unet, input=Input.Connection)
+    control: Union[ControlField, list[ControlField]] = InputField(
+        default=None, description=FieldDescriptions.control, input=Input.Connection
+    )
+    latents: Optional[LatentsField] = InputField(description=FieldDescriptions.latents, input=Input.Connection)
+    mask: Optional[ImageField] = InputField(
+        default=None,
+        description=FieldDescriptions.mask,
+    )
 
     @validator("cfg_scale")
     def ge_one(cls, v):
@@ -133,33 +143,20 @@ class TextToLatentsInvocation(BaseInvocation):
                 raise ValueError("cfg_scale must be greater than 1")
         return v
 
-    # Schema customisation
-    class Config(InvocationConfig):
-        schema_extra = {
-            "ui": {
-                "title": "Text To Latents",
-                "tags": ["latents"],
-                "type_hints": {
-                    "model": "model",
-                    "control": "control",
-                    # "cfg_scale": "float",
-                    "cfg_scale": "number",
-                },
-            },
-        }
-
     # TODO: pass this an emitter method or something? or a session for dispatching?
     def dispatch_progress(
         self,
         context: InvocationContext,
         source_node_id: str,
         intermediate_state: PipelineIntermediateState,
+        base_model: BaseModelType,
     ) -> None:
         stable_diffusion_step_callback(
             context=context,
             intermediate_state=intermediate_state,
             node=self.dict(),
             source_node_id=source_node_id,
+            base_model=base_model,
         )
 
     def get_conditioning_data(
@@ -167,13 +164,14 @@ class TextToLatentsInvocation(BaseInvocation):
         context: InvocationContext,
         scheduler,
         unet,
+        seed,
     ) -> ConditioningData:
         positive_cond_data = context.services.latents.get(self.positive_conditioning.conditioning_name)
-        c = positive_cond_data.conditionings[0].embeds.to(device=unet.device, dtype=unet.dtype)
-        extra_conditioning_info = positive_cond_data.conditionings[0].extra_conditioning
+        c = positive_cond_data.conditionings[0].to(device=unet.device, dtype=unet.dtype)
+        extra_conditioning_info = c.extra_conditioning
 
         negative_cond_data = context.services.latents.get(self.negative_conditioning.conditioning_name)
-        uc = negative_cond_data.conditionings[0].embeds.to(device=unet.device, dtype=unet.dtype)
+        uc = negative_cond_data.conditionings[0].to(device=unet.device, dtype=unet.dtype)
 
         conditioning_data = ConditioningData(
             unconditioned_embeddings=uc,
@@ -193,7 +191,8 @@ class TextToLatentsInvocation(BaseInvocation):
             # for ddim scheduler
             eta=0.0,  # ddim_eta
             # for ancestral and sde schedulers
-            generator=torch.Generator(device=unet.device).manual_seed(0),
+            # flip all bits to have noise different from initial
+            generator=torch.Generator(device=unet.device).manual_seed(seed ^ 0xFFFFFFFF),
         )
         return conditioning_data
 
@@ -304,110 +303,83 @@ class TextToLatentsInvocation(BaseInvocation):
                 # MultiControlNetModel has been refactored out, just need list[ControlNetData]
         return control_data
 
-    @torch.no_grad()
-    def invoke(self, context: InvocationContext) -> LatentsOutput:
-        with SilenceWarnings():
-            noise = context.services.latents.get(self.noise.latents_name)
+    # original idea by https://github.com/AmericanPresidentJimmyCarter
+    # TODO: research more for second order schedulers timesteps
+    def init_scheduler(self, scheduler, device, steps, denoising_start, denoising_end):
+        num_inference_steps = steps
+        if scheduler.config.get("cpu_only", False):
+            scheduler.set_timesteps(num_inference_steps, device="cpu")
+            timesteps = scheduler.timesteps.to(device=device)
+        else:
+            scheduler.set_timesteps(num_inference_steps, device=device)
+            timesteps = scheduler.timesteps
 
-            # Get the source node id (we are invoking the prepared node)
-            graph_execution_state = context.services.graph_execution_manager.get(context.graph_execution_state_id)
-            source_node_id = graph_execution_state.prepared_source_mapping[self.id]
+        # apply denoising_start
+        t_start_val = int(round(scheduler.config.num_train_timesteps * (1 - denoising_start)))
+        t_start_idx = len(list(filter(lambda ts: ts >= t_start_val, timesteps)))
+        timesteps = timesteps[t_start_idx:]
+        if scheduler.order == 2 and t_start_idx > 0:
+            timesteps = timesteps[1:]
 
-            def step_callback(state: PipelineIntermediateState):
-                self.dispatch_progress(context, source_node_id, state)
+        # save start timestep to apply noise
+        init_timestep = timesteps[:1]
 
-            def _lora_loader():
-                for lora in self.unet.loras:
-                    lora_info = context.services.model_manager.get_model(
-                        **lora.dict(exclude={"weight"}),
-                        context=context,
-                    )
-                    yield (lora_info.context.model, lora.weight)
-                    del lora_info
-                return
+        # apply denoising_end
+        t_end_val = int(round(scheduler.config.num_train_timesteps * (1 - denoising_end)))
+        t_end_idx = len(list(filter(lambda ts: ts >= t_end_val, timesteps)))
+        if scheduler.order == 2 and t_end_idx > 0:
+            t_end_idx += 1
+        timesteps = timesteps[:t_end_idx]
 
-            unet_info = context.services.model_manager.get_model(
-                **self.unet.unet.dict(),
-                context=context,
-            )
-            with ExitStack() as exit_stack, ModelPatcher.apply_lora_unet(
-                unet_info.context.model, _lora_loader()
-            ), unet_info as unet:
-                noise = noise.to(device=unet.device, dtype=unet.dtype)
+        # calculate step count based on scheduler order
+        num_inference_steps = len(timesteps)
+        if scheduler.order == 2:
+            num_inference_steps += num_inference_steps % 2
+            num_inference_steps = num_inference_steps // 2
 
-                scheduler = get_scheduler(
-                    context=context,
-                    scheduler_info=self.unet.scheduler,
-                    scheduler_name=self.scheduler,
-                )
+        return num_inference_steps, timesteps, init_timestep
 
-                pipeline = self.create_pipeline(unet, scheduler)
-                conditioning_data = self.get_conditioning_data(context, scheduler, unet)
+    def prep_mask_tensor(self, mask, context, lantents):
+        if mask is None:
+            return None
 
-                control_data = self.prep_control_data(
-                    model=pipeline,
-                    context=context,
-                    control_input=self.control,
-                    latents_shape=noise.shape,
-                    # do_classifier_free_guidance=(self.cfg_scale >= 1.0))
-                    do_classifier_free_guidance=True,
-                    exit_stack=exit_stack,
-                )
-
-                # TODO: Verify the noise is the right size
-                result_latents, result_attention_map_saver = pipeline.latents_from_embeddings(
-                    latents=torch.zeros_like(noise, dtype=torch_dtype(unet.device)),
-                    noise=noise,
-                    num_inference_steps=self.steps,
-                    conditioning_data=conditioning_data,
-                    control_data=control_data,  # list[ControlNetData]
-                    callback=step_callback,
-                )
-
-            # https://discuss.huggingface.co/t/memory-usage-by-later-pipeline-stages/23699
-            result_latents = result_latents.to("cpu")
-            torch.cuda.empty_cache()
-
-            name = f"{context.graph_execution_state_id}__{self.id}"
-            context.services.latents.save(name, result_latents)
-            return build_latents_output(latents_name=name, latents=result_latents)
-
-
-class LatentsToLatentsInvocation(TextToLatentsInvocation):
-    """Generates latents using latents as base image."""
-
-    type: Literal["l2l"] = "l2l"
-
-    # Inputs
-    latents: Optional[LatentsField] = Field(description="The latents to use as a base image")
-    strength: float = Field(default=0.7, ge=0, le=1, description="The strength of the latents to use")
-
-    # Schema customisation
-    class Config(InvocationConfig):
-        schema_extra = {
-            "ui": {
-                "title": "Latent To Latents",
-                "tags": ["latents"],
-                "type_hints": {
-                    "model": "model",
-                    "control": "control",
-                    "cfg_scale": "number",
-                },
-            },
-        }
+        mask_image = context.services.images.get_pil_image(mask.image_name)
+        if mask_image.mode != "L":
+            # FIXME: why do we get passed an RGB image here? We can only use single-channel.
+            mask_image = mask_image.convert("L")
+        mask_tensor = image_resized_to_grid_as_tensor(mask_image, normalize=False)
+        if mask_tensor.dim() == 3:
+            mask_tensor = mask_tensor.unsqueeze(0)
+        mask_tensor = tv_resize(mask_tensor, lantents.shape[-2:], T.InterpolationMode.BILINEAR)
+        return 1 - mask_tensor
 
     @torch.no_grad()
     def invoke(self, context: InvocationContext) -> LatentsOutput:
         with SilenceWarnings():  # this quenches NSFW nag from diffusers
-            noise = context.services.latents.get(self.noise.latents_name)
-            latent = context.services.latents.get(self.latents.latents_name)
+            seed = None
+            noise = None
+            if self.noise is not None:
+                noise = context.services.latents.get(self.noise.latents_name)
+                seed = self.noise.seed
+
+            if self.latents is not None:
+                latents = context.services.latents.get(self.latents.latents_name)
+                if seed is None:
+                    seed = self.latents.seed
+            else:
+                latents = torch.zeros_like(noise)
+
+            if seed is None:
+                seed = 0
+
+            mask = self.prep_mask_tensor(self.mask, context, latents)
 
             # Get the source node id (we are invoking the prepared node)
             graph_execution_state = context.services.graph_execution_manager.get(context.graph_execution_state_id)
             source_node_id = graph_execution_state.prepared_source_mapping[self.id]
 
             def step_callback(state: PipelineIntermediateState):
-                self.dispatch_progress(context, source_node_id, state)
+                self.dispatch_progress(context, source_node_id, state, self.unet.unet.base_model)
 
             def _lora_loader():
                 for lora in self.unet.loras:
@@ -426,44 +398,48 @@ class LatentsToLatentsInvocation(TextToLatentsInvocation):
             with ExitStack() as exit_stack, ModelPatcher.apply_lora_unet(
                 unet_info.context.model, _lora_loader()
             ), unet_info as unet:
-                noise = noise.to(device=unet.device, dtype=unet.dtype)
-                latent = latent.to(device=unet.device, dtype=unet.dtype)
+                latents = latents.to(device=unet.device, dtype=unet.dtype)
+                if noise is not None:
+                    noise = noise.to(device=unet.device, dtype=unet.dtype)
+                if mask is not None:
+                    mask = mask.to(device=unet.device, dtype=unet.dtype)
 
                 scheduler = get_scheduler(
                     context=context,
                     scheduler_info=self.unet.scheduler,
                     scheduler_name=self.scheduler,
+                    seed=seed,
                 )
 
                 pipeline = self.create_pipeline(unet, scheduler)
-                conditioning_data = self.get_conditioning_data(context, scheduler, unet)
+                conditioning_data = self.get_conditioning_data(context, scheduler, unet, seed)
 
                 control_data = self.prep_control_data(
                     model=pipeline,
                     context=context,
                     control_input=self.control,
-                    latents_shape=noise.shape,
+                    latents_shape=latents.shape,
                     # do_classifier_free_guidance=(self.cfg_scale >= 1.0))
                     do_classifier_free_guidance=True,
                     exit_stack=exit_stack,
                 )
 
-                # TODO: Verify the noise is the right size
-                initial_latents = (
-                    latent if self.strength < 1.0 else torch.zeros_like(latent, device=unet.device, dtype=latent.dtype)
-                )
-
-                timesteps, _ = pipeline.get_img2img_timesteps(
-                    self.steps,
-                    self.strength,
+                num_inference_steps, timesteps, init_timestep = self.init_scheduler(
+                    scheduler,
                     device=unet.device,
+                    steps=self.steps,
+                    denoising_start=self.denoising_start,
+                    denoising_end=self.denoising_end,
                 )
 
                 result_latents, result_attention_map_saver = pipeline.latents_from_embeddings(
-                    latents=initial_latents,
+                    latents=latents,
                     timesteps=timesteps,
+                    init_timestep=init_timestep,
                     noise=noise,
-                    num_inference_steps=self.steps,
+                    seed=seed,
+                    mask=mask,
+                    num_inference_steps=num_inference_steps,
                     conditioning_data=conditioning_data,
                     control_data=control_data,  # list[ControlNetData]
                     callback=step_callback,
@@ -475,32 +451,32 @@ class LatentsToLatentsInvocation(TextToLatentsInvocation):
 
             name = f"{context.graph_execution_state_id}__{self.id}"
             context.services.latents.save(name, result_latents)
-        return build_latents_output(latents_name=name, latents=result_latents)
+        return build_latents_output(latents_name=name, latents=result_latents, seed=seed)
 
 
-# Latent to image
+@title("Latents to Image")
+@tags("latents", "image", "vae")
 class LatentsToImageInvocation(BaseInvocation):
     """Generates an image from latents."""
 
     type: Literal["l2i"] = "l2i"
 
     # Inputs
-    latents: Optional[LatentsField] = Field(description="The latents to generate an image from")
-    vae: VaeField = Field(default=None, description="Vae submodel")
-    tiled: bool = Field(default=False, description="Decode latents by overlaping tiles (less memory consumption)")
-    fp32: bool = Field(DEFAULT_PRECISION == "float32", description="Decode in full precision")
-    metadata: Optional[CoreMetadata] = Field(
-        default=None, description="Optional core metadata to be written to the image"
+    latents: LatentsField = InputField(
+        description=FieldDescriptions.latents,
+        input=Input.Connection,
+    )
+    vae: VaeField = InputField(
+        description=FieldDescriptions.vae,
+        input=Input.Connection,
+    )
+    tiled: bool = InputField(default=False, description=FieldDescriptions.tiled)
+    fp32: bool = InputField(default=DEFAULT_PRECISION == "float32", description=FieldDescriptions.fp32)
+    metadata: CoreMetadata = InputField(
+        default=None,
+        description=FieldDescriptions.core_metadata,
+        ui_hidden=True,
     )
-
-    # Schema customisation
-    class Config(InvocationConfig):
-        schema_extra = {
-            "ui": {
-                "title": "Latents To Image",
-                "tags": ["latents", "image"],
-            },
-        }
 
     @torch.no_grad()
     def invoke(self, context: InvocationContext) -> ImageOutput:
@@ -578,24 +554,30 @@ class LatentsToImageInvocation(BaseInvocation):
 LATENTS_INTERPOLATION_MODE = Literal["nearest", "linear", "bilinear", "bicubic", "trilinear", "area", "nearest-exact"]
 
 
+@title("Resize Latents")
+@tags("latents", "resize")
 class ResizeLatentsInvocation(BaseInvocation):
     """Resizes latents to explicit width/height (in pixels). Provided dimensions are floor-divided by 8."""
 
     type: Literal["lresize"] = "lresize"
 
     # Inputs
-    latents: Optional[LatentsField] = Field(description="The latents to resize")
-    width: Union[int, None] = Field(default=512, ge=64, multiple_of=8, description="The width to resize to (px)")
-    height: Union[int, None] = Field(default=512, ge=64, multiple_of=8, description="The height to resize to (px)")
-    mode: LATENTS_INTERPOLATION_MODE = Field(default="bilinear", description="The interpolation mode")
-    antialias: bool = Field(
-        default=False, description="Whether or not to antialias (applied in bilinear and bicubic modes only)"
+    latents: LatentsField = InputField(
+        description=FieldDescriptions.latents,
+        input=Input.Connection,
     )
-
-    class Config(InvocationConfig):
-        schema_extra = {
-            "ui": {"title": "Resize Latents", "tags": ["latents", "resize"]},
-        }
+    width: int = InputField(
+        ge=64,
+        multiple_of=8,
+        description=FieldDescriptions.width,
+    )
+    height: int = InputField(
+        ge=64,
+        multiple_of=8,
+        description=FieldDescriptions.width,
+    )
+    mode: LATENTS_INTERPOLATION_MODE = InputField(default="bilinear", description=FieldDescriptions.interp_mode)
+    antialias: bool = InputField(default=False, description=FieldDescriptions.torch_antialias)
 
     def invoke(self, context: InvocationContext) -> LatentsOutput:
         latents = context.services.latents.get(self.latents.latents_name)
@@ -617,26 +599,24 @@ class ResizeLatentsInvocation(BaseInvocation):
         name = f"{context.graph_execution_state_id}__{self.id}"
         # context.services.latents.set(name, resized_latents)
         context.services.latents.save(name, resized_latents)
-        return build_latents_output(latents_name=name, latents=resized_latents)
+        return build_latents_output(latents_name=name, latents=resized_latents, seed=self.latents.seed)
 
 
+@title("Scale Latents")
+@tags("latents", "resize")
 class ScaleLatentsInvocation(BaseInvocation):
     """Scales latents by a given factor."""
 
     type: Literal["lscale"] = "lscale"
 
     # Inputs
-    latents: Optional[LatentsField] = Field(description="The latents to scale")
-    scale_factor: float = Field(gt=0, description="The factor by which to scale the latents")
-    mode: LATENTS_INTERPOLATION_MODE = Field(default="bilinear", description="The interpolation mode")
-    antialias: bool = Field(
-        default=False, description="Whether or not to antialias (applied in bilinear and bicubic modes only)"
+    latents: LatentsField = InputField(
+        description=FieldDescriptions.latents,
+        input=Input.Connection,
     )
-
-    class Config(InvocationConfig):
-        schema_extra = {
-            "ui": {"title": "Scale Latents", "tags": ["latents", "scale"]},
-        }
+    scale_factor: float = InputField(gt=0, description=FieldDescriptions.scale_factor)
+    mode: LATENTS_INTERPOLATION_MODE = InputField(default="bilinear", description=FieldDescriptions.interp_mode)
+    antialias: bool = InputField(default=False, description=FieldDescriptions.torch_antialias)
 
     def invoke(self, context: InvocationContext) -> LatentsOutput:
         latents = context.services.latents.get(self.latents.latents_name)
@@ -659,25 +639,26 @@ class ScaleLatentsInvocation(BaseInvocation):
         name = f"{context.graph_execution_state_id}__{self.id}"
         # context.services.latents.set(name, resized_latents)
         context.services.latents.save(name, resized_latents)
-        return build_latents_output(latents_name=name, latents=resized_latents)
+        return build_latents_output(latents_name=name, latents=resized_latents, seed=self.latents.seed)
 
 
+@title("Image to Latents")
+@tags("latents", "image", "vae")
 class ImageToLatentsInvocation(BaseInvocation):
     """Encodes an image into latents."""
 
     type: Literal["i2l"] = "i2l"
 
     # Inputs
-    image: Optional[ImageField] = Field(description="The image to encode")
-    vae: VaeField = Field(default=None, description="Vae submodel")
-    tiled: bool = Field(default=False, description="Encode latents by overlaping tiles(less memory consumption)")
-    fp32: bool = Field(DEFAULT_PRECISION == "float32", description="Decode in full precision")
-
-    # Schema customisation
-    class Config(InvocationConfig):
-        schema_extra = {
-            "ui": {"title": "Image To Latents", "tags": ["latents", "image"]},
-        }
+    image: ImageField = InputField(
+        description="The image to encode",
+    )
+    vae: VaeField = InputField(
+        description=FieldDescriptions.vae,
+        input=Input.Connection,
+    )
+    tiled: bool = InputField(default=False, description=FieldDescriptions.tiled)
+    fp32: bool = InputField(default=DEFAULT_PRECISION == "float32", description=FieldDescriptions.fp32)
 
     @torch.no_grad()
     def invoke(self, context: InvocationContext) -> LatentsOutput:
@@ -740,4 +721,4 @@ class ImageToLatentsInvocation(BaseInvocation):
         name = f"{context.graph_execution_state_id}__{self.id}"
         latents = latents.to("cpu")
         context.services.latents.save(name, latents)
-        return build_latents_output(latents_name=name, latents=latents)
+        return build_latents_output(latents_name=name, latents=latents, seed=None)

invokeai/app/invocations/math.py

@@ -2,134 +2,83 @@
 
 from typing import Literal
 
-from pydantic import BaseModel, Field
 import numpy as np
 
-from .baseinvocation import (
-    BaseInvocation,
-    BaseInvocationOutput,
-    InvocationContext,
-    InvocationConfig,
-)
+from invokeai.app.invocations.primitives import IntegerOutput
+
+from .baseinvocation import BaseInvocation, FieldDescriptions, InputField, InvocationContext, tags, title
 
 
-class MathInvocationConfig(BaseModel):
-    """Helper class to provide all math invocations with additional config"""
-
-    # Schema customisation
-    class Config(InvocationConfig):
-        schema_extra = {
-            "ui": {
-                "tags": ["math"],
-            }
-        }
-
-
-class IntOutput(BaseInvocationOutput):
-    """An integer output"""
-
-    # fmt: off
-    type: Literal["int_output"] = "int_output"
-    a: int = Field(default=None, description="The output integer")
-    # fmt: on
-
-
-class FloatOutput(BaseInvocationOutput):
-    """A float output"""
-
-    # fmt: off
-    type: Literal["float_output"] = "float_output"
-    param: float = Field(default=None, description="The output float")
-    # fmt: on
-
-
-class AddInvocation(BaseInvocation, MathInvocationConfig):
+@title("Add Integers")
+@tags("math")
+class AddInvocation(BaseInvocation):
     """Adds two numbers"""
 
-    # fmt: off
     type: Literal["add"] = "add"
-    a: int = Field(default=0, description="The first number")
-    b: int = Field(default=0, description="The second number")
-    # fmt: on
 
-    class Config(InvocationConfig):
-        schema_extra = {
-            "ui": {"title": "Add", "tags": ["math", "add"]},
-        }
+    # Inputs
+    a: int = InputField(default=0, description=FieldDescriptions.num_1)
+    b: int = InputField(default=0, description=FieldDescriptions.num_2)
 
-    def invoke(self, context: InvocationContext) -> IntOutput:
-        return IntOutput(a=self.a + self.b)
+    def invoke(self, context: InvocationContext) -> IntegerOutput:
+        return IntegerOutput(a=self.a + self.b)
 
 
-class SubtractInvocation(BaseInvocation, MathInvocationConfig):
+@title("Subtract Integers")
+@tags("math")
+class SubtractInvocation(BaseInvocation):
     """Subtracts two numbers"""
 
-    # fmt: off
     type: Literal["sub"] = "sub"
-    a: int = Field(default=0, description="The first number")
-    b: int = Field(default=0, description="The second number")
-    # fmt: on
 
-    class Config(InvocationConfig):
-        schema_extra = {
-            "ui": {"title": "Subtract", "tags": ["math", "subtract"]},
-        }
+    # Inputs
+    a: int = InputField(default=0, description=FieldDescriptions.num_1)
+    b: int = InputField(default=0, description=FieldDescriptions.num_2)
 
-    def invoke(self, context: InvocationContext) -> IntOutput:
-        return IntOutput(a=self.a - self.b)
+    def invoke(self, context: InvocationContext) -> IntegerOutput:
+        return IntegerOutput(a=self.a - self.b)
 
 
-class MultiplyInvocation(BaseInvocation, MathInvocationConfig):
+@title("Multiply Integers")
+@tags("math")
+class MultiplyInvocation(BaseInvocation):
     """Multiplies two numbers"""
 
-    # fmt: off
     type: Literal["mul"] = "mul"
-    a: int = Field(default=0, description="The first number")
-    b: int = Field(default=0, description="The second number")
-    # fmt: on
 
-    class Config(InvocationConfig):
-        schema_extra = {
-            "ui": {"title": "Multiply", "tags": ["math", "multiply"]},
-        }
+    # Inputs
+    a: int = InputField(default=0, description=FieldDescriptions.num_1)
+    b: int = InputField(default=0, description=FieldDescriptions.num_2)
 
-    def invoke(self, context: InvocationContext) -> IntOutput:
-        return IntOutput(a=self.a * self.b)
+    def invoke(self, context: InvocationContext) -> IntegerOutput:
+        return IntegerOutput(a=self.a * self.b)
 
 
-class DivideInvocation(BaseInvocation, MathInvocationConfig):
+@title("Divide Integers")
+@tags("math")
+class DivideInvocation(BaseInvocation):
     """Divides two numbers"""
 
-    # fmt: off
     type: Literal["div"] = "div"
-    a: int = Field(default=0, description="The first number")
-    b: int = Field(default=0, description="The second number")
-    # fmt: on
 
-    class Config(InvocationConfig):
-        schema_extra = {
-            "ui": {"title": "Divide", "tags": ["math", "divide"]},
-        }
+    # Inputs
+    a: int = InputField(default=0, description=FieldDescriptions.num_1)
+    b: int = InputField(default=0, description=FieldDescriptions.num_2)
 
-    def invoke(self, context: InvocationContext) -> IntOutput:
-        return IntOutput(a=int(self.a / self.b))
+    def invoke(self, context: InvocationContext) -> IntegerOutput:
+        return IntegerOutput(a=int(self.a / self.b))
 
 
+@title("Random Integer")
+@tags("math")
 class RandomIntInvocation(BaseInvocation):
     """Outputs a single random integer."""
 
-    # fmt: off
     type: Literal["rand_int"] = "rand_int"
-    low: int = Field(default=0, description="The inclusive low value")
-    high: int = Field(
-        default=np.iinfo(np.int32).max, description="The exclusive high value"
-    )
-    # fmt: on
 
-    class Config(InvocationConfig):
-        schema_extra = {
-            "ui": {"title": "Random Integer", "tags": ["math", "random", "integer"]},
-        }
+    # Inputs
+    low: int = InputField(default=0, description="The inclusive low value")
+    high: int = InputField(default=np.iinfo(np.int32).max, description="The exclusive high value")
 
-    def invoke(self, context: InvocationContext) -> IntOutput:
-        return IntOutput(a=np.random.randint(self.low, self.high))
+    def invoke(self, context: InvocationContext) -> IntegerOutput:
+        return IntegerOutput(a=np.random.randint(self.low, self.high))

invokeai/app/invocations/metadata.py

@@ -1,18 +1,22 @@
-from typing import Literal, Optional, Union
+from typing import Literal, Optional
 
 from pydantic import Field
 
-from ...version import __version__
 from invokeai.app.invocations.baseinvocation import (
     BaseInvocation,
     BaseInvocationOutput,
-    InvocationConfig,
+    InputField,
     InvocationContext,
+    OutputField,
+    tags,
+    title,
 )
 from invokeai.app.invocations.controlnet_image_processors import ControlField
 from invokeai.app.invocations.model import LoRAModelField, MainModelField, VAEModelField
 from invokeai.app.util.model_exclude_null import BaseModelExcludeNull
 
+from ...version import __version__
+
 
 class LoRAMetadataField(BaseModelExcludeNull):
     """LoRA metadata for an image generated in InvokeAI."""
@@ -43,34 +47,37 @@ class CoreMetadata(BaseModelExcludeNull):
     model: MainModelField = Field(description="The main model used for inference")
     controlnets: list[ControlField] = Field(description="The ControlNets used for inference")
     loras: list[LoRAMetadataField] = Field(description="The LoRAs used for inference")
-    vae: Union[VAEModelField, None] = Field(
+    vae: Optional[VAEModelField] = Field(
         default=None,
         description="The VAE used for decoding, if the main model's default was not used",
     )
 
     # Latents-to-Latents
-    strength: Union[float, None] = Field(
+    strength: Optional[float] = Field(
         default=None,
         description="The strength used for latents-to-latents",
     )
-    init_image: Union[str, None] = Field(default=None, description="The name of the initial image")
+    init_image: Optional[str] = Field(default=None, description="The name of the initial image")
 
     # SDXL
-    positive_style_prompt: Union[str, None] = Field(default=None, description="The positive style prompt parameter")
-    negative_style_prompt: Union[str, None] = Field(default=None, description="The negative style prompt parameter")
+    positive_style_prompt: Optional[str] = Field(default=None, description="The positive style prompt parameter")
+    negative_style_prompt: Optional[str] = Field(default=None, description="The negative style prompt parameter")
 
     # SDXL Refiner
-    refiner_model: Union[MainModelField, None] = Field(default=None, description="The SDXL Refiner model used")
-    refiner_cfg_scale: Union[float, None] = Field(
+    refiner_model: Optional[MainModelField] = Field(default=None, description="The SDXL Refiner model used")
+    refiner_cfg_scale: Optional[float] = Field(
         default=None,
         description="The classifier-free guidance scale parameter used for the refiner",
     )
-    refiner_steps: Union[int, None] = Field(default=None, description="The number of steps used for the refiner")
-    refiner_scheduler: Union[str, None] = Field(default=None, description="The scheduler used for the refiner")
-    refiner_aesthetic_store: Union[float, None] = Field(
+    refiner_steps: Optional[int] = Field(default=None, description="The number of steps used for the refiner")
+    refiner_scheduler: Optional[str] = Field(default=None, description="The scheduler used for the refiner")
+    refiner_positive_aesthetic_store: Optional[float] = Field(
         default=None, description="The aesthetic score used for the refiner"
     )
-    refiner_start: Union[float, None] = Field(default=None, description="The start value used for refiner denoising")
+    refiner_negative_aesthetic_store: Optional[float] = Field(
+        default=None, description="The aesthetic score used for the refiner"
+    )
+    refiner_start: Optional[float] = Field(default=None, description="The start value used for refiner denoising")
 
 
 class ImageMetadata(BaseModelExcludeNull):
@@ -88,66 +95,86 @@ class MetadataAccumulatorOutput(BaseInvocationOutput):
 
     type: Literal["metadata_accumulator_output"] = "metadata_accumulator_output"
 
-    metadata: CoreMetadata = Field(description="The core metadata for the image")
+    metadata: CoreMetadata = OutputField(description="The core metadata for the image")
 
 
+@title("Metadata Accumulator")
+@tags("metadata")
 class MetadataAccumulatorInvocation(BaseInvocation):
     """Outputs a Core Metadata Object"""
 
     type: Literal["metadata_accumulator"] = "metadata_accumulator"
 
-    generation_mode: str = Field(
+    generation_mode: str = InputField(
         description="The generation mode that output this image",
     )
-    positive_prompt: str = Field(description="The positive prompt parameter")
-    negative_prompt: str = Field(description="The negative prompt parameter")
-    width: int = Field(description="The width parameter")
-    height: int = Field(description="The height parameter")
-    seed: int = Field(description="The seed used for noise generation")
-    rand_device: str = Field(description="The device used for random number generation")
-    cfg_scale: float = Field(description="The classifier-free guidance scale parameter")
-    steps: int = Field(description="The number of steps used for inference")
-    scheduler: str = Field(description="The scheduler used for inference")
-    clip_skip: int = Field(
+    positive_prompt: str = InputField(description="The positive prompt parameter")
+    negative_prompt: str = InputField(description="The negative prompt parameter")
+    width: int = InputField(description="The width parameter")
+    height: int = InputField(description="The height parameter")
+    seed: int = InputField(description="The seed used for noise generation")
+    rand_device: str = InputField(description="The device used for random number generation")
+    cfg_scale: float = InputField(description="The classifier-free guidance scale parameter")
+    steps: int = InputField(description="The number of steps used for inference")
+    scheduler: str = InputField(description="The scheduler used for inference")
+    clip_skip: int = InputField(
         description="The number of skipped CLIP layers",
     )
-    model: MainModelField = Field(description="The main model used for inference")
-    controlnets: list[ControlField] = Field(description="The ControlNets used for inference")
-    loras: list[LoRAMetadataField] = Field(description="The LoRAs used for inference")
-    strength: Union[float, None] = Field(
+    model: MainModelField = InputField(description="The main model used for inference")
+    controlnets: list[ControlField] = InputField(description="The ControlNets used for inference")
+    loras: list[LoRAMetadataField] = InputField(description="The LoRAs used for inference")
+    strength: Optional[float] = InputField(
         default=None,
         description="The strength used for latents-to-latents",
     )
-    init_image: Union[str, None] = Field(default=None, description="The name of the initial image")
-    vae: Union[VAEModelField, None] = Field(
+    init_image: Optional[str] = InputField(
+        default=None,
+        description="The name of the initial image",
+    )
+    vae: Optional[VAEModelField] = InputField(
         default=None,
         description="The VAE used for decoding, if the main model's default was not used",
     )
 
     # SDXL
-    positive_style_prompt: Union[str, None] = Field(default=None, description="The positive style prompt parameter")
-    negative_style_prompt: Union[str, None] = Field(default=None, description="The negative style prompt parameter")
+    positive_style_prompt: Optional[str] = InputField(
+        default=None,
+        description="The positive style prompt parameter",
+    )
+    negative_style_prompt: Optional[str] = InputField(
+        default=None,
+        description="The negative style prompt parameter",
+    )
 
     # SDXL Refiner
-    refiner_model: Union[MainModelField, None] = Field(default=None, description="The SDXL Refiner model used")
-    refiner_cfg_scale: Union[float, None] = Field(
+    refiner_model: Optional[MainModelField] = InputField(
+        default=None,
+        description="The SDXL Refiner model used",
+    )
+    refiner_cfg_scale: Optional[float] = InputField(
         default=None,
         description="The classifier-free guidance scale parameter used for the refiner",
     )
-    refiner_steps: Union[int, None] = Field(default=None, description="The number of steps used for the refiner")
-    refiner_scheduler: Union[str, None] = Field(default=None, description="The scheduler used for the refiner")
-    refiner_aesthetic_store: Union[float, None] = Field(
-        default=None, description="The aesthetic score used for the refiner"
+    refiner_steps: Optional[int] = InputField(
+        default=None,
+        description="The number of steps used for the refiner",
+    )
+    refiner_scheduler: Optional[str] = InputField(
+        default=None,
+        description="The scheduler used for the refiner",
+    )
+    refiner_positive_aesthetic_store: Optional[float] = InputField(
+        default=None,
+        description="The aesthetic score used for the refiner",
+    )
+    refiner_negative_aesthetic_store: Optional[float] = InputField(
+        default=None,
+        description="The aesthetic score used for the refiner",
+    )
+    refiner_start: Optional[float] = InputField(
+        default=None,
+        description="The start value used for refiner denoising",
     )
-    refiner_start: Union[float, None] = Field(default=None, description="The start value used for refiner denoising")
-
-    class Config(InvocationConfig):
-        schema_extra = {
-            "ui": {
-                "title": "Metadata Accumulator",
-                "tags": ["image", "metadata", "generation"],
-            },
-        }
 
     def invoke(self, context: InvocationContext) -> MetadataAccumulatorOutput:
         """Collects and outputs a CoreMetadata object"""

invokeai/app/invocations/model.py

@@ -4,7 +4,18 @@ from typing import List, Literal, Optional, Union
 from pydantic import BaseModel, Field
 
 from ...backend.model_management import BaseModelType, ModelType, SubModelType
-from .baseinvocation import BaseInvocation, BaseInvocationOutput, InvocationConfig, InvocationContext
+from .baseinvocation import (
+    BaseInvocation,
+    BaseInvocationOutput,
+    FieldDescriptions,
+    InputField,
+    Input,
+    InvocationContext,
+    OutputField,
+    UIType,
+    tags,
+    title,
+)
 
 
 class ModelInfo(BaseModel):
@@ -39,13 +50,11 @@ class VaeField(BaseModel):
 class ModelLoaderOutput(BaseInvocationOutput):
     """Model loader output"""
 
-    # fmt: off
     type: Literal["model_loader_output"] = "model_loader_output"
 
-    unet: UNetField = Field(default=None, description="UNet submodel")
-    clip: ClipField = Field(default=None, description="Tokenizer and text_encoder submodels")
-    vae: VaeField = Field(default=None, description="Vae submodel")
-    # fmt: on
+    unet: UNetField = OutputField(description=FieldDescriptions.unet, title="UNet")
+    clip: ClipField = OutputField(description=FieldDescriptions.clip, title="CLIP")
+    vae: VaeField = OutputField(description=FieldDescriptions.vae, title="VAE")
 
 
 class MainModelField(BaseModel):
@@ -63,24 +72,17 @@ class LoRAModelField(BaseModel):
     base_model: BaseModelType = Field(description="Base model")
 
 
+@title("Main Model Loader")
+@tags("model")
 class MainModelLoaderInvocation(BaseInvocation):
     """Loads a main model, outputting its submodels."""
 
     type: Literal["main_model_loader"] = "main_model_loader"
 
-    model: MainModelField = Field(description="The model to load")
+    # Inputs
+    model: MainModelField = InputField(description=FieldDescriptions.main_model, input=Input.Direct)
     # TODO: precision?
 
-    # Schema customisation
-    class Config(InvocationConfig):
-        schema_extra = {
-            "ui": {
-                "title": "Model Loader",
-                "tags": ["model", "loader"],
-                "type_hints": {"model": "model"},
-            },
-        }
-
     def invoke(self, context: InvocationContext) -> ModelLoaderOutput:
         base_model = self.model.base_model
         model_name = self.model.model_name
@@ -155,22 +157,6 @@ class MainModelLoaderInvocation(BaseInvocation):
                 loras=[],
                 skipped_layers=0,
             ),
-            clip2=ClipField(
-                tokenizer=ModelInfo(
-                    model_name=model_name,
-                    base_model=base_model,
-                    model_type=model_type,
-                    submodel=SubModelType.Tokenizer2,
-                ),
-                text_encoder=ModelInfo(
-                    model_name=model_name,
-                    base_model=base_model,
-                    model_type=model_type,
-                    submodel=SubModelType.TextEncoder2,
-                ),
-                loras=[],
-                skipped_layers=0,
-            ),
             vae=VaeField(
                 vae=ModelInfo(
                     model_name=model_name,
@@ -188,30 +174,27 @@ class LoraLoaderOutput(BaseInvocationOutput):
     # fmt: off
     type: Literal["lora_loader_output"] = "lora_loader_output"
 
-    unet: Optional[UNetField] = Field(default=None, description="UNet submodel")
-    clip: Optional[ClipField] = Field(default=None, description="Tokenizer and text_encoder submodels")
+    unet: Optional[UNetField] = OutputField(default=None, description=FieldDescriptions.unet, title="UNet")
+    clip: Optional[ClipField] = OutputField(default=None, description=FieldDescriptions.clip, title="CLIP")
     # fmt: on
 
 
+@title("LoRA Loader")
+@tags("lora", "model")
 class LoraLoaderInvocation(BaseInvocation):
     """Apply selected lora to unet and text_encoder."""
 
     type: Literal["lora_loader"] = "lora_loader"
 
-    lora: Union[LoRAModelField, None] = Field(default=None, description="Lora model name")
-    weight: float = Field(default=0.75, description="With what weight to apply lora")
-
-    unet: Optional[UNetField] = Field(description="UNet model for applying lora")
-    clip: Optional[ClipField] = Field(description="Clip model for applying lora")
-
-    class Config(InvocationConfig):
-        schema_extra = {
-            "ui": {
-                "title": "Lora Loader",
-                "tags": ["lora", "loader"],
-                "type_hints": {"lora": "lora_model"},
-            },
-        }
+    # Inputs
+    lora: LoRAModelField = InputField(description=FieldDescriptions.lora_model, input=Input.Direct, title="LoRA")
+    weight: float = InputField(default=0.75, description=FieldDescriptions.lora_weight)
+    unet: Optional[UNetField] = InputField(
+        default=None, description=FieldDescriptions.unet, input=Input.Connection, title="UNet"
+    )
+    clip: Optional[ClipField] = InputField(
+        default=None, description=FieldDescriptions.clip, input=Input.Connection, title="CLIP"
+    )
 
     def invoke(self, context: InvocationContext) -> LoraLoaderOutput:
         if self.lora is None:
@@ -263,37 +246,35 @@ class LoraLoaderInvocation(BaseInvocation):
 
 
 class SDXLLoraLoaderOutput(BaseInvocationOutput):
-    """Model loader output"""
+    """SDXL LoRA Loader Output"""
 
     # fmt: off
     type: Literal["sdxl_lora_loader_output"] = "sdxl_lora_loader_output"
 
-    unet: Optional[UNetField] = Field(default=None, description="UNet submodel")
-    clip: Optional[ClipField] = Field(default=None, description="Tokenizer and text_encoder submodels")
-    clip2: Optional[ClipField] = Field(default=None, description="Tokenizer2 and text_encoder2 submodels")
+    unet: Optional[UNetField] = OutputField(default=None, description=FieldDescriptions.unet, title="UNet")
+    clip: Optional[ClipField] = OutputField(default=None, description=FieldDescriptions.clip, title="CLIP 1")
+    clip2: Optional[ClipField] = OutputField(default=None, description=FieldDescriptions.clip, title="CLIP 2")
     # fmt: on
 
 
+@title("SDXL LoRA Loader")
+@tags("sdxl", "lora", "model")
 class SDXLLoraLoaderInvocation(BaseInvocation):
     """Apply selected lora to unet and text_encoder."""
 
     type: Literal["sdxl_lora_loader"] = "sdxl_lora_loader"
 
-    lora: Union[LoRAModelField, None] = Field(default=None, description="Lora model name")
-    weight: float = Field(default=0.75, description="With what weight to apply lora")
-
-    unet: Optional[UNetField] = Field(description="UNet model for applying lora")
-    clip: Optional[ClipField] = Field(description="Clip model for applying lora")
-    clip2: Optional[ClipField] = Field(description="Clip2 model for applying lora")
-
-    class Config(InvocationConfig):
-        schema_extra = {
-            "ui": {
-                "title": "SDXL Lora Loader",
-                "tags": ["lora", "loader"],
-                "type_hints": {"lora": "lora_model"},
-            },
-        }
+    lora: LoRAModelField = InputField(description=FieldDescriptions.lora_model, input=Input.Direct, title="LoRA")
+    weight: float = Field(default=0.75, description=FieldDescriptions.lora_weight)
+    unet: Optional[UNetField] = Field(
+        default=None, description=FieldDescriptions.unet, input=Input.Connection, title="UNET"
+    )
+    clip: Optional[ClipField] = Field(
+        default=None, description=FieldDescriptions.clip, input=Input.Connection, title="CLIP 1"
+    )
+    clip2: Optional[ClipField] = Field(
+        default=None, description=FieldDescriptions.clip, input=Input.Connection, title="CLIP 2"
+    )
 
     def invoke(self, context: InvocationContext) -> SDXLLoraLoaderOutput:
         if self.lora is None:
@@ -369,29 +350,23 @@ class VAEModelField(BaseModel):
 class VaeLoaderOutput(BaseInvocationOutput):
     """Model loader output"""
 
-    # fmt: off
     type: Literal["vae_loader_output"] = "vae_loader_output"
 
-    vae: VaeField = Field(default=None, description="Vae model")
-    # fmt: on
+    # Outputs
+    vae: VaeField = OutputField(description=FieldDescriptions.vae, title="VAE")
 
 
+@title("VAE Loader")
+@tags("vae", "model")
 class VaeLoaderInvocation(BaseInvocation):
     """Loads a VAE model, outputting a VaeLoaderOutput"""
 
     type: Literal["vae_loader"] = "vae_loader"
 
-    vae_model: VAEModelField = Field(description="The VAE to load")
-
-    # Schema customisation
-    class Config(InvocationConfig):
-        schema_extra = {
-            "ui": {
-                "title": "VAE Loader",
-                "tags": ["vae", "loader"],
-                "type_hints": {"vae_model": "vae_model"},
-            },
-        }
+    # Inputs
+    vae_model: VAEModelField = InputField(
+        description=FieldDescriptions.vae_model, input=Input.Direct, ui_type=UIType.VaeModel, title="VAE"
+    )
 
     def invoke(self, context: InvocationContext) -> VaeLoaderOutput:
         base_model = self.vae_model.base_model

invokeai/app/invocations/noise.py

@@ -1,19 +1,24 @@
 # Copyright (c) 2023 Kyle Schouviller (https://github.com/kyle0654) & the InvokeAI Team
 
-import math
 from typing import Literal
 
-from pydantic import Field, validator
 import torch
-from invokeai.app.invocations.latent import LatentsField
+from pydantic import validator
 
+from invokeai.app.invocations.latent import LatentsField
 from invokeai.app.util.misc import SEED_MAX, get_random_seed
+
 from ...backend.util.devices import choose_torch_device, torch_dtype
 from .baseinvocation import (
     BaseInvocation,
     BaseInvocationOutput,
-    InvocationConfig,
+    FieldDescriptions,
+    InputField,
     InvocationContext,
+    OutputField,
+    UIType,
+    tags,
+    title,
 )
 
 """
@@ -61,62 +66,53 @@ Nodes
 class NoiseOutput(BaseInvocationOutput):
     """Invocation noise output"""
 
-    # fmt: off
-    type:  Literal["noise_output"] = "noise_output"
+    type: Literal["noise_output"] = "noise_output"
 
     # Inputs
-    noise: LatentsField            = Field(default=None, description="The output noise")
-    width:                     int = Field(description="The width of the noise in pixels")
-    height:                    int = Field(description="The height of the noise in pixels")
-    # fmt: on
+    noise: LatentsField = OutputField(default=None, description=FieldDescriptions.noise)
+    width: int = OutputField(description=FieldDescriptions.width)
+    height: int = OutputField(description=FieldDescriptions.height)
 
 
-def build_noise_output(latents_name: str, latents: torch.Tensor):
+def build_noise_output(latents_name: str, latents: torch.Tensor, seed: int):
     return NoiseOutput(
-        noise=LatentsField(latents_name=latents_name),
+        noise=LatentsField(latents_name=latents_name, seed=seed),
         width=latents.size()[3] * 8,
         height=latents.size()[2] * 8,
     )
 
 
+@title("Noise")
+@tags("latents", "noise")
 class NoiseInvocation(BaseInvocation):
     """Generates latent noise."""
 
     type: Literal["noise"] = "noise"
 
     # Inputs
-    seed: int = Field(
+    seed: int = InputField(
         ge=0,
         le=SEED_MAX,
-        description="The seed to use",
+        description=FieldDescriptions.seed,
         default_factory=get_random_seed,
     )
-    width: int = Field(
+    width: int = InputField(
         default=512,
         multiple_of=8,
         gt=0,
-        description="The width of the resulting noise",
+        description=FieldDescriptions.width,
     )
-    height: int = Field(
+    height: int = InputField(
         default=512,
         multiple_of=8,
         gt=0,
-        description="The height of the resulting noise",
+        description=FieldDescriptions.height,
     )
-    use_cpu: bool = Field(
+    use_cpu: bool = InputField(
         default=True,
         description="Use CPU for noise generation (for reproducible results across platforms)",
     )
 
-    # Schema customisation
-    class Config(InvocationConfig):
-        schema_extra = {
-            "ui": {
-                "title": "Noise",
-                "tags": ["latents", "noise"],
-            },
-        }
-
     @validator("seed", pre=True)
     def modulo_seed(cls, v):
         """Returns the seed modulo (SEED_MAX + 1) to ensure it is within the valid range."""
@@ -132,4 +128,4 @@ class NoiseInvocation(BaseInvocation):
         )
         name = f"{context.graph_execution_state_id}__{self.id}"
         context.services.latents.save(name, noise)
-        return build_noise_output(latents_name=name, latents=noise)
+        return build_noise_output(latents_name=name, latents=noise, seed=self.seed)

invokeai/app/invocations/onnx.py

@@ -1,37 +1,43 @@
 # Copyright (c) 2023 Borisov Sergey (https://github.com/StAlKeR7779)
 
+import inspect
+import re
 from contextlib import ExitStack
 from typing import List, Literal, Optional, Union
 
-import re
-import inspect
-
-from pydantic import BaseModel, Field, validator
-import torch
 import numpy as np
+import torch
 from diffusers import ControlNetModel, DPMSolverMultistepScheduler
 from diffusers.image_processor import VaeImageProcessor
 from diffusers.schedulers import SchedulerMixin as Scheduler
-
-from ..models.image import ImageCategory, ImageField, ResourceOrigin
-from ...backend.model_management import ONNXModelPatcher
-from ...backend.util import choose_torch_device
-from .baseinvocation import BaseInvocation, BaseInvocationOutput, InvocationConfig, InvocationContext
-from .compel import ConditioningField
-from .controlnet_image_processors import ControlField
-from .image import ImageOutput
-from .model import ModelInfo, UNetField, VaeField
+from pydantic import BaseModel, Field, validator
+from tqdm import tqdm
 
 from invokeai.app.invocations.metadata import CoreMetadata
-from invokeai.backend import BaseModelType, ModelType, SubModelType
+from invokeai.app.invocations.primitives import ConditioningField, ConditioningOutput, ImageField, ImageOutput
 from invokeai.app.util.step_callback import stable_diffusion_step_callback
+from invokeai.backend import BaseModelType, ModelType, SubModelType
+
+from ...backend.model_management import ONNXModelPatcher
 from ...backend.stable_diffusion import PipelineIntermediateState
-
-from tqdm import tqdm
-from .model import ClipField
-from .latent import LatentsField, LatentsOutput, build_latents_output, get_scheduler, SAMPLER_NAME_VALUES
-from .compel import CompelOutput
-
+from ...backend.util import choose_torch_device
+from ..models.image import ImageCategory, ResourceOrigin
+from .baseinvocation import (
+    BaseInvocation,
+    BaseInvocationOutput,
+    FieldDescriptions,
+    InputField,
+    Input,
+    InvocationContext,
+    OutputField,
+    UIComponent,
+    UIType,
+    tags,
+    title,
+)
+from .controlnet_image_processors import ControlField
+from .latent import SAMPLER_NAME_VALUES, LatentsField, LatentsOutput, build_latents_output, get_scheduler
+from .model import ClipField, ModelInfo, UNetField, VaeField
 
 ORT_TO_NP_TYPE = {
     "tensor(bool)": np.bool_,
@@ -51,13 +57,15 @@ ORT_TO_NP_TYPE = {
 PRECISION_VALUES = Literal[tuple(list(ORT_TO_NP_TYPE.keys()))]
 
 
+@title("ONNX Prompt (Raw)")
+@tags("onnx", "prompt")
 class ONNXPromptInvocation(BaseInvocation):
     type: Literal["prompt_onnx"] = "prompt_onnx"
 
-    prompt: str = Field(default="", description="Prompt")
-    clip: ClipField = Field(None, description="Clip to use")
+    prompt: str = InputField(default="", description=FieldDescriptions.raw_prompt, ui_component=UIComponent.Textarea)
+    clip: ClipField = InputField(description=FieldDescriptions.clip, input=Input.Connection)
 
-    def invoke(self, context: InvocationContext) -> CompelOutput:
+    def invoke(self, context: InvocationContext) -> ConditioningOutput:
         tokenizer_info = context.services.model_manager.get_model(
             **self.clip.tokenizer.dict(),
         )
@@ -126,7 +134,7 @@ class ONNXPromptInvocation(BaseInvocation):
         # TODO: hacky but works ;D maybe rename latents somehow?
         context.services.latents.save(conditioning_name, (prompt_embeds, None))
 
-        return CompelOutput(
+        return ConditioningOutput(
             conditioning=ConditioningField(
                 conditioning_name=conditioning_name,
             ),
@@ -134,25 +142,48 @@ class ONNXPromptInvocation(BaseInvocation):
 
 
 # Text to image
+@title("ONNX Text to Latents")
+@tags("latents", "inference", "txt2img", "onnx")
 class ONNXTextToLatentsInvocation(BaseInvocation):
     """Generates latents from conditionings."""
 
     type: Literal["t2l_onnx"] = "t2l_onnx"
 
     # Inputs
-    # fmt: off
-    positive_conditioning: Optional[ConditioningField] = Field(description="Positive conditioning for generation")
-    negative_conditioning: Optional[ConditioningField] = Field(description="Negative conditioning for generation")
-    noise: Optional[LatentsField] = Field(description="The noise to use")
-    steps:       int = Field(default=10, gt=0, description="The number of steps to use to generate the image")
-    cfg_scale: Union[float, List[float]] = Field(default=7.5, ge=1, description="The Classifier-Free Guidance, higher values may result in a result closer to the prompt", )
-    scheduler: SAMPLER_NAME_VALUES = Field(default="euler", description="The scheduler to use" )
-    precision: PRECISION_VALUES = Field(default = "tensor(float16)", description="The precision to use when generating latents")
-    unet: UNetField = Field(default=None, description="UNet submodel")
-    control: Union[ControlField, list[ControlField]] = Field(default=None, description="The control to use")
-    # seamless:   bool = Field(default=False, description="Whether or not to generate an image that can tile without seams", )
-    # seamless_axes: str = Field(default="", description="The axes to tile the image on, 'x' and/or 'y'")
-    # fmt: on
+    positive_conditioning: ConditioningField = InputField(
+        description=FieldDescriptions.positive_cond,
+        input=Input.Connection,
+    )
+    negative_conditioning: ConditioningField = InputField(
+        description=FieldDescriptions.negative_cond,
+        input=Input.Connection,
+    )
+    noise: LatentsField = InputField(
+        description=FieldDescriptions.noise,
+        input=Input.Connection,
+    )
+    steps: int = InputField(default=10, gt=0, description=FieldDescriptions.steps)
+    cfg_scale: Union[float, List[float]] = InputField(
+        default=7.5,
+        ge=1,
+        description=FieldDescriptions.cfg_scale,
+        ui_type=UIType.Float,
+    )
+    scheduler: SAMPLER_NAME_VALUES = InputField(
+        default="euler", description=FieldDescriptions.scheduler, input=Input.Direct
+    )
+    precision: PRECISION_VALUES = InputField(default="tensor(float16)", description=FieldDescriptions.precision)
+    unet: UNetField = InputField(
+        description=FieldDescriptions.unet,
+        input=Input.Connection,
+    )
+    control: Optional[Union[ControlField, list[ControlField]]] = InputField(
+        default=None,
+        description=FieldDescriptions.control,
+        ui_type=UIType.Control,
+    )
+    # seamless:   bool = InputField(default=False, description="Whether or not to generate an image that can tile without seams", )
+    # seamless_axes: str = InputField(default="", description="The axes to tile the image on, 'x' and/or 'y'")
 
     @validator("cfg_scale")
     def ge_one(cls, v):
@@ -166,20 +197,6 @@ class ONNXTextToLatentsInvocation(BaseInvocation):
                 raise ValueError("cfg_scale must be greater than 1")
         return v
 
-    # Schema customisation
-    class Config(InvocationConfig):
-        schema_extra = {
-            "ui": {
-                "tags": ["latents"],
-                "type_hints": {
-                    "model": "model",
-                    "control": "control",
-                    # "cfg_scale": "float",
-                    "cfg_scale": "number",
-                },
-            },
-        }
-
     # based on
     # https://github.com/huggingface/diffusers/blob/3ebbaf7c96801271f9e6c21400033b6aa5ffcf29/src/diffusers/pipelines/stable_diffusion/pipeline_onnx_stable_diffusion.py#L375
     def invoke(self, context: InvocationContext) -> LatentsOutput:
@@ -212,6 +229,7 @@ class ONNXTextToLatentsInvocation(BaseInvocation):
             context=context,
             scheduler_info=self.unet.scheduler,
             scheduler_name=self.scheduler,
+            seed=0,  # TODO: refactor this node
         )
 
         def torch2numpy(latent: torch.Tensor):
@@ -299,26 +317,28 @@ class ONNXTextToLatentsInvocation(BaseInvocation):
 
 
 # Latent to image
+@title("ONNX Latents to Image")
+@tags("latents", "image", "vae", "onnx")
 class ONNXLatentsToImageInvocation(BaseInvocation):
     """Generates an image from latents."""
 
     type: Literal["l2i_onnx"] = "l2i_onnx"
 
     # Inputs
-    latents: Optional[LatentsField] = Field(description="The latents to generate an image from")
-    vae: VaeField = Field(default=None, description="Vae submodel")
-    metadata: Optional[CoreMetadata] = Field(
-        default=None, description="Optional core metadata to be written to the image"
+    latents: LatentsField = InputField(
+        description=FieldDescriptions.denoised_latents,
+        input=Input.Connection,
     )
-    # tiled: bool = Field(default=False, description="Decode latents by overlaping tiles(less memory consumption)")
-
-    # Schema customisation
-    class Config(InvocationConfig):
-        schema_extra = {
-            "ui": {
-                "tags": ["latents", "image"],
-            },
-        }
+    vae: VaeField = InputField(
+        description=FieldDescriptions.vae,
+        input=Input.Connection,
+    )
+    metadata: Optional[CoreMetadata] = InputField(
+        default=None,
+        description=FieldDescriptions.core_metadata,
+        ui_hidden=True,
+    )
+    # tiled: bool = InputField(default=False, description="Decode latents by overlaping tiles(less memory consumption)")
 
     def invoke(self, context: InvocationContext) -> ImageOutput:
         latents = context.services.latents.get(self.latents.latents_name)
@@ -372,89 +392,13 @@ class ONNXModelLoaderOutput(BaseInvocationOutput):
     # fmt: off
     type: Literal["model_loader_output_onnx"] = "model_loader_output_onnx"
 
-    unet: UNetField = Field(default=None, description="UNet submodel")
-    clip: ClipField = Field(default=None, description="Tokenizer and text_encoder submodels")
-    vae_decoder: VaeField = Field(default=None, description="Vae submodel")
-    vae_encoder: VaeField = Field(default=None, description="Vae submodel")
+    unet: UNetField = OutputField(default=None, description=FieldDescriptions.unet, title="UNet")
+    clip: ClipField = OutputField(default=None, description=FieldDescriptions.clip, title="CLIP")
+    vae_decoder: VaeField = OutputField(default=None, description=FieldDescriptions.vae, title="VAE Decoder")
+    vae_encoder: VaeField = OutputField(default=None, description=FieldDescriptions.vae, title="VAE Encoder")
     # fmt: on
 
 
-class ONNXSD1ModelLoaderInvocation(BaseInvocation):
-    """Loading submodels of selected model."""
-
-    type: Literal["sd1_model_loader_onnx"] = "sd1_model_loader_onnx"
-
-    model_name: str = Field(default="", description="Model to load")
-    # TODO: precision?
-
-    # Schema customisation
-    class Config(InvocationConfig):
-        schema_extra = {
-            "ui": {"tags": ["model", "loader"], "type_hints": {"model_name": "model"}},  # TODO: rename to model_name?
-        }
-
-    def invoke(self, context: InvocationContext) -> ONNXModelLoaderOutput:
-        model_name = "stable-diffusion-v1-5"
-        base_model = BaseModelType.StableDiffusion1
-
-        # TODO: not found exceptions
-        if not context.services.model_manager.model_exists(
-            model_name=model_name,
-            base_model=BaseModelType.StableDiffusion1,
-            model_type=ModelType.ONNX,
-        ):
-            raise Exception(f"Unkown model name: {model_name}!")
-
-        return ONNXModelLoaderOutput(
-            unet=UNetField(
-                unet=ModelInfo(
-                    model_name=model_name,
-                    base_model=base_model,
-                    model_type=ModelType.ONNX,
-                    submodel=SubModelType.UNet,
-                ),
-                scheduler=ModelInfo(
-                    model_name=model_name,
-                    base_model=base_model,
-                    model_type=ModelType.ONNX,
-                    submodel=SubModelType.Scheduler,
-                ),
-                loras=[],
-            ),
-            clip=ClipField(
-                tokenizer=ModelInfo(
-                    model_name=model_name,
-                    base_model=base_model,
-                    model_type=ModelType.ONNX,
-                    submodel=SubModelType.Tokenizer,
-                ),
-                text_encoder=ModelInfo(
-                    model_name=model_name,
-                    base_model=base_model,
-                    model_type=ModelType.ONNX,
-                    submodel=SubModelType.TextEncoder,
-                ),
-                loras=[],
-            ),
-            vae_decoder=VaeField(
-                vae=ModelInfo(
-                    model_name=model_name,
-                    base_model=base_model,
-                    model_type=ModelType.ONNX,
-                    submodel=SubModelType.VaeDecoder,
-                ),
-            ),
-            vae_encoder=VaeField(
-                vae=ModelInfo(
-                    model_name=model_name,
-                    base_model=base_model,
-                    model_type=ModelType.ONNX,
-                    submodel=SubModelType.VaeEncoder,
-                ),
-            ),
-        )
-
-
 class OnnxModelField(BaseModel):
     """Onnx model field"""
 
@@ -463,22 +407,17 @@ class OnnxModelField(BaseModel):
     model_type: ModelType = Field(description="Model Type")
 
 
+@title("ONNX Model Loader")
+@tags("onnx", "model")
 class OnnxModelLoaderInvocation(BaseInvocation):
     """Loads a main model, outputting its submodels."""
 
     type: Literal["onnx_model_loader"] = "onnx_model_loader"
 
-    model: OnnxModelField = Field(description="The model to load")
-
-    # Schema customisation
-    class Config(InvocationConfig):
-        schema_extra = {
-            "ui": {
-                "title": "Onnx Model Loader",
-                "tags": ["model", "loader"],
-                "type_hints": {"model": "model"},
-            },
-        }
+    # Inputs
+    model: OnnxModelField = InputField(
+        description=FieldDescriptions.onnx_main_model, input=Input.Direct, ui_type=UIType.ONNXModel
+    )
 
     def invoke(self, context: InvocationContext) -> ONNXModelLoaderOutput:
         base_model = self.model.base_model

invokeai/app/invocations/param_easing.py

@@ -1,73 +1,64 @@
 import io
-from typing import Literal, Optional, Any
+from typing import Literal, Optional
 
-# from PIL.Image import Image
-import PIL.Image
-from matplotlib.ticker import MaxNLocator
-from matplotlib.figure import Figure
-
-from pydantic import BaseModel, Field
-import numpy as np
 import matplotlib.pyplot as plt
+import numpy as np
 
+import PIL.Image
 from easing_functions import (
-    LinearInOut,
-    QuadEaseInOut,
-    QuadEaseIn,
-    QuadEaseOut,
-    CubicEaseInOut,
-    CubicEaseIn,
-    CubicEaseOut,
-    QuarticEaseInOut,
-    QuarticEaseIn,
-    QuarticEaseOut,
-    QuinticEaseInOut,
-    QuinticEaseIn,
-    QuinticEaseOut,
-    SineEaseInOut,
-    SineEaseIn,
-    SineEaseOut,
-    CircularEaseIn,
-    CircularEaseInOut,
-    CircularEaseOut,
-    ExponentialEaseInOut,
-    ExponentialEaseIn,
-    ExponentialEaseOut,
-    ElasticEaseIn,
-    ElasticEaseInOut,
-    ElasticEaseOut,
     BackEaseIn,
     BackEaseInOut,
     BackEaseOut,
     BounceEaseIn,
     BounceEaseInOut,
     BounceEaseOut,
+    CircularEaseIn,
+    CircularEaseInOut,
+    CircularEaseOut,
+    CubicEaseIn,
+    CubicEaseInOut,
+    CubicEaseOut,
+    ElasticEaseIn,
+    ElasticEaseInOut,
+    ElasticEaseOut,
+    ExponentialEaseIn,
+    ExponentialEaseInOut,
+    ExponentialEaseOut,
+    LinearInOut,
+    QuadEaseIn,
+    QuadEaseInOut,
+    QuadEaseOut,
+    QuarticEaseIn,
+    QuarticEaseInOut,
+    QuarticEaseOut,
+    QuinticEaseIn,
+    QuinticEaseInOut,
+    QuinticEaseOut,
+    SineEaseIn,
+    SineEaseInOut,
+    SineEaseOut,
 )
+from matplotlib.figure import Figure
+from matplotlib.ticker import MaxNLocator
+from pydantic import BaseModel, Field
+
+from invokeai.app.invocations.primitives import FloatCollectionOutput
 
-from .baseinvocation import (
-    BaseInvocation,
-    BaseInvocationOutput,
-    InvocationContext,
-    InvocationConfig,
-)
 from ...backend.util.logging import InvokeAILogger
-from .collections import FloatCollectionOutput
+from .baseinvocation import BaseInvocation, InputField, InvocationContext, tags, title
 
 
+@title("Float Range")
+@tags("math", "range")
 class FloatLinearRangeInvocation(BaseInvocation):
     """Creates a range"""
 
     type: Literal["float_range"] = "float_range"
 
     # Inputs
-    start: float = Field(default=5, description="The first value of the range")
-    stop: float = Field(default=10, description="The last value of the range")
-    steps: int = Field(default=30, description="number of values to interpolate over (including start and stop)")
-
-    class Config(InvocationConfig):
-        schema_extra = {
-            "ui": {"title": "Linear Range (Float)", "tags": ["math", "float", "linear", "range"]},
-        }
+    start: float = InputField(default=5, description="The first value of the range")
+    stop: float = InputField(default=10, description="The last value of the range")
+    steps: int = InputField(default=30, description="number of values to interpolate over (including start and stop)")
 
     def invoke(self, context: InvocationContext) -> FloatCollectionOutput:
         param_list = list(np.linspace(self.start, self.stop, self.steps))
@@ -108,37 +99,32 @@ EASING_FUNCTIONS_MAP = {
     "BounceInOut": BounceEaseInOut,
 }
 
-EASING_FUNCTION_KEYS: Any = Literal[tuple(list(EASING_FUNCTIONS_MAP.keys()))]
+EASING_FUNCTION_KEYS = Literal[tuple(list(EASING_FUNCTIONS_MAP.keys()))]
 
 
 # actually I think for now could just use CollectionOutput (which is list[Any]
+@title("Step Param Easing")
+@tags("step", "easing")
 class StepParamEasingInvocation(BaseInvocation):
     """Experimental per-step parameter easing for denoising steps"""
 
     type: Literal["step_param_easing"] = "step_param_easing"
 
     # Inputs
-    # fmt: off
-    easing: EASING_FUNCTION_KEYS = Field(default="Linear", description="The easing function to use")
-    num_steps: int = Field(default=20, description="number of denoising steps")
-    start_value: float = Field(default=0.0, description="easing starting value")
-    end_value: float = Field(default=1.0, description="easing ending value")
-    start_step_percent: float = Field(default=0.0, description="fraction of steps at which to start easing")
-    end_step_percent: float = Field(default=1.0, description="fraction of steps after which to end easing")
+    easing: EASING_FUNCTION_KEYS = InputField(default="Linear", description="The easing function to use")
+    num_steps: int = InputField(default=20, description="number of denoising steps")
+    start_value: float = InputField(default=0.0, description="easing starting value")
+    end_value: float = InputField(default=1.0, description="easing ending value")
+    start_step_percent: float = InputField(default=0.0, description="fraction of steps at which to start easing")
+    end_step_percent: float = InputField(default=1.0, description="fraction of steps after which to end easing")
     # if None, then start_value is used prior to easing start
-    pre_start_value: Optional[float] = Field(default=None, description="value before easing start")
+    pre_start_value: Optional[float] = InputField(default=None, description="value before easing start")
     # if None, then end value is used prior to easing end
-    post_end_value: Optional[float] = Field(default=None, description="value after easing end")
-    mirror: bool = Field(default=False, description="include mirror of easing function")
+    post_end_value: Optional[float] = InputField(default=None, description="value after easing end")
+    mirror: bool = InputField(default=False, description="include mirror of easing function")
     # FIXME: add alt_mirror option (alternative to default or mirror), or remove entirely
-    # alt_mirror: bool = Field(default=False, description="alternative mirroring by dual easing")
-    show_easing_plot: bool = Field(default=False, description="show easing plot")
-    # fmt: on
-
-    class Config(InvocationConfig):
-        schema_extra = {
-            "ui": {"title": "Param Easing By Step", "tags": ["param", "step", "easing"]},
-        }
+    # alt_mirror: bool = InputField(default=False, description="alternative mirroring by dual easing")
+    show_easing_plot: bool = InputField(default=False, description="show easing plot")
 
     def invoke(self, context: InvocationContext) -> FloatCollectionOutput:
         log_diagnostics = False

invokeai/app/invocations/params.py

@@ -1,83 +0,0 @@
-# Copyright (c) 2023 Kyle Schouviller (https://github.com/kyle0654)
-
-from typing import Literal
-
-from pydantic import Field
-
-from invokeai.app.invocations.prompt import PromptOutput
-
-from .baseinvocation import BaseInvocation, BaseInvocationOutput, InvocationConfig, InvocationContext
-from .math import FloatOutput, IntOutput
-
-# Pass-through parameter nodes - used by subgraphs
-
-
-class ParamIntInvocation(BaseInvocation):
-    """An integer parameter"""
-
-    # fmt: off
-    type: Literal["param_int"] = "param_int"
-    a: int = Field(default=0, description="The integer value")
-    # fmt: on
-
-    class Config(InvocationConfig):
-        schema_extra = {
-            "ui": {"tags": ["param", "integer"], "title": "Integer Parameter"},
-        }
-
-    def invoke(self, context: InvocationContext) -> IntOutput:
-        return IntOutput(a=self.a)
-
-
-class ParamFloatInvocation(BaseInvocation):
-    """A float parameter"""
-
-    # fmt: off
-    type: Literal["param_float"] = "param_float"
-    param: float = Field(default=0.0, description="The float value")
-    # fmt: on
-
-    class Config(InvocationConfig):
-        schema_extra = {
-            "ui": {"tags": ["param", "float"], "title": "Float Parameter"},
-        }
-
-    def invoke(self, context: InvocationContext) -> FloatOutput:
-        return FloatOutput(param=self.param)
-
-
-class StringOutput(BaseInvocationOutput):
-    """A string output"""
-
-    type: Literal["string_output"] = "string_output"
-    text: str = Field(default=None, description="The output string")
-
-
-class ParamStringInvocation(BaseInvocation):
-    """A string parameter"""
-
-    type: Literal["param_string"] = "param_string"
-    text: str = Field(default="", description="The string value")
-
-    class Config(InvocationConfig):
-        schema_extra = {
-            "ui": {"tags": ["param", "string"], "title": "String Parameter"},
-        }
-
-    def invoke(self, context: InvocationContext) -> StringOutput:
-        return StringOutput(text=self.text)
-
-
-class ParamPromptInvocation(BaseInvocation):
-    """A prompt input parameter"""
-
-    type: Literal["param_prompt"] = "param_prompt"
-    prompt: str = Field(default="", description="The prompt value")
-
-    class Config(InvocationConfig):
-        schema_extra = {
-            "ui": {"tags": ["param", "prompt"], "title": "Prompt"},
-        }
-
-    def invoke(self, context: InvocationContext) -> PromptOutput:
-        return PromptOutput(prompt=self.prompt)

invokeai/app/invocations/primitives.py

@@ -0,0 +1,494 @@
+# Copyright (c) 2023 Kyle Schouviller (https://github.com/kyle0654)
+
+from typing import Literal, Optional, Tuple, Union
+from anyio import Condition
+
+from pydantic import BaseModel, Field
+import torch
+
+from .baseinvocation import (
+    BaseInvocation,
+    BaseInvocationOutput,
+    FieldDescriptions,
+    Input,
+    InputField,
+    InvocationContext,
+    OutputField,
+    UIComponent,
+    UIType,
+    tags,
+    title,
+)
+
+"""
+Primitives: Boolean, Integer, Float, String, Image, Latents, Conditioning, Color
+- primitive nodes
+- primitive outputs
+- primitive collection outputs
+"""
+
+# region Boolean
+
+
+class BooleanOutput(BaseInvocationOutput):
+    """Base class for nodes that output a single boolean"""
+
+    type: Literal["boolean_output"] = "boolean_output"
+    a: bool = OutputField(description="The output boolean")
+
+
+class BooleanCollectionOutput(BaseInvocationOutput):
+    """Base class for nodes that output a collection of booleans"""
+
+    type: Literal["boolean_collection_output"] = "boolean_collection_output"
+
+    # Outputs
+    collection: list[bool] = OutputField(
+        default_factory=list, description="The output boolean collection", ui_type=UIType.BooleanCollection
+    )
+
+
+@title("Boolean Primitive")
+@tags("primitives", "boolean")
+class BooleanInvocation(BaseInvocation):
+    """A boolean primitive value"""
+
+    type: Literal["boolean"] = "boolean"
+
+    # Inputs
+    a: bool = InputField(default=False, description="The boolean value")
+
+    def invoke(self, context: InvocationContext) -> BooleanOutput:
+        return BooleanOutput(a=self.a)
+
+
+@title("Boolean Primitive Collection")
+@tags("primitives", "boolean", "collection")
+class BooleanCollectionInvocation(BaseInvocation):
+    """A collection of boolean primitive values"""
+
+    type: Literal["boolean_collection"] = "boolean_collection"
+
+    # Inputs
+    collection: list[bool] = InputField(
+        default=False, description="The collection of boolean values", ui_type=UIType.BooleanCollection
+    )
+
+    def invoke(self, context: InvocationContext) -> BooleanCollectionOutput:
+        return BooleanCollectionOutput(collection=self.collection)
+
+
+# endregion
+
+# region Integer
+
+
+class IntegerOutput(BaseInvocationOutput):
+    """Base class for nodes that output a single integer"""
+
+    type: Literal["integer_output"] = "integer_output"
+    a: int = OutputField(description="The output integer")
+
+
+class IntegerCollectionOutput(BaseInvocationOutput):
+    """Base class for nodes that output a collection of integers"""
+
+    type: Literal["integer_collection_output"] = "integer_collection_output"
+
+    # Outputs
+    collection: list[int] = OutputField(
+        default_factory=list, description="The int collection", ui_type=UIType.IntegerCollection
+    )
+
+
+@title("Integer Primitive")
+@tags("primitives", "integer")
+class IntegerInvocation(BaseInvocation):
+    """An integer primitive value"""
+
+    type: Literal["integer"] = "integer"
+
+    # Inputs
+    a: int = InputField(default=0, description="The integer value")
+
+    def invoke(self, context: InvocationContext) -> IntegerOutput:
+        return IntegerOutput(a=self.a)
+
+
+@title("Integer Primitive Collection")
+@tags("primitives", "integer", "collection")
+class IntegerCollectionInvocation(BaseInvocation):
+    """A collection of integer primitive values"""
+
+    type: Literal["integer_collection"] = "integer_collection"
+
+    # Inputs
+    collection: list[int] = InputField(
+        default=0, description="The collection of integer values", ui_type=UIType.IntegerCollection
+    )
+
+    def invoke(self, context: InvocationContext) -> IntegerCollectionOutput:
+        return IntegerCollectionOutput(collection=self.collection)
+
+
+# endregion
+
+# region Float
+
+
+class FloatOutput(BaseInvocationOutput):
+    """Base class for nodes that output a single float"""
+
+    type: Literal["float_output"] = "float_output"
+    a: float = OutputField(description="The output float")
+
+
+class FloatCollectionOutput(BaseInvocationOutput):
+    """Base class for nodes that output a collection of floats"""
+
+    type: Literal["float_collection_output"] = "float_collection_output"
+
+    # Outputs
+    collection: list[float] = OutputField(
+        default_factory=list, description="The float collection", ui_type=UIType.FloatCollection
+    )
+
+
+@title("Float Primitive")
+@tags("primitives", "float")
+class FloatInvocation(BaseInvocation):
+    """A float primitive value"""
+
+    type: Literal["float"] = "float"
+
+    # Inputs
+    param: float = InputField(default=0.0, description="The float value")
+
+    def invoke(self, context: InvocationContext) -> FloatOutput:
+        return FloatOutput(a=self.param)
+
+
+@title("Float Primitive Collection")
+@tags("primitives", "float", "collection")
+class FloatCollectionInvocation(BaseInvocation):
+    """A collection of float primitive values"""
+
+    type: Literal["float_collection"] = "float_collection"
+
+    # Inputs
+    collection: list[float] = InputField(
+        default=0, description="The collection of float values", ui_type=UIType.FloatCollection
+    )
+
+    def invoke(self, context: InvocationContext) -> FloatCollectionOutput:
+        return FloatCollectionOutput(collection=self.collection)
+
+
+# endregion
+
+# region String
+
+
+class StringOutput(BaseInvocationOutput):
+    """Base class for nodes that output a single string"""
+
+    type: Literal["string_output"] = "string_output"
+    text: str = OutputField(description="The output string")
+
+
+class StringCollectionOutput(BaseInvocationOutput):
+    """Base class for nodes that output a collection of strings"""
+
+    type: Literal["string_collection_output"] = "string_collection_output"
+
+    # Outputs
+    collection: list[str] = OutputField(
+        default_factory=list, description="The output strings", ui_type=UIType.StringCollection
+    )
+
+
+@title("String Primitive")
+@tags("primitives", "string")
+class StringInvocation(BaseInvocation):
+    """A string primitive value"""
+
+    type: Literal["string"] = "string"
+
+    # Inputs
+    text: str = InputField(default="", description="The string value", ui_component=UIComponent.Textarea)
+
+    def invoke(self, context: InvocationContext) -> StringOutput:
+        return StringOutput(text=self.text)
+
+
+@title("String Primitive Collection")
+@tags("primitives", "string", "collection")
+class StringCollectionInvocation(BaseInvocation):
+    """A collection of string primitive values"""
+
+    type: Literal["string_collection"] = "string_collection"
+
+    # Inputs
+    collection: list[str] = InputField(
+        default=0, description="The collection of string values", ui_type=UIType.StringCollection
+    )
+
+    def invoke(self, context: InvocationContext) -> StringCollectionOutput:
+        return StringCollectionOutput(collection=self.collection)
+
+
+# endregion
+
+# region Image
+
+
+class ImageField(BaseModel):
+    """An image primitive field"""
+
+    image_name: str = Field(description="The name of the image")
+
+
+class ImageOutput(BaseInvocationOutput):
+    """Base class for nodes that output a single image"""
+
+    type: Literal["image_output"] = "image_output"
+    image: ImageField = OutputField(description="The output image")
+    width: int = OutputField(description="The width of the image in pixels")
+    height: int = OutputField(description="The height of the image in pixels")
+
+
+class ImageCollectionOutput(BaseInvocationOutput):
+    """Base class for nodes that output a collection of images"""
+
+    type: Literal["image_collection_output"] = "image_collection_output"
+
+    # Outputs
+    collection: list[ImageField] = OutputField(
+        default_factory=list, description="The output images", ui_type=UIType.ImageCollection
+    )
+
+
+@title("Image Primitive")
+@tags("primitives", "image")
+class ImageInvocation(BaseInvocation):
+    """An image primitive value"""
+
+    # Metadata
+    type: Literal["image"] = "image"
+
+    # Inputs
+    image: ImageField = InputField(description="The image to load")
+
+    def invoke(self, context: InvocationContext) -> ImageOutput:
+        image = context.services.images.get_pil_image(self.image.image_name)
+
+        return ImageOutput(
+            image=ImageField(image_name=self.image.image_name),
+            width=image.width,
+            height=image.height,
+        )
+
+
+@title("Image Primitive Collection")
+@tags("primitives", "image", "collection")
+class ImageCollectionInvocation(BaseInvocation):
+    """A collection of image primitive values"""
+
+    type: Literal["image_collection"] = "image_collection"
+
+    # Inputs
+    collection: list[ImageField] = InputField(
+        default=0, description="The collection of image values", ui_type=UIType.ImageCollection
+    )
+
+    def invoke(self, context: InvocationContext) -> ImageCollectionOutput:
+        return ImageCollectionOutput(collection=self.collection)
+
+
+# endregion
+
+# region Latents
+
+
+class LatentsField(BaseModel):
+    """A latents tensor primitive field"""
+
+    latents_name: str = Field(description="The name of the latents")
+    seed: Optional[int] = Field(default=None, description="Seed used to generate this latents")
+
+
+class LatentsOutput(BaseInvocationOutput):
+    """Base class for nodes that output a single latents tensor"""
+
+    type: Literal["latents_output"] = "latents_output"
+
+    latents: LatentsField = OutputField(
+        description=FieldDescriptions.latents,
+    )
+    width: int = OutputField(description=FieldDescriptions.width)
+    height: int = OutputField(description=FieldDescriptions.height)
+
+
+class LatentsCollectionOutput(BaseInvocationOutput):
+    """Base class for nodes that output a collection of latents tensors"""
+
+    type: Literal["latents_collection_output"] = "latents_collection_output"
+
+    collection: list[LatentsField] = OutputField(
+        default_factory=list,
+        description=FieldDescriptions.latents,
+        ui_type=UIType.LatentsCollection,
+    )
+
+
+@title("Latents Primitive")
+@tags("primitives", "latents")
+class LatentsInvocation(BaseInvocation):
+    """A latents tensor primitive value"""
+
+    type: Literal["latents"] = "latents"
+
+    # Inputs
+    latents: LatentsField = InputField(description="The latents tensor", input=Input.Connection)
+
+    def invoke(self, context: InvocationContext) -> LatentsOutput:
+        latents = context.services.latents.get(self.latents.latents_name)
+
+        return build_latents_output(self.latents.latents_name, latents)
+
+
+@title("Latents Primitive Collection")
+@tags("primitives", "latents", "collection")
+class LatentsCollectionInvocation(BaseInvocation):
+    """A collection of latents tensor primitive values"""
+
+    type: Literal["latents_collection"] = "latents_collection"
+
+    # Inputs
+    collection: list[LatentsField] = InputField(
+        default=0, description="The collection of latents tensors", ui_type=UIType.LatentsCollection
+    )
+
+    def invoke(self, context: InvocationContext) -> LatentsCollectionOutput:
+        return LatentsCollectionOutput(collection=self.collection)
+
+
+def build_latents_output(latents_name: str, latents: torch.Tensor, seed: Optional[int] = None):
+    return LatentsOutput(
+        latents=LatentsField(latents_name=latents_name, seed=seed),
+        width=latents.size()[3] * 8,
+        height=latents.size()[2] * 8,
+    )
+
+
+# endregion
+
+# region Color
+
+
+class ColorField(BaseModel):
+    """A color primitive field"""
+
+    r: int = Field(ge=0, le=255, description="The red component")
+    g: int = Field(ge=0, le=255, description="The green component")
+    b: int = Field(ge=0, le=255, description="The blue component")
+    a: int = Field(ge=0, le=255, description="The alpha component")
+
+    def tuple(self) -> Tuple[int, int, int, int]:
+        return (self.r, self.g, self.b, self.a)
+
+
+class ColorOutput(BaseInvocationOutput):
+    """Base class for nodes that output a single color"""
+
+    type: Literal["color_output"] = "color_output"
+    color: ColorField = OutputField(description="The output color")
+
+
+class ColorCollectionOutput(BaseInvocationOutput):
+    """Base class for nodes that output a collection of colors"""
+
+    type: Literal["color_collection_output"] = "color_collection_output"
+
+    # Outputs
+    collection: list[ColorField] = OutputField(
+        default_factory=list, description="The output colors", ui_type=UIType.ColorCollection
+    )
+
+
+@title("Color Primitive")
+@tags("primitives", "color")
+class ColorInvocation(BaseInvocation):
+    """A color primitive value"""
+
+    type: Literal["color"] = "color"
+
+    # Inputs
+    color: ColorField = InputField(default=ColorField(r=0, g=0, b=0, a=255), description="The color value")
+
+    def invoke(self, context: InvocationContext) -> ColorOutput:
+        return ColorOutput(color=self.color)
+
+
+# endregion
+
+# region Conditioning
+
+
+class ConditioningField(BaseModel):
+    """A conditioning tensor primitive value"""
+
+    conditioning_name: str = Field(description="The name of conditioning tensor")
+
+
+class ConditioningOutput(BaseInvocationOutput):
+    """Base class for nodes that output a single conditioning tensor"""
+
+    type: Literal["conditioning_output"] = "conditioning_output"
+
+    conditioning: ConditioningField = OutputField(description=FieldDescriptions.cond)
+
+
+class ConditioningCollectionOutput(BaseInvocationOutput):
+    """Base class for nodes that output a collection of conditioning tensors"""
+
+    type: Literal["conditioning_collection_output"] = "conditioning_collection_output"
+
+    # Outputs
+    collection: list[ConditioningField] = OutputField(
+        default_factory=list,
+        description="The output conditioning tensors",
+        ui_type=UIType.ConditioningCollection,
+    )
+
+
+@title("Conditioning Primitive")
+@tags("primitives", "conditioning")
+class ConditioningInvocation(BaseInvocation):
+    """A conditioning tensor primitive value"""
+
+    type: Literal["conditioning"] = "conditioning"
+
+    conditioning: ConditioningField = InputField(description=FieldDescriptions.cond, input=Input.Connection)
+
+    def invoke(self, context: InvocationContext) -> ConditioningOutput:
+        return ConditioningOutput(conditioning=self.conditioning)
+
+
+@title("Conditioning Primitive Collection")
+@tags("primitives", "conditioning", "collection")
+class ConditioningCollectionInvocation(BaseInvocation):
+    """A collection of conditioning tensor primitive values"""
+
+    type: Literal["conditioning_collection"] = "conditioning_collection"
+
+    # Inputs
+    collection: list[ConditioningField] = InputField(
+        default=0, description="The collection of conditioning tensors", ui_type=UIType.ConditioningCollection
+    )
+
+    def invoke(self, context: InvocationContext) -> ConditioningCollectionOutput:
+        return ConditioningCollectionOutput(collection=self.collection)
+
+
+# endregion

invokeai/app/invocations/prompt.py

@@ -1,59 +1,28 @@
 from os.path import exists
-from typing import Literal, Optional
+from typing import Literal, Optional, Union
 
 import numpy as np
-from pydantic import Field, validator
+from dynamicprompts.generators import CombinatorialPromptGenerator, RandomPromptGenerator
+from pydantic import validator
 
-from .baseinvocation import BaseInvocation, BaseInvocationOutput, InvocationConfig, InvocationContext
-from dynamicprompts.generators import RandomPromptGenerator, CombinatorialPromptGenerator
-
-
-class PromptOutput(BaseInvocationOutput):
-    """Base class for invocations that output a prompt"""
-
-    # fmt: off
-    type: Literal["prompt"] = "prompt"
-
-    prompt: str = Field(default=None, description="The output prompt")
-    # fmt: on
-
-    class Config:
-        schema_extra = {
-            "required": [
-                "type",
-                "prompt",
-            ]
-        }
-
-
-class PromptCollectionOutput(BaseInvocationOutput):
-    """Base class for invocations that output a collection of prompts"""
-
-    # fmt: off
-    type: Literal["prompt_collection_output"] = "prompt_collection_output"
-
-    prompt_collection: list[str] = Field(description="The output prompt collection")
-    count: int = Field(description="The size of the prompt collection")
-    # fmt: on
-
-    class Config:
-        schema_extra = {"required": ["type", "prompt_collection", "count"]}
+from invokeai.app.invocations.primitives import StringCollectionOutput
+
+from .baseinvocation import BaseInvocation, InputField, InvocationContext, UIComponent, UIType, tags, title
 
 
+@title("Dynamic Prompt")
+@tags("prompt", "collection")
 class DynamicPromptInvocation(BaseInvocation):
     """Parses a prompt using adieyal/dynamicprompts' random or combinatorial generator"""
 
     type: Literal["dynamic_prompt"] = "dynamic_prompt"
-    prompt: str = Field(description="The prompt to parse with dynamicprompts")
-    max_prompts: int = Field(default=1, description="The number of prompts to generate")
-    combinatorial: bool = Field(default=False, description="Whether to use the combinatorial generator")
 
-    class Config(InvocationConfig):
-        schema_extra = {
-            "ui": {"title": "Dynamic Prompt", "tags": ["prompt", "dynamic"]},
-        }
+    # Inputs
+    prompt: str = InputField(description="The prompt to parse with dynamicprompts", ui_component=UIComponent.Textarea)
+    max_prompts: int = InputField(default=1, description="The number of prompts to generate")
+    combinatorial: bool = InputField(default=False, description="Whether to use the combinatorial generator")
 
-    def invoke(self, context: InvocationContext) -> PromptCollectionOutput:
+    def invoke(self, context: InvocationContext) -> StringCollectionOutput:
         if self.combinatorial:
             generator = CombinatorialPromptGenerator()
             prompts = generator.generate(self.prompt, max_prompts=self.max_prompts)
@@ -61,27 +30,26 @@ class DynamicPromptInvocation(BaseInvocation):
             generator = RandomPromptGenerator()
             prompts = generator.generate(self.prompt, num_images=self.max_prompts)
 
-        return PromptCollectionOutput(prompt_collection=prompts, count=len(prompts))
+        return StringCollectionOutput(collection=prompts)
 
 
+@title("Prompts from File")
+@tags("prompt", "file")
 class PromptsFromFileInvocation(BaseInvocation):
     """Loads prompts from a text file"""
 
-    # fmt: off
-    type: Literal['prompt_from_file'] = 'prompt_from_file'
+    type: Literal["prompt_from_file"] = "prompt_from_file"
 
     # Inputs
-    file_path: str = Field(description="Path to prompt text file")
-    pre_prompt: Optional[str] = Field(description="String to prepend to each prompt")
-    post_prompt: Optional[str] = Field(description="String to append to each prompt")
-    start_line: int = Field(default=1, ge=1, description="Line in the file to start start from")
-    max_prompts: int = Field(default=1, ge=0, description="Max lines to read from file (0=all)")
-    # fmt: on
-
-    class Config(InvocationConfig):
-        schema_extra = {
-            "ui": {"title": "Prompts From File", "tags": ["prompt", "file"]},
-        }
+    file_path: str = InputField(description="Path to prompt text file", ui_type=UIType.FilePath)
+    pre_prompt: Optional[str] = InputField(
+        default=None, description="String to prepend to each prompt", ui_component=UIComponent.Textarea
+    )
+    post_prompt: Optional[str] = InputField(
+        default=None, description="String to append to each prompt", ui_component=UIComponent.Textarea
+    )
+    start_line: int = InputField(default=1, ge=1, description="Line in the file to start start from")
+    max_prompts: int = InputField(default=1, ge=0, description="Max lines to read from file (0=all)")
 
     @validator("file_path")
     def file_path_exists(cls, v):
@@ -89,7 +57,14 @@ class PromptsFromFileInvocation(BaseInvocation):
             raise ValueError(FileNotFoundError)
         return v
 
-    def promptsFromFile(self, file_path: str, pre_prompt: str, post_prompt: str, start_line: int, max_prompts: int):
+    def promptsFromFile(
+        self,
+        file_path: str,
+        pre_prompt: Union[str, None],
+        post_prompt: Union[str, None],
+        start_line: int,
+        max_prompts: int,
+    ):
         prompts = []
         start_line -= 1
         end_line = start_line + max_prompts
@@ -103,8 +78,8 @@ class PromptsFromFileInvocation(BaseInvocation):
                     break
         return prompts
 
-    def invoke(self, context: InvocationContext) -> PromptCollectionOutput:
+    def invoke(self, context: InvocationContext) -> StringCollectionOutput:
         prompts = self.promptsFromFile(
             self.file_path, self.pre_prompt, self.post_prompt, self.start_line, self.max_prompts
         )
-        return PromptCollectionOutput(prompt_collection=prompts, count=len(prompts))
+        return StringCollectionOutput(collection=prompts)

invokeai/app/invocations/sdxl.py

@@ -1,62 +1,55 @@
-import torch
-import inspect
-from tqdm import tqdm
-from typing import List, Literal, Optional, Union
+from typing import Literal
 
-from pydantic import Field, validator
-
-from ...backend.model_management import ModelType, SubModelType, ModelPatcher
-from invokeai.app.util.step_callback import stable_diffusion_xl_step_callback
-from .baseinvocation import BaseInvocation, BaseInvocationOutput, InvocationConfig, InvocationContext
-
-from .model import UNetField, ClipField, VaeField, MainModelField, ModelInfo
-from .compel import ConditioningField
-from .latent import LatentsField, SAMPLER_NAME_VALUES, LatentsOutput, get_scheduler, build_latents_output
+from ...backend.model_management import ModelType, SubModelType
+from .baseinvocation import (
+    BaseInvocation,
+    BaseInvocationOutput,
+    FieldDescriptions,
+    Input,
+    InputField,
+    InvocationContext,
+    OutputField,
+    UIType,
+    tags,
+    title,
+)
+from .model import ClipField, MainModelField, ModelInfo, UNetField, VaeField
 
 
 class SDXLModelLoaderOutput(BaseInvocationOutput):
     """SDXL base model loader output"""
 
-    # fmt: off
     type: Literal["sdxl_model_loader_output"] = "sdxl_model_loader_output"
 
-    unet: UNetField = Field(default=None, description="UNet submodel")
-    clip: ClipField = Field(default=None, description="Tokenizer and text_encoder submodels")
-    clip2: ClipField = Field(default=None, description="Tokenizer and text_encoder submodels")
-    vae: VaeField = Field(default=None, description="Vae submodel")
-    # fmt: on
+    unet: UNetField = OutputField(description=FieldDescriptions.unet, title="UNet")
+    clip: ClipField = OutputField(description=FieldDescriptions.clip, title="CLIP 1")
+    clip2: ClipField = OutputField(description=FieldDescriptions.clip, title="CLIP 2")
+    vae: VaeField = OutputField(description=FieldDescriptions.vae, title="VAE")
 
 
 class SDXLRefinerModelLoaderOutput(BaseInvocationOutput):
     """SDXL refiner model loader output"""
 
-    # fmt: off
     type: Literal["sdxl_refiner_model_loader_output"] = "sdxl_refiner_model_loader_output"
-    unet: UNetField = Field(default=None, description="UNet submodel")
-    clip2: ClipField = Field(default=None, description="Tokenizer and text_encoder submodels")
-    vae: VaeField = Field(default=None, description="Vae submodel")
-    # fmt: on
-    # fmt: on
+
+    unet: UNetField = OutputField(description=FieldDescriptions.unet, title="UNet")
+    clip2: ClipField = OutputField(description=FieldDescriptions.clip, title="CLIP 2")
+    vae: VaeField = OutputField(description=FieldDescriptions.vae, title="VAE")
 
 
+@title("SDXL Main Model Loader")
+@tags("model", "sdxl")
 class SDXLModelLoaderInvocation(BaseInvocation):
     """Loads an sdxl base model, outputting its submodels."""
 
     type: Literal["sdxl_model_loader"] = "sdxl_model_loader"
 
-    model: MainModelField = Field(description="The model to load")
+    # Inputs
+    model: MainModelField = InputField(
+        description=FieldDescriptions.sdxl_main_model, input=Input.Direct, ui_type=UIType.SDXLMainModel
+    )
     # TODO: precision?
 
-    # Schema customisation
-    class Config(InvocationConfig):
-        schema_extra = {
-            "ui": {
-                "title": "SDXL Model Loader",
-                "tags": ["model", "loader", "sdxl"],
-                "type_hints": {"model": "model"},
-            },
-        }
-
     def invoke(self, context: InvocationContext) -> SDXLModelLoaderOutput:
         base_model = self.model.base_model
         model_name = self.model.model_name
@@ -129,24 +122,21 @@ class SDXLModelLoaderInvocation(BaseInvocation):
         )
 
 
+@title("SDXL Refiner Model Loader")
+@tags("model", "sdxl", "refiner")
 class SDXLRefinerModelLoaderInvocation(BaseInvocation):
     """Loads an sdxl refiner model, outputting its submodels."""
 
     type: Literal["sdxl_refiner_model_loader"] = "sdxl_refiner_model_loader"
 
-    model: MainModelField = Field(description="The model to load")
+    # Inputs
+    model: MainModelField = InputField(
+        description=FieldDescriptions.sdxl_refiner_model,
+        input=Input.Direct,
+        ui_type=UIType.SDXLRefinerModel,
+    )
     # TODO: precision?
 
-    # Schema customisation
-    class Config(InvocationConfig):
-        schema_extra = {
-            "ui": {
-                "title": "SDXL Refiner Model Loader",
-                "tags": ["model", "loader", "sdxl_refiner"],
-                "type_hints": {"model": "refiner_model"},
-            },
-        }
-
     def invoke(self, context: InvocationContext) -> SDXLRefinerModelLoaderOutput:
         base_model = self.model.base_model
         model_name = self.model.model_name
@@ -201,526 +191,3 @@ class SDXLRefinerModelLoaderInvocation(BaseInvocation):
                 ),
             ),
         )
-
-
-# Text to image
-class SDXLTextToLatentsInvocation(BaseInvocation):
-    """Generates latents from conditionings."""
-
-    type: Literal["t2l_sdxl"] = "t2l_sdxl"
-
-    # Inputs
-    # fmt: off
-    positive_conditioning: Optional[ConditioningField] = Field(description="Positive conditioning for generation")
-    negative_conditioning: Optional[ConditioningField] = Field(description="Negative conditioning for generation")
-    noise: Optional[LatentsField] = Field(description="The noise to use")
-    steps:       int = Field(default=10, gt=0, description="The number of steps to use to generate the image")
-    cfg_scale: Union[float, List[float]] = Field(default=7.5, ge=1, description="The Classifier-Free Guidance, higher values may result in a result closer to the prompt", )
-    scheduler: SAMPLER_NAME_VALUES = Field(default="euler", description="The scheduler to use" )
-    unet: UNetField = Field(default=None, description="UNet submodel")
-    denoising_end: float = Field(default=1.0, gt=0, le=1, description="")
-    # control: Union[ControlField, list[ControlField]] = Field(default=None, description="The control to use")
-    # seamless:   bool = Field(default=False, description="Whether or not to generate an image that can tile without seams", )
-    # seamless_axes: str = Field(default="", description="The axes to tile the image on, 'x' and/or 'y'")
-    # fmt: on
-
-    @validator("cfg_scale")
-    def ge_one(cls, v):
-        """validate that all cfg_scale values are >= 1"""
-        if isinstance(v, list):
-            for i in v:
-                if i < 1:
-                    raise ValueError("cfg_scale must be greater than 1")
-        else:
-            if v < 1:
-                raise ValueError("cfg_scale must be greater than 1")
-        return v
-
-    # Schema customisation
-    class Config(InvocationConfig):
-        schema_extra = {
-            "ui": {
-                "title": "SDXL Text To Latents",
-                "tags": ["latents"],
-                "type_hints": {
-                    "model": "model",
-                    # "cfg_scale": "float",
-                    "cfg_scale": "number",
-                },
-            },
-        }
-
-    def dispatch_progress(
-        self,
-        context: InvocationContext,
-        source_node_id: str,
-        sample,
-        step,
-        total_steps,
-    ) -> None:
-        stable_diffusion_xl_step_callback(
-            context=context,
-            node=self.dict(),
-            source_node_id=source_node_id,
-            sample=sample,
-            step=step,
-            total_steps=total_steps,
-        )
-
-    # based on
-    # https://github.com/huggingface/diffusers/blob/3ebbaf7c96801271f9e6c21400033b6aa5ffcf29/src/diffusers/pipelines/stable_diffusion/pipeline_onnx_stable_diffusion.py#L375
-    @torch.no_grad()
-    def invoke(self, context: InvocationContext) -> LatentsOutput:
-        graph_execution_state = context.services.graph_execution_manager.get(context.graph_execution_state_id)
-        source_node_id = graph_execution_state.prepared_source_mapping[self.id]
-        latents = context.services.latents.get(self.noise.latents_name)
-
-        positive_cond_data = context.services.latents.get(self.positive_conditioning.conditioning_name)
-        prompt_embeds = positive_cond_data.conditionings[0].embeds
-        pooled_prompt_embeds = positive_cond_data.conditionings[0].pooled_embeds
-        add_time_ids = positive_cond_data.conditionings[0].add_time_ids
-
-        negative_cond_data = context.services.latents.get(self.negative_conditioning.conditioning_name)
-        negative_prompt_embeds = negative_cond_data.conditionings[0].embeds
-        negative_pooled_prompt_embeds = negative_cond_data.conditionings[0].pooled_embeds
-        add_neg_time_ids = negative_cond_data.conditionings[0].add_time_ids
-
-        scheduler = get_scheduler(
-            context=context,
-            scheduler_info=self.unet.scheduler,
-            scheduler_name=self.scheduler,
-        )
-
-        num_inference_steps = self.steps
-
-        def _lora_loader():
-            for lora in self.unet.loras:
-                lora_info = context.services.model_manager.get_model(
-                    **lora.dict(exclude={"weight"}),
-                    context=context,
-                )
-                yield (lora_info.context.model, lora.weight)
-                del lora_info
-            return
-
-        unet_info = context.services.model_manager.get_model(**self.unet.unet.dict(), context=context)
-        do_classifier_free_guidance = True
-        cross_attention_kwargs = None
-        with ModelPatcher.apply_lora_unet(unet_info.context.model, _lora_loader()), unet_info as unet:
-            scheduler.set_timesteps(num_inference_steps, device=unet.device)
-            timesteps = scheduler.timesteps
-
-            latents = latents.to(device=unet.device, dtype=unet.dtype) * scheduler.init_noise_sigma
-
-            extra_step_kwargs = dict()
-            if "eta" in set(inspect.signature(scheduler.step).parameters.keys()):
-                extra_step_kwargs.update(
-                    eta=0.0,
-                )
-            if "generator" in set(inspect.signature(scheduler.step).parameters.keys()):
-                extra_step_kwargs.update(
-                    generator=torch.Generator(device=unet.device).manual_seed(0),
-                )
-
-            num_warmup_steps = len(timesteps) - self.steps * scheduler.order
-
-            # apply denoising_end
-            skipped_final_steps = int(round((1 - self.denoising_end) * self.steps))
-            num_inference_steps = num_inference_steps - skipped_final_steps
-            timesteps = timesteps[: num_warmup_steps + scheduler.order * num_inference_steps]
-
-            if not context.services.configuration.sequential_guidance:
-                prompt_embeds = torch.cat([negative_prompt_embeds, prompt_embeds], dim=0)
-                add_text_embeds = torch.cat([negative_pooled_prompt_embeds, pooled_prompt_embeds], dim=0)
-                add_time_ids = torch.cat([add_neg_time_ids, add_time_ids], dim=0)
-
-                prompt_embeds = prompt_embeds.to(device=unet.device, dtype=unet.dtype)
-                add_text_embeds = add_text_embeds.to(device=unet.device, dtype=unet.dtype)
-                add_time_ids = add_time_ids.to(device=unet.device, dtype=unet.dtype)
-                latents = latents.to(device=unet.device, dtype=unet.dtype)
-
-                with tqdm(total=num_inference_steps) as progress_bar:
-                    for i, t in enumerate(timesteps):
-                        # expand the latents if we are doing classifier free guidance
-                        latent_model_input = torch.cat([latents] * 2) if do_classifier_free_guidance else latents
-
-                        latent_model_input = scheduler.scale_model_input(latent_model_input, t)
-
-                        # predict the noise residual
-                        added_cond_kwargs = {"text_embeds": add_text_embeds, "time_ids": add_time_ids}
-                        noise_pred = unet(
-                            latent_model_input,
-                            t,
-                            encoder_hidden_states=prompt_embeds,
-                            cross_attention_kwargs=cross_attention_kwargs,
-                            added_cond_kwargs=added_cond_kwargs,
-                            return_dict=False,
-                        )[0]
-
-                        # perform guidance
-                        if do_classifier_free_guidance:
-                            noise_pred_uncond, noise_pred_text = noise_pred.chunk(2)
-                            noise_pred = noise_pred_uncond + self.cfg_scale * (noise_pred_text - noise_pred_uncond)
-                            # del noise_pred_uncond
-                            # del noise_pred_text
-
-                        # if do_classifier_free_guidance and guidance_rescale > 0.0:
-                        #    # Based on 3.4. in https://arxiv.org/pdf/2305.08891.pdf
-                        #    noise_pred = rescale_noise_cfg(noise_pred, noise_pred_text, guidance_rescale=guidance_rescale)
-
-                        # compute the previous noisy sample x_t -> x_t-1
-                        latents = scheduler.step(noise_pred, t, latents, **extra_step_kwargs, return_dict=False)[0]
-
-                        # call the callback, if provided
-                        if i == len(timesteps) - 1 or ((i + 1) > num_warmup_steps and (i + 1) % scheduler.order == 0):
-                            progress_bar.update()
-                            self.dispatch_progress(context, source_node_id, latents, i, num_inference_steps)
-                            # if callback is not None and i % callback_steps == 0:
-                            #    callback(i, t, latents)
-            else:
-                negative_pooled_prompt_embeds = negative_pooled_prompt_embeds.to(device=unet.device, dtype=unet.dtype)
-                negative_prompt_embeds = negative_prompt_embeds.to(device=unet.device, dtype=unet.dtype)
-                add_neg_time_ids = add_neg_time_ids.to(device=unet.device, dtype=unet.dtype)
-                pooled_prompt_embeds = pooled_prompt_embeds.to(device=unet.device, dtype=unet.dtype)
-                prompt_embeds = prompt_embeds.to(device=unet.device, dtype=unet.dtype)
-                add_time_ids = add_time_ids.to(device=unet.device, dtype=unet.dtype)
-                latents = latents.to(device=unet.device, dtype=unet.dtype)
-
-                with tqdm(total=num_inference_steps) as progress_bar:
-                    for i, t in enumerate(timesteps):
-                        # expand the latents if we are doing classifier free guidance
-                        # latent_model_input = torch.cat([latents] * 2) if do_classifier_free_guidance else latents
-
-                        latent_model_input = scheduler.scale_model_input(latents, t)
-
-                        # import gc
-                        # gc.collect()
-                        # torch.cuda.empty_cache()
-
-                        # predict the noise residual
-
-                        added_cond_kwargs = {"text_embeds": negative_pooled_prompt_embeds, "time_ids": add_neg_time_ids}
-                        noise_pred_uncond = unet(
-                            latent_model_input,
-                            t,
-                            encoder_hidden_states=negative_prompt_embeds,
-                            cross_attention_kwargs=cross_attention_kwargs,
-                            added_cond_kwargs=added_cond_kwargs,
-                            return_dict=False,
-                        )[0]
-
-                        added_cond_kwargs = {"text_embeds": pooled_prompt_embeds, "time_ids": add_time_ids}
-                        noise_pred_text = unet(
-                            latent_model_input,
-                            t,
-                            encoder_hidden_states=prompt_embeds,
-                            cross_attention_kwargs=cross_attention_kwargs,
-                            added_cond_kwargs=added_cond_kwargs,
-                            return_dict=False,
-                        )[0]
-
-                        # perform guidance
-                        noise_pred = noise_pred_uncond + self.cfg_scale * (noise_pred_text - noise_pred_uncond)
-
-                        # del noise_pred_text
-                        # del noise_pred_uncond
-                        # import gc
-                        # gc.collect()
-                        # torch.cuda.empty_cache()
-
-                        # if do_classifier_free_guidance and guidance_rescale > 0.0:
-                        #    # Based on 3.4. in https://arxiv.org/pdf/2305.08891.pdf
-                        #    noise_pred = rescale_noise_cfg(noise_pred, noise_pred_text, guidance_rescale=guidance_rescale)
-
-                        # compute the previous noisy sample x_t -> x_t-1
-                        latents = scheduler.step(noise_pred, t, latents, **extra_step_kwargs, return_dict=False)[0]
-
-                        # del noise_pred
-                        # import gc
-                        # gc.collect()
-                        # torch.cuda.empty_cache()
-
-                        # call the callback, if provided
-                        if i == len(timesteps) - 1 or ((i + 1) > num_warmup_steps and (i + 1) % scheduler.order == 0):
-                            progress_bar.update()
-                            self.dispatch_progress(context, source_node_id, latents, i, num_inference_steps)
-                            # if callback is not None and i % callback_steps == 0:
-                            #    callback(i, t, latents)
-
-        #################
-
-        latents = latents.to("cpu")
-        torch.cuda.empty_cache()
-
-        name = f"{context.graph_execution_state_id}__{self.id}"
-        context.services.latents.save(name, latents)
-        return build_latents_output(latents_name=name, latents=latents)
-
-
-class SDXLLatentsToLatentsInvocation(BaseInvocation):
-    """Generates latents from conditionings."""
-
-    type: Literal["l2l_sdxl"] = "l2l_sdxl"
-
-    # Inputs
-    # fmt: off
-    positive_conditioning: Optional[ConditioningField] = Field(description="Positive conditioning for generation")
-    negative_conditioning: Optional[ConditioningField] = Field(description="Negative conditioning for generation")
-    noise: Optional[LatentsField] = Field(description="The noise to use")
-    steps:       int = Field(default=10, gt=0, description="The number of steps to use to generate the image")
-    cfg_scale: Union[float, List[float]] = Field(default=7.5, ge=1, description="The Classifier-Free Guidance, higher values may result in a result closer to the prompt", )
-    scheduler: SAMPLER_NAME_VALUES = Field(default="euler", description="The scheduler to use" )
-    unet: UNetField = Field(default=None, description="UNet submodel")
-    latents: Optional[LatentsField] = Field(description="Initial latents")
-
-    denoising_start: float = Field(default=0.0, ge=0, le=1, description="")
-    denoising_end: float = Field(default=1.0, ge=0, le=1, description="")
-
-    # control: Union[ControlField, list[ControlField]] = Field(default=None, description="The control to use")
-    # seamless:   bool = Field(default=False, description="Whether or not to generate an image that can tile without seams", )
-    # seamless_axes: str = Field(default="", description="The axes to tile the image on, 'x' and/or 'y'")
-    # fmt: on
-
-    @validator("cfg_scale")
-    def ge_one(cls, v):
-        """validate that all cfg_scale values are >= 1"""
-        if isinstance(v, list):
-            for i in v:
-                if i < 1:
-                    raise ValueError("cfg_scale must be greater than 1")
-        else:
-            if v < 1:
-                raise ValueError("cfg_scale must be greater than 1")
-        return v
-
-    # Schema customisation
-    class Config(InvocationConfig):
-        schema_extra = {
-            "ui": {
-                "title": "SDXL Latents to Latents",
-                "tags": ["latents"],
-                "type_hints": {
-                    "model": "model",
-                    # "cfg_scale": "float",
-                    "cfg_scale": "number",
-                },
-            },
-        }
-
-    def dispatch_progress(
-        self,
-        context: InvocationContext,
-        source_node_id: str,
-        sample,
-        step,
-        total_steps,
-    ) -> None:
-        stable_diffusion_xl_step_callback(
-            context=context,
-            node=self.dict(),
-            source_node_id=source_node_id,
-            sample=sample,
-            step=step,
-            total_steps=total_steps,
-        )
-
-    # based on
-    # https://github.com/huggingface/diffusers/blob/3ebbaf7c96801271f9e6c21400033b6aa5ffcf29/src/diffusers/pipelines/stable_diffusion/pipeline_onnx_stable_diffusion.py#L375
-    @torch.no_grad()
-    def invoke(self, context: InvocationContext) -> LatentsOutput:
-        graph_execution_state = context.services.graph_execution_manager.get(context.graph_execution_state_id)
-        source_node_id = graph_execution_state.prepared_source_mapping[self.id]
-        latents = context.services.latents.get(self.latents.latents_name)
-
-        positive_cond_data = context.services.latents.get(self.positive_conditioning.conditioning_name)
-        prompt_embeds = positive_cond_data.conditionings[0].embeds
-        pooled_prompt_embeds = positive_cond_data.conditionings[0].pooled_embeds
-        add_time_ids = positive_cond_data.conditionings[0].add_time_ids
-
-        negative_cond_data = context.services.latents.get(self.negative_conditioning.conditioning_name)
-        negative_prompt_embeds = negative_cond_data.conditionings[0].embeds
-        negative_pooled_prompt_embeds = negative_cond_data.conditionings[0].pooled_embeds
-        add_neg_time_ids = negative_cond_data.conditionings[0].add_time_ids
-
-        scheduler = get_scheduler(
-            context=context,
-            scheduler_info=self.unet.scheduler,
-            scheduler_name=self.scheduler,
-        )
-
-        unet_info = context.services.model_manager.get_model(
-            **self.unet.unet.dict(),
-            context=context,
-        )
-
-        def _lora_loader():
-            for lora in self.unet.loras:
-                lora_info = context.services.model_manager.get_model(
-                    **lora.dict(exclude={"weight"}),
-                    context=context,
-                )
-                yield (lora_info.context.model, lora.weight)
-                del lora_info
-            return
-
-        do_classifier_free_guidance = True
-        cross_attention_kwargs = None
-        with ModelPatcher.apply_lora_unet(unet_info.context.model, _lora_loader()), unet_info as unet:
-            # apply denoising_start
-            num_inference_steps = self.steps
-            scheduler.set_timesteps(num_inference_steps, device=unet.device)
-
-            t_start = int(round(self.denoising_start * num_inference_steps))
-            timesteps = scheduler.timesteps[t_start * scheduler.order :]
-            num_inference_steps = num_inference_steps - t_start
-
-            # apply noise(if provided)
-            if self.noise is not None and timesteps.shape[0] > 0:
-                noise = context.services.latents.get(self.noise.latents_name)
-                latents = scheduler.add_noise(latents, noise, timesteps[:1])
-                del noise
-
-            # apply scheduler extra args
-            extra_step_kwargs = dict()
-            if "eta" in set(inspect.signature(scheduler.step).parameters.keys()):
-                extra_step_kwargs.update(
-                    eta=0.0,
-                )
-            if "generator" in set(inspect.signature(scheduler.step).parameters.keys()):
-                extra_step_kwargs.update(
-                    generator=torch.Generator(device=unet.device).manual_seed(0),
-                )
-
-            num_warmup_steps = max(len(timesteps) - num_inference_steps * scheduler.order, 0)
-
-            # apply denoising_end
-            skipped_final_steps = int(round((1 - self.denoising_end) * self.steps))
-            num_inference_steps = num_inference_steps - skipped_final_steps
-            timesteps = timesteps[: num_warmup_steps + scheduler.order * num_inference_steps]
-
-            if not context.services.configuration.sequential_guidance:
-                prompt_embeds = torch.cat([negative_prompt_embeds, prompt_embeds], dim=0)
-                add_text_embeds = torch.cat([negative_pooled_prompt_embeds, pooled_prompt_embeds], dim=0)
-                add_time_ids = torch.cat([add_neg_time_ids, add_time_ids], dim=0)
-
-                prompt_embeds = prompt_embeds.to(device=unet.device, dtype=unet.dtype)
-                add_text_embeds = add_text_embeds.to(device=unet.device, dtype=unet.dtype)
-                add_time_ids = add_time_ids.to(device=unet.device, dtype=unet.dtype)
-                latents = latents.to(device=unet.device, dtype=unet.dtype)
-
-                with tqdm(total=num_inference_steps) as progress_bar:
-                    for i, t in enumerate(timesteps):
-                        # expand the latents if we are doing classifier free guidance
-                        latent_model_input = torch.cat([latents] * 2) if do_classifier_free_guidance else latents
-
-                        latent_model_input = scheduler.scale_model_input(latent_model_input, t)
-
-                        # predict the noise residual
-                        added_cond_kwargs = {"text_embeds": add_text_embeds, "time_ids": add_time_ids}
-                        noise_pred = unet(
-                            latent_model_input,
-                            t,
-                            encoder_hidden_states=prompt_embeds,
-                            cross_attention_kwargs=cross_attention_kwargs,
-                            added_cond_kwargs=added_cond_kwargs,
-                            return_dict=False,
-                        )[0]
-
-                        # perform guidance
-                        if do_classifier_free_guidance:
-                            noise_pred_uncond, noise_pred_text = noise_pred.chunk(2)
-                            noise_pred = noise_pred_uncond + self.cfg_scale * (noise_pred_text - noise_pred_uncond)
-                            # del noise_pred_uncond
-                            # del noise_pred_text
-
-                        # if do_classifier_free_guidance and guidance_rescale > 0.0:
-                        #    # Based on 3.4. in https://arxiv.org/pdf/2305.08891.pdf
-                        #    noise_pred = rescale_noise_cfg(noise_pred, noise_pred_text, guidance_rescale=guidance_rescale)
-
-                        # compute the previous noisy sample x_t -> x_t-1
-                        latents = scheduler.step(noise_pred, t, latents, **extra_step_kwargs, return_dict=False)[0]
-
-                        # call the callback, if provided
-                        if i == len(timesteps) - 1 or ((i + 1) > num_warmup_steps and (i + 1) % scheduler.order == 0):
-                            progress_bar.update()
-                            self.dispatch_progress(context, source_node_id, latents, i, num_inference_steps)
-                            # if callback is not None and i % callback_steps == 0:
-                            #    callback(i, t, latents)
-            else:
-                negative_pooled_prompt_embeds = negative_pooled_prompt_embeds.to(device=unet.device, dtype=unet.dtype)
-                negative_prompt_embeds = negative_prompt_embeds.to(device=unet.device, dtype=unet.dtype)
-                add_neg_time_ids = add_neg_time_ids.to(device=unet.device, dtype=unet.dtype)
-                pooled_prompt_embeds = pooled_prompt_embeds.to(device=unet.device, dtype=unet.dtype)
-                prompt_embeds = prompt_embeds.to(device=unet.device, dtype=unet.dtype)
-                add_time_ids = add_time_ids.to(device=unet.device, dtype=unet.dtype)
-                latents = latents.to(device=unet.device, dtype=unet.dtype)
-
-                with tqdm(total=num_inference_steps) as progress_bar:
-                    for i, t in enumerate(timesteps):
-                        # expand the latents if we are doing classifier free guidance
-                        # latent_model_input = torch.cat([latents] * 2) if do_classifier_free_guidance else latents
-
-                        latent_model_input = scheduler.scale_model_input(latents, t)
-
-                        # import gc
-                        # gc.collect()
-                        # torch.cuda.empty_cache()
-
-                        # predict the noise residual
-
-                        added_cond_kwargs = {"text_embeds": negative_pooled_prompt_embeds, "time_ids": add_time_ids}
-                        noise_pred_uncond = unet(
-                            latent_model_input,
-                            t,
-                            encoder_hidden_states=negative_prompt_embeds,
-                            cross_attention_kwargs=cross_attention_kwargs,
-                            added_cond_kwargs=added_cond_kwargs,
-                            return_dict=False,
-                        )[0]
-
-                        added_cond_kwargs = {"text_embeds": pooled_prompt_embeds, "time_ids": add_time_ids}
-                        noise_pred_text = unet(
-                            latent_model_input,
-                            t,
-                            encoder_hidden_states=prompt_embeds,
-                            cross_attention_kwargs=cross_attention_kwargs,
-                            added_cond_kwargs=added_cond_kwargs,
-                            return_dict=False,
-                        )[0]
-
-                        # perform guidance
-                        noise_pred = noise_pred_uncond + self.cfg_scale * (noise_pred_text - noise_pred_uncond)
-
-                        # del noise_pred_text
-                        # del noise_pred_uncond
-                        # import gc
-                        # gc.collect()
-                        # torch.cuda.empty_cache()
-
-                        # if do_classifier_free_guidance and guidance_rescale > 0.0:
-                        #    # Based on 3.4. in https://arxiv.org/pdf/2305.08891.pdf
-                        #    noise_pred = rescale_noise_cfg(noise_pred, noise_pred_text, guidance_rescale=guidance_rescale)
-
-                        # compute the previous noisy sample x_t -> x_t-1
-                        latents = scheduler.step(noise_pred, t, latents, **extra_step_kwargs, return_dict=False)[0]
-
-                        # del noise_pred
-                        # import gc
-                        # gc.collect()
-                        # torch.cuda.empty_cache()
-
-                        # call the callback, if provided
-                        if i == len(timesteps) - 1 or ((i + 1) > num_warmup_steps and (i + 1) % scheduler.order == 0):
-                            progress_bar.update()
-                            self.dispatch_progress(context, source_node_id, latents, i, num_inference_steps)
-                            # if callback is not None and i % callback_steps == 0:
-                            #    callback(i, t, latents)
-
-        #################
-
-        latents = latents.to("cpu")
-        torch.cuda.empty_cache()
-
-        name = f"{context.graph_execution_state_id}__{self.id}"
-        context.services.latents.save(name, latents)
-        return build_latents_output(latents_name=name, latents=latents)

invokeai/app/invocations/upscale.py

@@ -6,13 +6,12 @@ import cv2 as cv
 import numpy as np
 from basicsr.archs.rrdbnet_arch import RRDBNet
 from PIL import Image
-from pydantic import Field
 from realesrgan import RealESRGANer
+from invokeai.app.invocations.primitives import ImageField, ImageOutput
 
-from invokeai.app.models.image import ImageCategory, ImageField, ResourceOrigin
+from invokeai.app.models.image import ImageCategory, ResourceOrigin
 
-from .baseinvocation import BaseInvocation, InvocationConfig, InvocationContext
-from .image import ImageOutput
+from .baseinvocation import BaseInvocation, InputField, InvocationContext, title, tags
 
 # TODO: Populate this from disk?
 # TODO: Use model manager to load?
@@ -24,17 +23,16 @@ ESRGAN_MODELS = Literal[
 ]
 
 
+@title("Upscale (RealESRGAN)")
+@tags("esrgan", "upscale")
 class ESRGANInvocation(BaseInvocation):
     """Upscales an image using RealESRGAN."""
 
     type: Literal["esrgan"] = "esrgan"
-    image: Union[ImageField, None] = Field(default=None, description="The input image")
-    model_name: ESRGAN_MODELS = Field(default="RealESRGAN_x4plus.pth", description="The Real-ESRGAN model to use")
 
-    class Config(InvocationConfig):
-        schema_extra = {
-            "ui": {"title": "Upscale (RealESRGAN)", "tags": ["image", "upscale", "realesrgan"]},
-        }
+    # Inputs
+    image: ImageField = InputField(description="The input image")
+    model_name: ESRGAN_MODELS = InputField(default="RealESRGAN_x4plus.pth", description="The Real-ESRGAN model to use")
 
     def invoke(self, context: InvocationContext) -> ImageOutput:
         image = context.services.images.get_pil_image(self.image.image_name)

invokeai/app/models/image.py

@@ -1,31 +1,8 @@
 from enum import Enum
-from typing import Optional, Tuple, Literal
+
 from pydantic import BaseModel, Field
 
 from invokeai.app.util.metaenum import MetaEnum
-from ..invocations.baseinvocation import (
-    BaseInvocationOutput,
-    InvocationConfig,
-)
-
-
-class ImageField(BaseModel):
-    """An image field used for passing image objects between invocations"""
-
-    image_name: Optional[str] = Field(default=None, description="The name of the image")
-
-    class Config:
-        schema_extra = {"required": ["image_name"]}
-
-
-class ColorField(BaseModel):
-    r: int = Field(ge=0, le=255, description="The red component")
-    g: int = Field(ge=0, le=255, description="The green component")
-    b: int = Field(ge=0, le=255, description="The blue component")
-    a: int = Field(ge=0, le=255, description="The alpha component")
-
-    def tuple(self) -> Tuple[int, int, int, int]:
-        return (self.r, self.g, self.b, self.a)
 
 
 class ProgressImage(BaseModel):
@@ -36,50 +13,6 @@ class ProgressImage(BaseModel):
     dataURL: str = Field(description="The image data as a b64 data URL")
 
 
-class PILInvocationConfig(BaseModel):
-    """Helper class to provide all PIL invocations with additional config"""
-
-    class Config(InvocationConfig):
-        schema_extra = {
-            "ui": {
-                "tags": ["PIL", "image"],
-            },
-        }
-
-
-class ImageOutput(BaseInvocationOutput):
-    """Base class for invocations that output an image"""
-
-    # fmt: off
-    type: Literal["image_output"] = "image_output"
-    image:      ImageField = Field(default=None, description="The output image")
-    width:             int = Field(description="The width of the image in pixels")
-    height:            int = Field(description="The height of the image in pixels")
-    # fmt: on
-
-    class Config:
-        schema_extra = {"required": ["type", "image", "width", "height"]}
-
-
-class MaskOutput(BaseInvocationOutput):
-    """Base class for invocations that output a mask"""
-
-    # fmt: off
-    type: Literal["mask"] = "mask"
-    mask:      ImageField = Field(default=None, description="The output mask")
-    width:            int = Field(description="The width of the mask in pixels")
-    height:           int = Field(description="The height of the mask in pixels")
-    # fmt: on
-
-    class Config:
-        schema_extra = {
-            "required": [
-                "type",
-                "mask",
-            ]
-        }
-
-
 class ResourceOrigin(str, Enum, metaclass=MetaEnum):
     """The origin of a resource (eg image).
 

invokeai/app/services/default_graphs.py

@@ -1,8 +1,8 @@
-from ..invocations.latent import LatentsToImageInvocation, TextToLatentsInvocation
+from ..invocations.latent import LatentsToImageInvocation, DenoiseLatentsInvocation
 from ..invocations.image import ImageNSFWBlurInvocation
 from ..invocations.noise import NoiseInvocation
 from ..invocations.compel import CompelInvocation
-from ..invocations.params import ParamIntInvocation
+from ..invocations.primitives import IntegerInvocation
 from .graph import Edge, EdgeConnection, ExposedNodeInput, ExposedNodeOutput, Graph, LibraryGraph
 from .item_storage import ItemStorageABC
 
@@ -17,13 +17,13 @@ def create_text_to_image() -> LibraryGraph:
         description="Converts text to an image",
         graph=Graph(
             nodes={
-                "width": ParamIntInvocation(id="width", a=512),
-                "height": ParamIntInvocation(id="height", a=512),
-                "seed": ParamIntInvocation(id="seed", a=-1),
+                "width": IntegerInvocation(id="width", a=512),
+                "height": IntegerInvocation(id="height", a=512),
+                "seed": IntegerInvocation(id="seed", a=-1),
                 "3": NoiseInvocation(id="3"),
                 "4": CompelInvocation(id="4"),
                 "5": CompelInvocation(id="5"),
-                "6": TextToLatentsInvocation(id="6"),
+                "6": DenoiseLatentsInvocation(id="6"),
                 "7": LatentsToImageInvocation(id="7"),
                 "8": ImageNSFWBlurInvocation(id="8"),
             },

invokeai/app/services/events.py

@@ -35,6 +35,7 @@ class EventServiceBase:
         source_node_id: str,
         progress_image: Optional[ProgressImage],
         step: int,
+        order: int,
         total_steps: int,
     ) -> None:
         """Emitted when there is generation progress"""
@@ -46,6 +47,7 @@ class EventServiceBase:
                 source_node_id=source_node_id,
                 progress_image=progress_image.dict() if progress_image is not None else None,
                 step=step,
+                order=order,
                 total_steps=total_steps,
             ),
         )

invokeai/app/services/graph.py

@@ -3,16 +3,7 @@
 import copy
 import itertools
 import uuid
-from typing import (
-    Annotated,
-    Any,
-    Literal,
-    Optional,
-    Union,
-    get_args,
-    get_origin,
-    get_type_hints,
-)
+from typing import Annotated, Any, Literal, Optional, Union, get_args, get_origin, get_type_hints
 
 import networkx as nx
 from pydantic import BaseModel, root_validator, validator
@@ -22,7 +13,11 @@ from ..invocations import *
 from ..invocations.baseinvocation import (
     BaseInvocation,
     BaseInvocationOutput,
+    Input,
+    InputField,
     InvocationContext,
+    OutputField,
+    UIType,
 )
 
 # in 3.10 this would be "from types import NoneType"
@@ -183,15 +178,9 @@ class IterateInvocationOutput(BaseInvocationOutput):
 
     type: Literal["iterate_output"] = "iterate_output"
 
-    item: Any = Field(description="The item being iterated over")
-
-    class Config:
-        schema_extra = {
-            "required": [
-                "type",
-                "item",
-            ]
-        }
+    item: Any = OutputField(
+        description="The item being iterated over", title="Collection Item", ui_type=UIType.CollectionItem
+    )
 
 
 # TODO: Fill this out and move to invocations
@@ -200,8 +189,10 @@ class IterateInvocation(BaseInvocation):
 
     type: Literal["iterate"] = "iterate"
 
-    collection: list[Any] = Field(description="The list of items to iterate over", default_factory=list)
-    index: int = Field(description="The index, will be provided on executed iterators", default=0)
+    collection: list[Any] = InputField(
+        description="The list of items to iterate over", default_factory=list, ui_type=UIType.Collection
+    )
+    index: int = InputField(description="The index, will be provided on executed iterators", default=0, ui_hidden=True)
 
     def invoke(self, context: InvocationContext) -> IterateInvocationOutput:
         """Produces the outputs as values"""
@@ -211,15 +202,9 @@ class IterateInvocation(BaseInvocation):
 class CollectInvocationOutput(BaseInvocationOutput):
     type: Literal["collect_output"] = "collect_output"
 
-    collection: list[Any] = Field(description="The collection of input items")
-
-    class Config:
-        schema_extra = {
-            "required": [
-                "type",
-                "collection",
-            ]
-        }
+    collection: list[Any] = OutputField(
+        description="The collection of input items", title="Collection", ui_type=UIType.Collection
+    )
 
 
 class CollectInvocation(BaseInvocation):
@@ -227,13 +212,14 @@ class CollectInvocation(BaseInvocation):
 
     type: Literal["collect"] = "collect"
 
-    item: Any = Field(
+    item: Any = InputField(
         description="The item to collect (all inputs must be of the same type)",
-        default=None,
+        ui_type=UIType.CollectionItem,
+        title="Collection Item",
+        input=Input.Connection,
     )
-    collection: list[Any] = Field(
-        description="The collection, will be provided on execution",
-        default_factory=list,
+    collection: list[Any] = InputField(
+        description="The collection, will be provided on execution", default_factory=list, ui_hidden=True
     )
 
     def invoke(self, context: InvocationContext) -> CollectInvocationOutput:

invokeai/app/services/image_record_storage.py

@@ -67,6 +67,7 @@ IMAGE_DTO_COLS = ", ".join(
                 "created_at",
                 "updated_at",
                 "deleted_at",
+                "starred",
             ],
         )
     )
@@ -139,6 +140,7 @@ class ImageRecordStorageBase(ABC):
         node_id: Optional[str],
         metadata: Optional[dict],
         is_intermediate: bool = False,
+        starred: bool = False,
     ) -> datetime:
         """Saves an image record."""
         pass
@@ -200,6 +202,16 @@ class SqliteImageRecordStorage(ImageRecordStorageBase):
             """
         )
 
+        self._cursor.execute("PRAGMA table_info(images)")
+        columns = [column[1] for column in self._cursor.fetchall()]
+
+        if "starred" not in columns:
+            self._cursor.execute(
+                """--sql
+                ALTER TABLE images ADD COLUMN starred BOOLEAN DEFAULT FALSE;
+                """
+            )
+
         # Create the `images` table indices.
         self._cursor.execute(
             """--sql
@@ -222,6 +234,12 @@ class SqliteImageRecordStorage(ImageRecordStorageBase):
             """
         )
 
+        self._cursor.execute(
+            """--sql
+            CREATE INDEX IF NOT EXISTS idx_images_starred ON images(starred);
+            """
+        )
+
         # Add trigger for `updated_at`.
         self._cursor.execute(
             """--sql
@@ -321,6 +339,17 @@ class SqliteImageRecordStorage(ImageRecordStorageBase):
                     (changes.is_intermediate, image_name),
                 )
 
+            # Change the image's `starred`` state
+            if changes.starred is not None:
+                self._cursor.execute(
+                    f"""--sql
+                    UPDATE images
+                    SET starred = ?
+                    WHERE image_name = ?;
+                    """,
+                    (changes.starred, image_name),
+                )
+
             self._conn.commit()
         except sqlite3.Error as e:
             self._conn.rollback()
@@ -397,7 +426,7 @@ class SqliteImageRecordStorage(ImageRecordStorageBase):
                 query_params.append(board_id)
 
             query_pagination = """--sql
-            ORDER BY images.created_at DESC LIMIT ? OFFSET ?
+            ORDER BY images.starred DESC, images.created_at DESC LIMIT ? OFFSET ?
             """
 
             # Final images query with pagination
@@ -500,6 +529,7 @@ class SqliteImageRecordStorage(ImageRecordStorageBase):
         node_id: Optional[str],
         metadata: Optional[dict],
         is_intermediate: bool = False,
+        starred: bool = False,
     ) -> datetime:
         try:
             metadata_json = None if metadata is None else json.dumps(metadata)
@@ -515,9 +545,10 @@ class SqliteImageRecordStorage(ImageRecordStorageBase):
                     node_id,
                     session_id,
                     metadata,
-                    is_intermediate
+                    is_intermediate,
+                    starred
                     )
-                VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?);
+                VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?, ?);
                 """,
                 (
                     image_name,
@@ -529,6 +560,7 @@ class SqliteImageRecordStorage(ImageRecordStorageBase):
                     session_id,
                     metadata_json,
                     is_intermediate,
+                    starred,
                 ),
             )
             self._conn.commit()

invokeai/app/services/invocation_stats.py

@@ -29,6 +29,7 @@ The abstract base class for this class is InvocationStatsServiceBase. An impleme
 writes to the system log is stored in InvocationServices.performance_statistics.
 """
 
+import psutil
 import time
 from abc import ABC, abstractmethod
 from contextlib import AbstractContextManager
@@ -42,6 +43,11 @@ import invokeai.backend.util.logging as logger
 from ..invocations.baseinvocation import BaseInvocation
 from .graph import GraphExecutionState
 from .item_storage import ItemStorageABC
+from .model_manager_service import ModelManagerService
+from invokeai.backend.model_management.model_cache import CacheStats
+
+# size of GIG in bytes
+GIG = 1073741824
 
 
 class InvocationStatsServiceBase(ABC):
@@ -89,6 +95,8 @@ class InvocationStatsServiceBase(ABC):
         invocation_type: str,
         time_used: float,
         vram_used: float,
+        ram_used: float,
+        ram_changed: float,
     ):
         """
         Add timing information on execution of a node. Usually
@@ -97,6 +105,8 @@ class InvocationStatsServiceBase(ABC):
         :param invocation_type: String literal type of the node
         :param time_used: Time used by node's exection (sec)
         :param vram_used: Maximum VRAM used during exection (GB)
+        :param ram_used: Current RAM available (GB)
+        :param ram_changed: Change in RAM usage over course of the run (GB)
         """
         pass
 
@@ -115,6 +125,9 @@ class NodeStats:
     calls: int = 0
     time_used: float = 0.0  # seconds
     max_vram: float = 0.0  # GB
+    cache_hits: int = 0
+    cache_misses: int = 0
+    cache_high_watermark: int = 0
 
 
 @dataclass
@@ -133,31 +146,62 @@ class InvocationStatsService(InvocationStatsServiceBase):
         self.graph_execution_manager = graph_execution_manager
         # {graph_id => NodeLog}
         self._stats: Dict[str, NodeLog] = {}
+        self._cache_stats: Dict[str, CacheStats] = {}
+        self.ram_used: float = 0.0
+        self.ram_changed: float = 0.0
 
     class StatsContext:
-        def __init__(self, invocation: BaseInvocation, graph_id: str, collector: "InvocationStatsServiceBase"):
+        """Context manager for collecting statistics."""
+
+        invocation: BaseInvocation = None
+        collector: "InvocationStatsServiceBase" = None
+        graph_id: str = None
+        start_time: int = 0
+        ram_used: int = 0
+        model_manager: ModelManagerService = None
+
+        def __init__(
+            self,
+            invocation: BaseInvocation,
+            graph_id: str,
+            model_manager: ModelManagerService,
+            collector: "InvocationStatsServiceBase",
+        ):
+            """Initialize statistics for this run."""
             self.invocation = invocation
             self.collector = collector
             self.graph_id = graph_id
             self.start_time = 0
+            self.ram_used = 0
+            self.model_manager = model_manager
 
         def __enter__(self):
             self.start_time = time.time()
             if torch.cuda.is_available():
                 torch.cuda.reset_peak_memory_stats()
+            self.ram_used = psutil.Process().memory_info().rss
+            if self.model_manager:
+                self.model_manager.collect_cache_stats(self.collector._cache_stats[self.graph_id])
 
         def __exit__(self, *args):
+            """Called on exit from the context."""
+            ram_used = psutil.Process().memory_info().rss
+            self.collector.update_mem_stats(
+                ram_used=ram_used / GIG,
+                ram_changed=(ram_used - self.ram_used) / GIG,
+            )
             self.collector.update_invocation_stats(
-                self.graph_id,
-                self.invocation.type,
-                time.time() - self.start_time,
-                torch.cuda.max_memory_allocated() / 1e9 if torch.cuda.is_available() else 0.0,
+                graph_id=self.graph_id,
+                invocation_type=self.invocation.type,
+                time_used=time.time() - self.start_time,
+                vram_used=torch.cuda.max_memory_allocated() / GIG if torch.cuda.is_available() else 0.0,
             )
 
     def collect_stats(
         self,
         invocation: BaseInvocation,
         graph_execution_state_id: str,
+        model_manager: ModelManagerService,
     ) -> StatsContext:
         """
         Return a context object that will capture the statistics.
@@ -166,7 +210,8 @@ class InvocationStatsService(InvocationStatsServiceBase):
         """
         if not self._stats.get(graph_execution_state_id):  # first time we're seeing this
             self._stats[graph_execution_state_id] = NodeLog()
-        return self.StatsContext(invocation, graph_execution_state_id, self)
+            self._cache_stats[graph_execution_state_id] = CacheStats()
+        return self.StatsContext(invocation, graph_execution_state_id, model_manager, self)
 
     def reset_all_stats(self):
         """Zero all statistics"""
@@ -179,13 +224,36 @@ class InvocationStatsService(InvocationStatsServiceBase):
         except KeyError:
             logger.warning(f"Attempted to clear statistics for unknown graph {graph_execution_id}")
 
-    def update_invocation_stats(self, graph_id: str, invocation_type: str, time_used: float, vram_used: float):
+    def update_mem_stats(
+        self,
+        ram_used: float,
+        ram_changed: float,
+    ):
+        """
+        Update the collector with RAM memory usage info.
+
+        :param ram_used: How much RAM is currently in use.
+        :param ram_changed: How much RAM changed since last generation.
+        """
+        self.ram_used = ram_used
+        self.ram_changed = ram_changed
+
+    def update_invocation_stats(
+        self,
+        graph_id: str,
+        invocation_type: str,
+        time_used: float,
+        vram_used: float,
+    ):
         """
         Add timing information on execution of a node. Usually
         used internally.
         :param graph_id: ID of the graph that is currently executing
         :param invocation_type: String literal type of the node
-        :param time_used: Floating point seconds used by node's exection
+        :param time_used: Time used by node's exection (sec)
+        :param vram_used: Maximum VRAM used during exection (GB)
+        :param ram_used: Current RAM available (GB)
+        :param ram_changed: Change in RAM usage over course of the run (GB)
         """
         if not self._stats[graph_id].nodes.get(invocation_type):
             self._stats[graph_id].nodes[invocation_type] = NodeStats()
@@ -197,7 +265,7 @@ class InvocationStatsService(InvocationStatsServiceBase):
     def log_stats(self):
         """
         Send the statistics to the system logger at the info level.
-        Stats will only be printed if when the execution of the graph
+        Stats will only be printed when the execution of the graph
         is complete.
         """
         completed = set()
@@ -208,16 +276,30 @@ class InvocationStatsService(InvocationStatsServiceBase):
 
             total_time = 0
             logger.info(f"Graph stats: {graph_id}")
-            logger.info("Node                 Calls    Seconds VRAM Used")
+            logger.info(f"{'Node':>30} {'Calls':>7}{'Seconds':>9} {'VRAM Used':>10}")
             for node_type, stats in self._stats[graph_id].nodes.items():
-                logger.info(f"{node_type:<20} {stats.calls:>5}   {stats.time_used:7.3f}s     {stats.max_vram:4.2f}G")
+                logger.info(f"{node_type:>30}  {stats.calls:>4}   {stats.time_used:7.3f}s     {stats.max_vram:4.3f}G")
                 total_time += stats.time_used
 
+            cache_stats = self._cache_stats[graph_id]
+            hwm = cache_stats.high_watermark / GIG
+            tot = cache_stats.cache_size / GIG
+            loaded = sum([v for v in cache_stats.loaded_model_sizes.values()]) / GIG
+
             logger.info(f"TOTAL GRAPH EXECUTION TIME:  {total_time:7.3f}s")
+            logger.info("RAM used by InvokeAI process: " + "%4.2fG" % self.ram_used + f" ({self.ram_changed:+5.3f}G)")
+            logger.info(f"RAM used to load models: {loaded:4.2f}G")
             if torch.cuda.is_available():
-                logger.info("Current VRAM utilization " + "%4.2fG" % (torch.cuda.memory_allocated() / 1e9))
+                logger.info("VRAM in use: " + "%4.3fG" % (torch.cuda.memory_allocated() / GIG))
+            logger.info("RAM cache statistics:")
+            logger.info(f"   Model cache hits: {cache_stats.hits}")
+            logger.info(f"   Model cache misses: {cache_stats.misses}")
+            logger.info(f"   Models cached: {cache_stats.in_cache}")
+            logger.info(f"   Models cleared from cache: {cache_stats.cleared}")
+            logger.info(f"   Cache high water mark: {hwm:4.2f}/{tot:4.2f}G")
 
             completed.add(graph_id)
 
         for graph_id in completed:
             del self._stats[graph_id]
+            del self._cache_stats[graph_id]

invokeai/app/services/model_manager_service.py

@@ -22,6 +22,7 @@ from invokeai.backend.model_management import (
     ModelNotFoundException,
 )
 from invokeai.backend.model_management.model_search import FindModels
+from invokeai.backend.model_management.model_cache import CacheStats
 
 import torch
 from invokeai.app.models.exceptions import CanceledException
@@ -276,6 +277,13 @@ class ModelManagerServiceBase(ABC):
         """
         pass
 
+    @abstractmethod
+    def collect_cache_stats(self, cache_stats: CacheStats):
+        """
+        Reset model cache statistics for graph with graph_id.
+        """
+        pass
+
     @abstractmethod
     def commit(self, conf_file: Optional[Path] = None) -> None:
         """
@@ -500,6 +508,12 @@ class ModelManagerService(ModelManagerServiceBase):
         self.logger.debug(f"convert model {model_name}")
         return self.mgr.convert_model(model_name, base_model, model_type, convert_dest_directory)
 
+    def collect_cache_stats(self, cache_stats: CacheStats):
+        """
+        Reset model cache statistics for graph with graph_id.
+        """
+        self.mgr.cache.stats = cache_stats
+
     def commit(self, conf_file: Optional[Path] = None):
         """
         Write current configuration out to the indicated file.

invokeai/app/services/models/image_record.py

@@ -39,6 +39,8 @@ class ImageRecord(BaseModelExcludeNull):
         description="The node ID that generated this image, if it is a generated image.",
     )
     """The node ID that generated this image, if it is a generated image."""
+    starred: bool = Field(description="Whether this image is starred.")
+    """Whether this image is starred."""
 
 
 class ImageRecordChanges(BaseModelExcludeNull, extra=Extra.forbid):
@@ -48,6 +50,7 @@ class ImageRecordChanges(BaseModelExcludeNull, extra=Extra.forbid):
       - `image_category`: change the category of an image
       - `session_id`: change the session associated with an image
       - `is_intermediate`: change the image's `is_intermediate` flag
+      - `starred`: change whether the image is starred
     """
 
     image_category: Optional[ImageCategory] = Field(description="The image's new category.")
@@ -59,6 +62,8 @@ class ImageRecordChanges(BaseModelExcludeNull, extra=Extra.forbid):
     """The image's new session ID."""
     is_intermediate: Optional[StrictBool] = Field(default=None, description="The image's new `is_intermediate` flag.")
     """The image's new `is_intermediate` flag."""
+    starred: Optional[StrictBool] = Field(default=None, description="The image's new `starred` state")
+    """The image's new `starred` state."""
 
 
 class ImageUrlsDTO(BaseModelExcludeNull):
@@ -113,6 +118,7 @@ def deserialize_image_record(image_dict: dict) -> ImageRecord:
     updated_at = image_dict.get("updated_at", get_iso_timestamp())
     deleted_at = image_dict.get("deleted_at", get_iso_timestamp())
     is_intermediate = image_dict.get("is_intermediate", False)
+    starred = image_dict.get("starred", False)
 
     return ImageRecord(
         image_name=image_name,
@@ -126,4 +132,5 @@ def deserialize_image_record(image_dict: dict) -> ImageRecord:
         updated_at=updated_at,
         deleted_at=deleted_at,
         is_intermediate=is_intermediate,
+        starred=starred,
     )

invokeai/app/services/processor.py

@@ -86,8 +86,13 @@ class DefaultInvocationProcessor(InvocationProcessorABC):
 
                 # Invoke
                 try:
-                    with statistics.collect_stats(invocation, graph_execution_state.id):
-                        outputs = invocation.invoke(
+                    graph_id = graph_execution_state.id
+                    model_manager = self.__invoker.services.model_manager
+                    with statistics.collect_stats(invocation, graph_id, model_manager):
+                        # use the internal invoke_internal(), which wraps the node's invoke() method in
+                        # this accomodates nodes which require a value, but get it only from a
+                        # connection
+                        outputs = invocation.invoke_internal(
                             InvocationContext(
                                 services=self.__invoker.services,
                                 graph_execution_state_id=graph_execution_state.id,

invokeai/app/services/sqlite.py

@@ -1,6 +1,7 @@
 import sqlite3
+import json
 from threading import Lock
-from typing import Generic, Optional, TypeVar, get_args
+from typing import Generic, Optional, TypeVar, Union, get_args
 
 from pydantic import BaseModel, parse_raw_as
 
@@ -49,7 +50,8 @@ class SqliteItemStorage(ItemStorageABC, Generic[T]):
 
     def _parse_item(self, item: str) -> T:
         item_type = get_args(self.__orig_class__)[0]
-        return parse_raw_as(item_type, item)
+        parsed = parse_raw_as(item_type, item)
+        return parsed
 
     def set(self, item: T):
         try:
@@ -98,7 +100,7 @@ class SqliteItemStorage(ItemStorageABC, Generic[T]):
             self._lock.release()
         self._on_deleted(id)
 
-    def list(self, page: int = 0, per_page: int = 10) -> PaginatedResults[T]:
+    def list(self, page: int = 0, per_page: int = 10) -> PaginatedResults[dict]:
         try:
             self._lock.acquire()
             self._cursor.execute(
@@ -107,7 +109,7 @@ class SqliteItemStorage(ItemStorageABC, Generic[T]):
             )
             result = self._cursor.fetchall()
 
-            items = list(map(lambda r: self._parse_item(r[0]), result))
+            items = [json.loads(r[0]) for r in result]
 
             self._cursor.execute(f"""SELECT count(*) FROM {self._table_name};""")
             count = self._cursor.fetchone()[0]
@@ -116,9 +118,9 @@ class SqliteItemStorage(ItemStorageABC, Generic[T]):
 
         pageCount = int(count / per_page) + 1
 
-        return PaginatedResults[T](items=items, page=page, pages=pageCount, per_page=per_page, total=count)
+        return PaginatedResults[dict](items=items, page=page, pages=pageCount, per_page=per_page, total=count)
 
-    def search(self, query: str, page: int = 0, per_page: int = 10) -> PaginatedResults[T]:
+    def search(self, query: str, page: int = 0, per_page: int = 10) -> PaginatedResults[dict]:
         try:
             self._lock.acquire()
             self._cursor.execute(
@@ -127,7 +129,7 @@ class SqliteItemStorage(ItemStorageABC, Generic[T]):
             )
             result = self._cursor.fetchall()
 
-            items = list(map(lambda r: self._parse_item(r[0]), result))
+            items = [json.loads(r[0]) for r in result]
 
             self._cursor.execute(
                 f"""SELECT count(*) FROM {self._table_name} WHERE item LIKE ?;""",
@@ -139,4 +141,4 @@ class SqliteItemStorage(ItemStorageABC, Generic[T]):
 
         pageCount = int(count / per_page) + 1
 
-        return PaginatedResults[T](items=items, page=page, pages=pageCount, per_page=per_page, total=count)
+        return PaginatedResults[dict](items=items, page=page, pages=pageCount, per_page=per_page, total=count)

invokeai/app/util/step_callback.py

@@ -4,9 +4,9 @@ from invokeai.app.models.exceptions import CanceledException
 from invokeai.app.models.image import ProgressImage
 from ..invocations.baseinvocation import InvocationContext
 from ...backend.util.util import image_to_dataURL
-from ...backend.generator.base import Generator
 from ...backend.stable_diffusion import PipelineIntermediateState
 from invokeai.app.services.config import InvokeAIAppConfig
+from ...backend.model_management.models import BaseModelType
 
 
 def sample_to_lowres_estimated_image(samples, latent_rgb_factors, smooth_matrix=None):
@@ -29,6 +29,7 @@ def stable_diffusion_step_callback(
     intermediate_state: PipelineIntermediateState,
     node: dict,
     source_node_id: str,
+    base_model: BaseModelType,
 ):
     if context.services.queue.is_canceled(context.graph_execution_state_id):
         raise CanceledException
@@ -56,23 +57,50 @@ def stable_diffusion_step_callback(
 
     # TODO: only output a preview image when requested
 
-    # origingally adapted from code by @erucipe and @keturn here:
-    # https://discuss.huggingface.co/t/decoding-latents-to-rgb-without-upscaling/23204/7
+    if base_model in [BaseModelType.StableDiffusionXL, BaseModelType.StableDiffusionXLRefiner]:
+        # fast latents preview matrix for sdxl
+        # generated by @StAlKeR7779
+        sdxl_latent_rgb_factors = torch.tensor(
+            [
+                #   R        G        B
+                [0.3816, 0.4930, 0.5320],
+                [-0.3753, 0.1631, 0.1739],
+                [0.1770, 0.3588, -0.2048],
+                [-0.4350, -0.2644, -0.4289],
+            ],
+            dtype=sample.dtype,
+            device=sample.device,
+        )
 
-    # these updated numbers for v1.5 are from @torridgristle
-    v1_5_latent_rgb_factors = torch.tensor(
-        [
-            #    R        G        B
-            [0.3444, 0.1385, 0.0670],  # L1
-            [0.1247, 0.4027, 0.1494],  # L2
-            [-0.3192, 0.2513, 0.2103],  # L3
-            [-0.1307, -0.1874, -0.7445],  # L4
-        ],
-        dtype=sample.dtype,
-        device=sample.device,
-    )
+        sdxl_smooth_matrix = torch.tensor(
+            [
+                [0.0358, 0.0964, 0.0358],
+                [0.0964, 0.4711, 0.0964],
+                [0.0358, 0.0964, 0.0358],
+            ],
+            dtype=sample.dtype,
+            device=sample.device,
+        )
 
-    image = sample_to_lowres_estimated_image(sample, v1_5_latent_rgb_factors)
+        image = sample_to_lowres_estimated_image(sample, sdxl_latent_rgb_factors, sdxl_smooth_matrix)
+    else:
+        # origingally adapted from code by @erucipe and @keturn here:
+        # https://discuss.huggingface.co/t/decoding-latents-to-rgb-without-upscaling/23204/7
+
+        # these updated numbers for v1.5 are from @torridgristle
+        v1_5_latent_rgb_factors = torch.tensor(
+            [
+                #    R        G        B
+                [0.3444, 0.1385, 0.0670],  # L1
+                [0.1247, 0.4027, 0.1494],  # L2
+                [-0.3192, 0.2513, 0.2103],  # L3
+                [-0.1307, -0.1874, -0.7445],  # L4
+            ],
+            dtype=sample.dtype,
+            device=sample.device,
+        )
+
+        image = sample_to_lowres_estimated_image(sample, v1_5_latent_rgb_factors)
 
     (width, height) = image.size
     width *= 8
@@ -86,59 +114,6 @@ def stable_diffusion_step_callback(
         source_node_id=source_node_id,
         progress_image=ProgressImage(width=width, height=height, dataURL=dataURL),
         step=intermediate_state.step,
-        total_steps=node["steps"],
-    )
-
-
-def stable_diffusion_xl_step_callback(
-    context: InvocationContext,
-    node: dict,
-    source_node_id: str,
-    sample,
-    step,
-    total_steps,
-):
-    if context.services.queue.is_canceled(context.graph_execution_state_id):
-        raise CanceledException
-
-    sdxl_latent_rgb_factors = torch.tensor(
-        [
-            #   R        G        B
-            [0.3816, 0.4930, 0.5320],
-            [-0.3753, 0.1631, 0.1739],
-            [0.1770, 0.3588, -0.2048],
-            [-0.4350, -0.2644, -0.4289],
-        ],
-        dtype=sample.dtype,
-        device=sample.device,
-    )
-
-    sdxl_smooth_matrix = torch.tensor(
-        [
-            # [ 0.0478,  0.1285,  0.0478],
-            # [ 0.1285,  0.2948,  0.1285],
-            # [ 0.0478,  0.1285,  0.0478],
-            [0.0358, 0.0964, 0.0358],
-            [0.0964, 0.4711, 0.0964],
-            [0.0358, 0.0964, 0.0358],
-        ],
-        dtype=sample.dtype,
-        device=sample.device,
-    )
-
-    image = sample_to_lowres_estimated_image(sample, sdxl_latent_rgb_factors, sdxl_smooth_matrix)
-
-    (width, height) = image.size
-    width *= 8
-    height *= 8
-
-    dataURL = image_to_dataURL(image, image_format="JPEG")
-
-    context.services.events.emit_generator_progress(
-        graph_execution_state_id=context.graph_execution_state_id,
-        node=node,
-        source_node_id=source_node_id,
-        progress_image=ProgressImage(width=width, height=height, dataURL=dataURL),
-        step=step,
-        total_steps=total_steps,
+        order=intermediate_state.order,
+        total_steps=intermediate_state.total_steps,
     )

invokeai/backend/__init__.py

@@ -1,6 +1,5 @@
 """
 Initialization file for invokeai.backend
 """
-from .generator import InvokeAIGeneratorBasicParams, InvokeAIGenerator, InvokeAIGeneratorOutput, Img2Img, Inpaint
 from .model_management import ModelManager, ModelCache, BaseModelType, ModelType, SubModelType, ModelInfo
 from .model_management.models import SilenceWarnings

invokeai/backend/generator/__init__.py

@@ -1,12 +0,0 @@
-"""
-Initialization file for the invokeai.generator package
-"""
-from .base import (
-    InvokeAIGenerator,
-    InvokeAIGeneratorBasicParams,
-    InvokeAIGeneratorOutput,
-    Img2Img,
-    Inpaint,
-    Generator,
-)
-from .inpaint import infill_methods

invokeai/backend/generator/base.py

@@ -1,559 +0,0 @@
-"""
-Base class for invokeai.backend.generator.*
-including img2img, txt2img, and inpaint
-"""
-from __future__ import annotations
-
-import itertools
-import dataclasses
-import diffusers
-import os
-import random
-import traceback
-from abc import ABCMeta
-from argparse import Namespace
-from contextlib import nullcontext
-
-import cv2
-import numpy as np
-import torch
-from PIL import Image, ImageChops, ImageFilter
-from accelerate.utils import set_seed
-from diffusers import DiffusionPipeline
-from tqdm import trange
-from typing import Callable, List, Iterator, Optional, Type, Union
-from dataclasses import dataclass, field
-from diffusers.schedulers import SchedulerMixin as Scheduler
-
-import invokeai.backend.util.logging as logger
-from ..image_util import configure_model_padding
-from ..util.util import rand_perlin_2d
-from ..stable_diffusion.diffusers_pipeline import StableDiffusionGeneratorPipeline
-from ..stable_diffusion.schedulers import SCHEDULER_MAP
-
-downsampling = 8
-
-
-@dataclass
-class InvokeAIGeneratorBasicParams:
-    seed: Optional[int] = None
-    width: int = 512
-    height: int = 512
-    cfg_scale: float = 7.5
-    steps: int = 20
-    ddim_eta: float = 0.0
-    scheduler: str = "ddim"
-    precision: str = "float16"
-    perlin: float = 0.0
-    threshold: float = 0.0
-    seamless: bool = False
-    seamless_axes: List[str] = field(default_factory=lambda: ["x", "y"])
-    h_symmetry_time_pct: Optional[float] = None
-    v_symmetry_time_pct: Optional[float] = None
-    variation_amount: float = 0.0
-    with_variations: list = field(default_factory=list)
-
-
-@dataclass
-class InvokeAIGeneratorOutput:
-    """
-    InvokeAIGeneratorOutput is a dataclass that contains the outputs of a generation
-    operation, including the image, its seed, the model name used to generate the image
-    and the model hash, as well as all the generate() parameters that went into
-    generating the image (in .params, also available as attributes)
-    """
-
-    image: Image.Image
-    seed: int
-    model_hash: str
-    attention_maps_images: List[Image.Image]
-    params: Namespace
-
-
-# we are interposing a wrapper around the original Generator classes so that
-# old code that calls Generate will continue to work.
-class InvokeAIGenerator(metaclass=ABCMeta):
-    def __init__(
-        self,
-        model_info: dict,
-        params: InvokeAIGeneratorBasicParams = InvokeAIGeneratorBasicParams(),
-        **kwargs,
-    ):
-        self.model_info = model_info
-        self.params = params
-        self.kwargs = kwargs
-
-    def generate(
-        self,
-        conditioning: tuple,
-        scheduler,
-        callback: Optional[Callable] = None,
-        step_callback: Optional[Callable] = None,
-        iterations: int = 1,
-        **keyword_args,
-    ) -> Iterator[InvokeAIGeneratorOutput]:
-        """
-        Return an iterator across the indicated number of generations.
-        Each time the iterator is called it will return an InvokeAIGeneratorOutput
-        object. Use like this:
-
-           outputs = txt2img.generate(prompt='banana sushi', iterations=5)
-           for result in outputs:
-               print(result.image, result.seed)
-
-        In the typical case of wanting to get just a single image, iterations
-        defaults to 1 and do:
-
-           output = next(txt2img.generate(prompt='banana sushi')
-
-        Pass None to get an infinite iterator.
-
-           outputs = txt2img.generate(prompt='banana sushi', iterations=None)
-           for o in outputs:
-               print(o.image, o.seed)
-
-        """
-        generator_args = dataclasses.asdict(self.params)
-        generator_args.update(keyword_args)
-
-        model_info = self.model_info
-        model_name = model_info.name
-        model_hash = model_info.hash
-        with model_info.context as model:
-            gen_class = self._generator_class()
-            generator = gen_class(model, self.params.precision, **self.kwargs)
-            if self.params.variation_amount > 0:
-                generator.set_variation(
-                    generator_args.get("seed"),
-                    generator_args.get("variation_amount"),
-                    generator_args.get("with_variations"),
-                )
-
-            if isinstance(model, DiffusionPipeline):
-                for component in [model.unet, model.vae]:
-                    configure_model_padding(
-                        component, generator_args.get("seamless", False), generator_args.get("seamless_axes")
-                    )
-            else:
-                configure_model_padding(
-                    model, generator_args.get("seamless", False), generator_args.get("seamless_axes")
-                )
-
-            iteration_count = range(iterations) if iterations else itertools.count(start=0, step=1)
-            for i in iteration_count:
-                results = generator.generate(
-                    conditioning=conditioning,
-                    step_callback=step_callback,
-                    sampler=scheduler,
-                    **generator_args,
-                )
-                output = InvokeAIGeneratorOutput(
-                    image=results[0][0],
-                    seed=results[0][1],
-                    attention_maps_images=results[0][2],
-                    model_hash=model_hash,
-                    params=Namespace(model_name=model_name, **generator_args),
-                )
-                if callback:
-                    callback(output)
-            yield output
-
-    @classmethod
-    def schedulers(self) -> List[str]:
-        """
-        Return list of all the schedulers that we currently handle.
-        """
-        return list(SCHEDULER_MAP.keys())
-
-    def load_generator(self, model: StableDiffusionGeneratorPipeline, generator_class: Type[Generator]):
-        return generator_class(model, self.params.precision)
-
-    @classmethod
-    def _generator_class(cls) -> Type[Generator]:
-        """
-        In derived classes return the name of the generator to apply.
-        If you don't override will return the name of the derived
-        class, which nicely parallels the generator class names.
-        """
-        return Generator
-
-
-# ------------------------------------
-class Img2Img(InvokeAIGenerator):
-    def generate(
-        self, init_image: Union[Image.Image, torch.FloatTensor], strength: float = 0.75, **keyword_args
-    ) -> Iterator[InvokeAIGeneratorOutput]:
-        return super().generate(init_image=init_image, strength=strength, **keyword_args)
-
-    @classmethod
-    def _generator_class(cls):
-        from .img2img import Img2Img
-
-        return Img2Img
-
-
-# ------------------------------------
-# Takes all the arguments of Img2Img and adds the mask image and the seam/infill stuff
-class Inpaint(Img2Img):
-    def generate(
-        self,
-        mask_image: Union[Image.Image, torch.FloatTensor],
-        # Seam settings - when 0, doesn't fill seam
-        seam_size: int = 96,
-        seam_blur: int = 16,
-        seam_strength: float = 0.7,
-        seam_steps: int = 30,
-        tile_size: int = 32,
-        inpaint_replace=False,
-        infill_method=None,
-        inpaint_width=None,
-        inpaint_height=None,
-        inpaint_fill: tuple(int) = (0x7F, 0x7F, 0x7F, 0xFF),
-        **keyword_args,
-    ) -> Iterator[InvokeAIGeneratorOutput]:
-        return super().generate(
-            mask_image=mask_image,
-            seam_size=seam_size,
-            seam_blur=seam_blur,
-            seam_strength=seam_strength,
-            seam_steps=seam_steps,
-            tile_size=tile_size,
-            inpaint_replace=inpaint_replace,
-            infill_method=infill_method,
-            inpaint_width=inpaint_width,
-            inpaint_height=inpaint_height,
-            inpaint_fill=inpaint_fill,
-            **keyword_args,
-        )
-
-    @classmethod
-    def _generator_class(cls):
-        from .inpaint import Inpaint
-
-        return Inpaint
-
-
-class Generator:
-    downsampling_factor: int
-    latent_channels: int
-    precision: str
-    model: DiffusionPipeline
-
-    def __init__(self, model: DiffusionPipeline, precision: str, **kwargs):
-        self.model = model
-        self.precision = precision
-        self.seed = None
-        self.latent_channels = model.unet.config.in_channels
-        self.downsampling_factor = downsampling  # BUG: should come from model or config
-        self.perlin = 0.0
-        self.threshold = 0
-        self.variation_amount = 0
-        self.with_variations = []
-        self.use_mps_noise = False
-        self.free_gpu_mem = None
-
-    # this is going to be overridden in img2img.py, txt2img.py and inpaint.py
-    def get_make_image(self, **kwargs):
-        """
-        Returns a function returning an image derived from the prompt and the initial image
-        Return value depends on the seed at the time you call it
-        """
-        raise NotImplementedError("image_iterator() must be implemented in a descendent class")
-
-    def set_variation(self, seed, variation_amount, with_variations):
-        self.seed = seed
-        self.variation_amount = variation_amount
-        self.with_variations = with_variations
-
-    def generate(
-        self,
-        width,
-        height,
-        sampler,
-        init_image=None,
-        iterations=1,
-        seed=None,
-        image_callback=None,
-        step_callback=None,
-        threshold=0.0,
-        perlin=0.0,
-        h_symmetry_time_pct=None,
-        v_symmetry_time_pct=None,
-        free_gpu_mem: bool = False,
-        **kwargs,
-    ):
-        scope = nullcontext
-        self.free_gpu_mem = free_gpu_mem
-        attention_maps_images = []
-        attention_maps_callback = lambda saver: attention_maps_images.append(saver.get_stacked_maps_image())
-        make_image = self.get_make_image(
-            sampler=sampler,
-            init_image=init_image,
-            width=width,
-            height=height,
-            step_callback=step_callback,
-            threshold=threshold,
-            perlin=perlin,
-            h_symmetry_time_pct=h_symmetry_time_pct,
-            v_symmetry_time_pct=v_symmetry_time_pct,
-            attention_maps_callback=attention_maps_callback,
-            **kwargs,
-        )
-        results = []
-        seed = seed if seed is not None and seed >= 0 else self.new_seed()
-        first_seed = seed
-        seed, initial_noise = self.generate_initial_noise(seed, width, height)
-
-        # There used to be an additional self.model.ema_scope() here, but it breaks
-        # the inpaint-1.5 model. Not sure what it did.... ?
-        with scope(self.model.device.type):
-            for n in trange(iterations, desc="Generating"):
-                x_T = None
-                if self.variation_amount > 0:
-                    set_seed(seed)
-                    target_noise = self.get_noise(width, height)
-                    x_T = self.slerp(self.variation_amount, initial_noise, target_noise)
-                elif initial_noise is not None:
-                    # i.e. we specified particular variations
-                    x_T = initial_noise
-                else:
-                    set_seed(seed)
-                    try:
-                        x_T = self.get_noise(width, height)
-                    except:
-                        logger.error("An error occurred while getting initial noise")
-                        print(traceback.format_exc())
-
-                # Pass on the seed in case a layer beneath us needs to generate noise on its own.
-                image = make_image(x_T, seed)
-
-                results.append([image, seed, attention_maps_images])
-
-                if image_callback is not None:
-                    attention_maps_image = None if len(attention_maps_images) == 0 else attention_maps_images[-1]
-                    image_callback(
-                        image,
-                        seed,
-                        first_seed=first_seed,
-                        attention_maps_image=attention_maps_image,
-                    )
-
-                seed = self.new_seed()
-
-                # Free up memory from the last generation.
-                clear_cuda_cache = kwargs["clear_cuda_cache"] if "clear_cuda_cache" in kwargs else None
-                if clear_cuda_cache is not None:
-                    clear_cuda_cache()
-
-        return results
-
-    def sample_to_image(self, samples) -> Image.Image:
-        """
-        Given samples returned from a sampler, converts
-        it into a PIL Image
-        """
-        with torch.inference_mode():
-            image = self.model.decode_latents(samples)
-        return self.model.numpy_to_pil(image)[0]
-
-    def repaste_and_color_correct(
-        self,
-        result: Image.Image,
-        init_image: Image.Image,
-        init_mask: Image.Image,
-        mask_blur_radius: int = 8,
-    ) -> Image.Image:
-        if init_image is None or init_mask is None:
-            return result
-
-        # Get the original alpha channel of the mask if there is one.
-        # Otherwise it is some other black/white image format ('1', 'L' or 'RGB')
-        pil_init_mask = init_mask.getchannel("A") if init_mask.mode == "RGBA" else init_mask.convert("L")
-        pil_init_image = init_image.convert("RGBA")  # Add an alpha channel if one doesn't exist
-
-        # Build an image with only visible pixels from source to use as reference for color-matching.
-        init_rgb_pixels = np.asarray(init_image.convert("RGB"), dtype=np.uint8)
-        init_a_pixels = np.asarray(pil_init_image.getchannel("A"), dtype=np.uint8)
-        init_mask_pixels = np.asarray(pil_init_mask, dtype=np.uint8)
-
-        # Get numpy version of result
-        np_image = np.asarray(result, dtype=np.uint8)
-
-        # Mask and calculate mean and standard deviation
-        mask_pixels = init_a_pixels * init_mask_pixels > 0
-        np_init_rgb_pixels_masked = init_rgb_pixels[mask_pixels, :]
-        np_image_masked = np_image[mask_pixels, :]
-
-        if np_init_rgb_pixels_masked.size > 0:
-            init_means = np_init_rgb_pixels_masked.mean(axis=0)
-            init_std = np_init_rgb_pixels_masked.std(axis=0)
-            gen_means = np_image_masked.mean(axis=0)
-            gen_std = np_image_masked.std(axis=0)
-
-            # Color correct
-            np_matched_result = np_image.copy()
-            np_matched_result[:, :, :] = (
-                (
-                    (
-                        (np_matched_result[:, :, :].astype(np.float32) - gen_means[None, None, :])
-                        / gen_std[None, None, :]
-                    )
-                    * init_std[None, None, :]
-                    + init_means[None, None, :]
-                )
-                .clip(0, 255)
-                .astype(np.uint8)
-            )
-            matched_result = Image.fromarray(np_matched_result, mode="RGB")
-        else:
-            matched_result = Image.fromarray(np_image, mode="RGB")
-
-        # Blur the mask out (into init image) by specified amount
-        if mask_blur_radius > 0:
-            nm = np.asarray(pil_init_mask, dtype=np.uint8)
-            nmd = cv2.erode(
-                nm,
-                kernel=np.ones((3, 3), dtype=np.uint8),
-                iterations=int(mask_blur_radius / 2),
-            )
-            pmd = Image.fromarray(nmd, mode="L")
-            blurred_init_mask = pmd.filter(ImageFilter.BoxBlur(mask_blur_radius))
-        else:
-            blurred_init_mask = pil_init_mask
-
-        multiplied_blurred_init_mask = ImageChops.multiply(blurred_init_mask, self.pil_image.split()[-1])
-
-        # Paste original on color-corrected generation (using blurred mask)
-        matched_result.paste(init_image, (0, 0), mask=multiplied_blurred_init_mask)
-        return matched_result
-
-    @staticmethod
-    def sample_to_lowres_estimated_image(samples):
-        # origingally adapted from code by @erucipe and @keturn here:
-        # https://discuss.huggingface.co/t/decoding-latents-to-rgb-without-upscaling/23204/7
-
-        # these updated numbers for v1.5 are from @torridgristle
-        v1_5_latent_rgb_factors = torch.tensor(
-            [
-                #    R        G        B
-                [0.3444, 0.1385, 0.0670],  # L1
-                [0.1247, 0.4027, 0.1494],  # L2
-                [-0.3192, 0.2513, 0.2103],  # L3
-                [-0.1307, -0.1874, -0.7445],  # L4
-            ],
-            dtype=samples.dtype,
-            device=samples.device,
-        )
-
-        latent_image = samples[0].permute(1, 2, 0) @ v1_5_latent_rgb_factors
-        latents_ubyte = (
-            ((latent_image + 1) / 2).clamp(0, 1).mul(0xFF).byte()  # change scale from -1..1 to 0..1  # to 0..255
-        ).cpu()
-
-        return Image.fromarray(latents_ubyte.numpy())
-
-    def generate_initial_noise(self, seed, width, height):
-        initial_noise = None
-        if self.variation_amount > 0 or len(self.with_variations) > 0:
-            # use fixed initial noise plus random noise per iteration
-            set_seed(seed)
-            initial_noise = self.get_noise(width, height)
-            for v_seed, v_weight in self.with_variations:
-                seed = v_seed
-                set_seed(seed)
-                next_noise = self.get_noise(width, height)
-                initial_noise = self.slerp(v_weight, initial_noise, next_noise)
-            if self.variation_amount > 0:
-                random.seed()  # reset RNG to an actually random state, so we can get a random seed for variations
-                seed = random.randrange(0, np.iinfo(np.uint32).max)
-        return (seed, initial_noise)
-
-    def get_perlin_noise(self, width, height):
-        fixdevice = "cpu" if (self.model.device.type == "mps") else self.model.device
-        # limit noise to only the diffusion image channels, not the mask channels
-        input_channels = min(self.latent_channels, 4)
-        # round up to the nearest block of 8
-        temp_width = int((width + 7) / 8) * 8
-        temp_height = int((height + 7) / 8) * 8
-        noise = torch.stack(
-            [
-                rand_perlin_2d((temp_height, temp_width), (8, 8), device=self.model.device).to(fixdevice)
-                for _ in range(input_channels)
-            ],
-            dim=0,
-        ).to(self.model.device)
-        return noise[0:4, 0:height, 0:width]
-
-    def new_seed(self):
-        self.seed = random.randrange(0, np.iinfo(np.uint32).max)
-        return self.seed
-
-    def slerp(self, t, v0, v1, DOT_THRESHOLD=0.9995):
-        """
-        Spherical linear interpolation
-        Args:
-            t (float/np.ndarray): Float value between 0.0 and 1.0
-            v0 (np.ndarray): Starting vector
-            v1 (np.ndarray): Final vector
-            DOT_THRESHOLD (float): Threshold for considering the two vectors as
-                                colineal. Not recommended to alter this.
-        Returns:
-            v2 (np.ndarray): Interpolation vector between v0 and v1
-        """
-        inputs_are_torch = False
-        if not isinstance(v0, np.ndarray):
-            inputs_are_torch = True
-            v0 = v0.detach().cpu().numpy()
-        if not isinstance(v1, np.ndarray):
-            inputs_are_torch = True
-            v1 = v1.detach().cpu().numpy()
-
-        dot = np.sum(v0 * v1 / (np.linalg.norm(v0) * np.linalg.norm(v1)))
-        if np.abs(dot) > DOT_THRESHOLD:
-            v2 = (1 - t) * v0 + t * v1
-        else:
-            theta_0 = np.arccos(dot)
-            sin_theta_0 = np.sin(theta_0)
-            theta_t = theta_0 * t
-            sin_theta_t = np.sin(theta_t)
-            s0 = np.sin(theta_0 - theta_t) / sin_theta_0
-            s1 = sin_theta_t / sin_theta_0
-            v2 = s0 * v0 + s1 * v1
-
-        if inputs_are_torch:
-            v2 = torch.from_numpy(v2).to(self.model.device)
-
-        return v2
-
-    # this is a handy routine for debugging use. Given a generated sample,
-    # convert it into a PNG image and store it at the indicated path
-    def save_sample(self, sample, filepath):
-        image = self.sample_to_image(sample)
-        dirname = os.path.dirname(filepath) or "."
-        if not os.path.exists(dirname):
-            logger.info(f"creating directory {dirname}")
-            os.makedirs(dirname, exist_ok=True)
-        image.save(filepath, "PNG")
-
-    def torch_dtype(self) -> torch.dtype:
-        return torch.float16 if self.precision == "float16" else torch.float32
-
-    # returns a tensor filled with random numbers from a normal distribution
-    def get_noise(self, width, height):
-        device = self.model.device
-        # limit noise to only the diffusion image channels, not the mask channels
-        input_channels = min(self.latent_channels, 4)
-        x = torch.randn(
-            [
-                1,
-                input_channels,
-                height // self.downsampling_factor,
-                width // self.downsampling_factor,
-            ],
-            dtype=self.torch_dtype(),
-            device=device,
-        )
-        if self.perlin > 0.0:
-            perlin_noise = self.get_perlin_noise(width // self.downsampling_factor, height // self.downsampling_factor)
-            x = (1 - self.perlin) * x + self.perlin * perlin_noise
-        return x

invokeai/backend/generator/img2img.py

@@ -1,31 +0,0 @@
-"""
-invokeai.backend.generator.img2img descends from .generator
-"""
-
-from .base import Generator
-
-
-class Img2Img(Generator):
-    def get_make_image(
-        self,
-        sampler,
-        steps,
-        cfg_scale,
-        ddim_eta,
-        conditioning,
-        init_image,
-        strength,
-        step_callback=None,
-        threshold=0.0,
-        warmup=0.2,
-        perlin=0.0,
-        h_symmetry_time_pct=None,
-        v_symmetry_time_pct=None,
-        attention_maps_callback=None,
-        **kwargs,
-    ):
-        """
-        Returns a function returning an image derived from the prompt and the initial image
-        Return value depends on the seed at the time you call it.
-        """
-        raise NotImplementedError("replaced by invokeai.app.invocations.latent.LatentsToLatentsInvocation")

invokeai/backend/generator/inpaint.py

@@ -1,387 +0,0 @@
-"""
-invokeai.backend.generator.inpaint descends from .generator
-"""
-from __future__ import annotations
-
-import math
-from typing import Tuple, Union, Optional
-
-import cv2
-import numpy as np
-import torch
-from PIL import Image, ImageChops, ImageFilter, ImageOps
-
-from ..image_util import PatchMatch, debug_image
-from ..stable_diffusion.diffusers_pipeline import (
-    ConditioningData,
-    StableDiffusionGeneratorPipeline,
-    image_resized_to_grid_as_tensor,
-)
-from .img2img import Img2Img
-
-
-def infill_methods() -> list[str]:
-    methods = [
-        "tile",
-        "solid",
-    ]
-    if PatchMatch.patchmatch_available():
-        methods.insert(0, "patchmatch")
-    return methods
-
-
-class Inpaint(Img2Img):
-    def __init__(self, model, precision):
-        self.inpaint_height = 0
-        self.inpaint_width = 0
-        self.enable_image_debugging = False
-        self.init_latent = None
-        self.pil_image = None
-        self.pil_mask = None
-        self.mask_blur_radius = 0
-        self.infill_method = None
-        super().__init__(model, precision)
-
-    # Outpaint support code
-    def get_tile_images(self, image: np.ndarray, width=8, height=8):
-        _nrows, _ncols, depth = image.shape
-        _strides = image.strides
-
-        nrows, _m = divmod(_nrows, height)
-        ncols, _n = divmod(_ncols, width)
-        if _m != 0 or _n != 0:
-            return None
-
-        return np.lib.stride_tricks.as_strided(
-            np.ravel(image),
-            shape=(nrows, ncols, height, width, depth),
-            strides=(height * _strides[0], width * _strides[1], *_strides),
-            writeable=False,
-        )
-
-    def infill_patchmatch(self, im: Image.Image) -> Image.Image:
-        if im.mode != "RGBA":
-            return im
-
-        # Skip patchmatch if patchmatch isn't available
-        if not PatchMatch.patchmatch_available():
-            return im
-
-        # Patchmatch (note, we may want to expose patch_size? Increasing it significantly impacts performance though)
-        im_patched_np = PatchMatch.inpaint(im.convert("RGB"), ImageOps.invert(im.split()[-1]), patch_size=3)
-        im_patched = Image.fromarray(im_patched_np, mode="RGB")
-        return im_patched
-
-    def tile_fill_missing(self, im: Image.Image, tile_size: int = 16, seed: Optional[int] = None) -> Image.Image:
-        # Only fill if there's an alpha layer
-        if im.mode != "RGBA":
-            return im
-
-        a = np.asarray(im, dtype=np.uint8)
-
-        tile_size_tuple = (tile_size, tile_size)
-
-        # Get the image as tiles of a specified size
-        tiles = self.get_tile_images(a, *tile_size_tuple).copy()
-
-        # Get the mask as tiles
-        tiles_mask = tiles[:, :, :, :, 3]
-
-        # Find any mask tiles with any fully transparent pixels (we will be replacing these later)
-        tmask_shape = tiles_mask.shape
-        tiles_mask = tiles_mask.reshape(math.prod(tiles_mask.shape))
-        n, ny = (math.prod(tmask_shape[0:2])), math.prod(tmask_shape[2:])
-        tiles_mask = tiles_mask > 0
-        tiles_mask = tiles_mask.reshape((n, ny)).all(axis=1)
-
-        # Get RGB tiles in single array and filter by the mask
-        tshape = tiles.shape
-        tiles_all = tiles.reshape((math.prod(tiles.shape[0:2]), *tiles.shape[2:]))
-        filtered_tiles = tiles_all[tiles_mask]
-
-        if len(filtered_tiles) == 0:
-            return im
-
-        # Find all invalid tiles and replace with a random valid tile
-        replace_count = (tiles_mask == False).sum()
-        rng = np.random.default_rng(seed=seed)
-        tiles_all[np.logical_not(tiles_mask)] = filtered_tiles[
-            rng.choice(filtered_tiles.shape[0], replace_count), :, :, :
-        ]
-
-        # Convert back to an image
-        tiles_all = tiles_all.reshape(tshape)
-        tiles_all = tiles_all.swapaxes(1, 2)
-        st = tiles_all.reshape(
-            (
-                math.prod(tiles_all.shape[0:2]),
-                math.prod(tiles_all.shape[2:4]),
-                tiles_all.shape[4],
-            )
-        )
-        si = Image.fromarray(st, mode="RGBA")
-
-        return si
-
-    def mask_edge(self, mask: Image.Image, edge_size: int, edge_blur: int) -> Image.Image:
-        npimg = np.asarray(mask, dtype=np.uint8)
-
-        # Detect any partially transparent regions
-        npgradient = np.uint8(255 * (1.0 - np.floor(np.abs(0.5 - np.float32(npimg) / 255.0) * 2.0)))
-
-        # Detect hard edges
-        npedge = cv2.Canny(npimg, threshold1=100, threshold2=200)
-
-        # Combine
-        npmask = npgradient + npedge
-
-        # Expand
-        npmask = cv2.dilate(npmask, np.ones((3, 3), np.uint8), iterations=int(edge_size / 2))
-
-        new_mask = Image.fromarray(npmask)
-
-        if edge_blur > 0:
-            new_mask = new_mask.filter(ImageFilter.BoxBlur(edge_blur))
-
-        return ImageOps.invert(new_mask)
-
-    def seam_paint(
-        self,
-        im: Image.Image,
-        seam_size: int,
-        seam_blur: int,
-        seed,
-        steps,
-        cfg_scale,
-        ddim_eta,
-        conditioning,
-        strength,
-        noise,
-        infill_method,
-        step_callback,
-    ) -> Image.Image:
-        hard_mask = self.pil_image.split()[-1].copy()
-        mask = self.mask_edge(hard_mask, seam_size, seam_blur)
-
-        make_image = self.get_make_image(
-            steps,
-            cfg_scale,
-            ddim_eta,
-            conditioning,
-            init_image=im.copy().convert("RGBA"),
-            mask_image=mask,
-            strength=strength,
-            mask_blur_radius=0,
-            seam_size=0,
-            step_callback=step_callback,
-            inpaint_width=im.width,
-            inpaint_height=im.height,
-            infill_method=infill_method,
-        )
-
-        seam_noise = self.get_noise(im.width, im.height)
-
-        result = make_image(seam_noise, seed=None)
-
-        return result
-
-    @torch.no_grad()
-    def get_make_image(
-        self,
-        steps,
-        cfg_scale,
-        ddim_eta,
-        conditioning,
-        init_image: Union[Image.Image, torch.FloatTensor],
-        mask_image: Union[Image.Image, torch.FloatTensor],
-        strength: float,
-        mask_blur_radius: int = 8,
-        # Seam settings - when 0, doesn't fill seam
-        seam_size: int = 96,
-        seam_blur: int = 16,
-        seam_strength: float = 0.7,
-        seam_steps: int = 30,
-        tile_size: int = 32,
-        step_callback=None,
-        inpaint_replace=False,
-        enable_image_debugging=False,
-        infill_method=None,
-        inpaint_width=None,
-        inpaint_height=None,
-        inpaint_fill: Tuple[int, int, int, int] = (0x7F, 0x7F, 0x7F, 0xFF),
-        attention_maps_callback=None,
-        **kwargs,
-    ):
-        """
-        Returns a function returning an image derived from the prompt and
-        the initial image + mask.  Return value depends on the seed at
-        the time you call it.  kwargs are 'init_latent' and 'strength'
-        """
-
-        self.enable_image_debugging = enable_image_debugging
-        infill_method = infill_method or infill_methods()[0]
-        self.infill_method = infill_method
-
-        self.inpaint_width = inpaint_width
-        self.inpaint_height = inpaint_height
-
-        if isinstance(init_image, Image.Image):
-            self.pil_image = init_image.copy()
-
-            # Do infill
-            if infill_method == "patchmatch" and PatchMatch.patchmatch_available():
-                init_filled = self.infill_patchmatch(self.pil_image.copy())
-            elif infill_method == "tile":
-                init_filled = self.tile_fill_missing(self.pil_image.copy(), seed=self.seed, tile_size=tile_size)
-            elif infill_method == "solid":
-                solid_bg = Image.new("RGBA", init_image.size, inpaint_fill)
-                init_filled = Image.alpha_composite(solid_bg, init_image)
-            else:
-                raise ValueError(f"Non-supported infill type {infill_method}", infill_method)
-            init_filled.paste(init_image, (0, 0), init_image.split()[-1])
-
-            # Resize if requested for inpainting
-            if inpaint_width and inpaint_height:
-                init_filled = init_filled.resize((inpaint_width, inpaint_height))
-
-            debug_image(init_filled, "init_filled", debug_status=self.enable_image_debugging)
-
-            # Create init tensor
-            init_image = image_resized_to_grid_as_tensor(init_filled.convert("RGB"))
-
-        if isinstance(mask_image, Image.Image):
-            self.pil_mask = mask_image.copy()
-            debug_image(
-                mask_image,
-                "mask_image BEFORE multiply with pil_image",
-                debug_status=self.enable_image_debugging,
-            )
-
-            init_alpha = self.pil_image.getchannel("A")
-            if mask_image.mode != "L":
-                # FIXME: why do we get passed an RGB image here? We can only use single-channel.
-                mask_image = mask_image.convert("L")
-            mask_image = ImageChops.multiply(mask_image, init_alpha)
-            self.pil_mask = mask_image
-
-            # Resize if requested for inpainting
-            if inpaint_width and inpaint_height:
-                mask_image = mask_image.resize((inpaint_width, inpaint_height))
-
-            debug_image(
-                mask_image,
-                "mask_image AFTER multiply with pil_image",
-                debug_status=self.enable_image_debugging,
-            )
-            mask: torch.FloatTensor = image_resized_to_grid_as_tensor(mask_image, normalize=False)
-        else:
-            mask: torch.FloatTensor = mask_image
-
-        self.mask_blur_radius = mask_blur_radius
-
-        # noinspection PyTypeChecker
-        pipeline: StableDiffusionGeneratorPipeline = self.model
-
-        # todo: support cross-attention control
-        uc, c, _ = conditioning
-        conditioning_data = ConditioningData(uc, c, cfg_scale).add_scheduler_args_if_applicable(
-            pipeline.scheduler, eta=ddim_eta
-        )
-
-        def make_image(x_T: torch.Tensor, seed: int):
-            pipeline_output = pipeline.inpaint_from_embeddings(
-                init_image=init_image,
-                mask=1 - mask,  # expects white means "paint here."
-                strength=strength,
-                num_inference_steps=steps,
-                conditioning_data=conditioning_data,
-                noise_func=self.get_noise_like,
-                callback=step_callback,
-                seed=seed,
-            )
-
-            if pipeline_output.attention_map_saver is not None and attention_maps_callback is not None:
-                attention_maps_callback(pipeline_output.attention_map_saver)
-
-            result = self.postprocess_size_and_mask(pipeline.numpy_to_pil(pipeline_output.images)[0])
-
-            # Seam paint if this is our first pass (seam_size set to 0 during seam painting)
-            if seam_size > 0:
-                old_image = self.pil_image or init_image
-                old_mask = self.pil_mask or mask_image
-
-                result = self.seam_paint(
-                    result,
-                    seam_size,
-                    seam_blur,
-                    seed,
-                    seam_steps,
-                    cfg_scale,
-                    ddim_eta,
-                    conditioning,
-                    seam_strength,
-                    x_T,
-                    infill_method,
-                    step_callback,
-                )
-
-                # Restore original settings
-                self.get_make_image(
-                    steps,
-                    cfg_scale,
-                    ddim_eta,
-                    conditioning,
-                    old_image,
-                    old_mask,
-                    strength,
-                    mask_blur_radius,
-                    seam_size,
-                    seam_blur,
-                    seam_strength,
-                    seam_steps,
-                    tile_size,
-                    step_callback,
-                    inpaint_replace,
-                    enable_image_debugging,
-                    inpaint_width=inpaint_width,
-                    inpaint_height=inpaint_height,
-                    infill_method=infill_method,
-                    **kwargs,
-                )
-
-            return result
-
-        return make_image
-
-    def sample_to_image(self, samples) -> Image.Image:
-        gen_result = super().sample_to_image(samples).convert("RGB")
-        return self.postprocess_size_and_mask(gen_result)
-
-    def postprocess_size_and_mask(self, gen_result: Image.Image) -> Image.Image:
-        debug_image(gen_result, "gen_result", debug_status=self.enable_image_debugging)
-
-        # Resize if necessary
-        if self.inpaint_width and self.inpaint_height:
-            gen_result = gen_result.resize(self.pil_image.size)
-
-        if self.pil_image is None or self.pil_mask is None:
-            return gen_result
-
-        corrected_result = self.repaste_and_color_correct(
-            gen_result, self.pil_image, self.pil_mask, self.mask_blur_radius
-        )
-        debug_image(
-            corrected_result,
-            "corrected_result",
-            debug_status=self.enable_image_debugging,
-        )
-
-        return corrected_result
-
-    def get_noise_like(self, like: torch.Tensor):
-        device = like.device
-        x = torch.randn_like(like, device=device)
-        if self.perlin > 0.0:
-            shape = like.shape
-            x = (1 - self.perlin) * x + self.perlin * self.get_perlin_noise(shape[3], shape[2])
-        return x

invokeai/backend/model_management/model_cache.py

@@ -21,12 +21,12 @@ import os
 import sys
 import hashlib
 from contextlib import suppress
+from dataclasses import dataclass, field
 from pathlib import Path
 from typing import Dict, Union, types, Optional, Type, Any
 
 import torch
 
-import logging
 import invokeai.backend.util.logging as logger
 from .models import BaseModelType, ModelType, SubModelType, ModelBase
 
@@ -41,6 +41,18 @@ DEFAULT_MAX_VRAM_CACHE_SIZE = 2.75
 GIG = 1073741824
 
 
+@dataclass
+class CacheStats(object):
+    hits: int = 0  # cache hits
+    misses: int = 0  # cache misses
+    high_watermark: int = 0  # amount of cache used
+    in_cache: int = 0  # number of models in cache
+    cleared: int = 0  # number of models cleared to make space
+    cache_size: int = 0  # total size of cache
+    # {submodel_key => size}
+    loaded_model_sizes: Dict[str, int] = field(default_factory=dict)
+
+
 class ModelLocker(object):
     "Forward declaration"
     pass
@@ -115,6 +127,9 @@ class ModelCache(object):
         self.sha_chunksize = sha_chunksize
         self.logger = logger
 
+        # used for stats collection
+        self.stats = None
+
         self._cached_models = dict()
         self._cache_stack = list()
 
@@ -181,13 +196,14 @@ class ModelCache(object):
             model_type=model_type,
             submodel_type=submodel,
         )
-
         # TODO: lock for no copies on simultaneous calls?
         cache_entry = self._cached_models.get(key, None)
         if cache_entry is None:
             self.logger.info(
                 f"Loading model {model_path}, type {base_model.value}:{model_type.value}{':'+submodel.value if submodel else ''}"
             )
+            if self.stats:
+                self.stats.misses += 1
 
             # this will remove older cached models until
             # there is sufficient room to load the requested model
@@ -201,6 +217,17 @@ class ModelCache(object):
 
             cache_entry = _CacheRecord(self, model, mem_used)
             self._cached_models[key] = cache_entry
+        else:
+            if self.stats:
+                self.stats.hits += 1
+
+        if self.stats:
+            self.stats.cache_size = self.max_cache_size * GIG
+            self.stats.high_watermark = max(self.stats.high_watermark, self._cache_size())
+            self.stats.in_cache = len(self._cached_models)
+            self.stats.loaded_model_sizes[key] = max(
+                self.stats.loaded_model_sizes.get(key, 0), model_info.get_size(submodel)
+            )
 
         with suppress(Exception):
             self._cache_stack.remove(key)
@@ -280,14 +307,14 @@ class ModelCache(object):
         """
         Given the HF repo id or path to a model on disk, returns a unique
         hash. Works for legacy checkpoint files, HF models on disk, and HF repo IDs
+
         :param model_path: Path to model file/directory on disk.
         """
         return self._local_model_hash(model_path)
 
     def cache_size(self) -> float:
-        "Return the current size of the cache, in GB"
-        current_cache_size = sum([m.size for m in self._cached_models.values()])
-        return current_cache_size / GIG
+        """Return the current size of the cache, in GB."""
+        return self._cache_size() / GIG
 
     def _has_cuda(self) -> bool:
         return self.execution_device.type == "cuda"
@@ -310,12 +337,15 @@ class ModelCache(object):
             f"Current VRAM/RAM usage: {vram}/{ram}; cached_models/loaded_models/locked_models/ = {cached_models}/{loaded_models}/{locked_models}"
         )
 
+    def _cache_size(self) -> int:
+        return sum([m.size for m in self._cached_models.values()])
+
     def _make_cache_room(self, model_size):
         # calculate how much memory this model will require
         # multiplier = 2 if self.precision==torch.float32 else 1
         bytes_needed = model_size
         maximum_size = self.max_cache_size * GIG  # stored in GB, convert to bytes
-        current_size = sum([m.size for m in self._cached_models.values()])
+        current_size = self._cache_size()
 
         if current_size + bytes_needed > maximum_size:
             self.logger.debug(
@@ -364,6 +394,8 @@ class ModelCache(object):
                     f"Unloading model {model_key} to free {(model_size/GIG):.2f} GB (-{(cache_entry.size/GIG):.2f} GB)"
                 )
                 current_size -= cache_entry.size
+                if self.stats:
+                    self.stats.cleared += 1
                 del self._cache_stack[pos]
                 del self._cached_models[model_key]
                 del cache_entry

invokeai/backend/model_management/model_merge.py

@@ -109,7 +109,7 @@ class ModelMerger(object):
             # pick up the first model's vae
             if mod == model_names[0]:
                 vae = info.get("vae")
-            model_paths.extend([config.root_path / info["path"]])
+            model_paths.extend([(config.root_path / info["path"]).as_posix()])
 
         merge_method = None if interp == "weighted_sum" else MergeInterpolationMethod(interp)
         logger.debug(f"interp = {interp}, merge_method={merge_method}")
@@ -120,11 +120,11 @@ class ModelMerger(object):
             else config.models_path / base_model.value / ModelType.Main.value
         )
         dump_path.mkdir(parents=True, exist_ok=True)
-        dump_path = dump_path / merged_model_name
+        dump_path = (dump_path / merged_model_name).as_posix()
 
         merged_pipe.save_pretrained(dump_path, safe_serialization=True)
         attributes = dict(
-            path=str(dump_path),
+            path=dump_path,
             description=f"Merge of models {', '.join(model_names)}",
             model_format="diffusers",
             variant=ModelVariantType.Normal.value,

invokeai/backend/model_management/model_probe.py

@@ -481,9 +481,19 @@ class ControlNetFolderProbe(FolderProbeBase):
         with open(config_file, "r") as file:
             config = json.load(file)
         # no obvious way to distinguish between sd2-base and sd2-768
-        return (
-            BaseModelType.StableDiffusion1 if config["cross_attention_dim"] == 768 else BaseModelType.StableDiffusion2
+        dimension = config["cross_attention_dim"]
+        base_model = (
+            BaseModelType.StableDiffusion1
+            if dimension == 768
+            else BaseModelType.StableDiffusion2
+            if dimension == 1024
+            else BaseModelType.StableDiffusionXL
+            if dimension == 2048
+            else None
         )
+        if not base_model:
+            raise InvalidModelException(f"Unable to determine model base for {self.folder_path}")
+        return base_model
 
 
 class LoRAFolderProbe(FolderProbeBase):

invokeai/backend/stable_diffusion/__init__.py

@@ -8,4 +8,4 @@ from .diffusers_pipeline import (
 )
 from .diffusion import InvokeAIDiffuserComponent
 from .diffusion.cross_attention_map_saving import AttentionMapSaver
-from .diffusion.shared_invokeai_diffusion import PostprocessingSettings
+from .diffusion.shared_invokeai_diffusion import PostprocessingSettings, BasicConditioningInfo, SDXLConditioningInfo

invokeai/backend/stable_diffusion/diffusers_pipeline.py

@@ -5,23 +5,19 @@ import inspect
 import math
 import secrets
 from dataclasses import dataclass, field
-from typing import Any, Callable, Generic, List, Optional, Type, TypeVar, Union
+from typing import Any, Callable, Generic, List, Optional, Type, Union
 
 import PIL.Image
 import einops
 import psutil
 import torch
 import torchvision.transforms as T
-from accelerate.utils import set_seed
 from diffusers.models import AutoencoderKL, UNet2DConditionModel
 from diffusers.models.controlnet import ControlNetModel
 from diffusers.pipelines.stable_diffusion import StableDiffusionPipelineOutput
 from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion import (
     StableDiffusionPipeline,
 )
-from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion_img2img import (
-    StableDiffusionImg2ImgPipeline,
-)
 from diffusers.pipelines.stable_diffusion.safety_checker import (
     StableDiffusionSafetyChecker,
 )
@@ -30,23 +26,23 @@ from diffusers.schedulers.scheduling_utils import SchedulerMixin, SchedulerOutpu
 from diffusers.utils.import_utils import is_xformers_available
 from diffusers.utils.outputs import BaseOutput
 from pydantic import Field
-from torchvision.transforms.functional import resize as tv_resize
 from transformers import CLIPFeatureExtractor, CLIPTextModel, CLIPTokenizer
-from typing_extensions import ParamSpec
 
 from invokeai.app.services.config import InvokeAIAppConfig
 from .diffusion import (
     AttentionMapSaver,
     InvokeAIDiffuserComponent,
     PostprocessingSettings,
+    BasicConditioningInfo,
 )
 from ..util import normalize_device
 
 
 @dataclass
 class PipelineIntermediateState:
-    run_id: str
     step: int
+    order: int
+    total_steps: int
     timestep: int
     latents: torch.Tensor
     predicted_original: Optional[torch.Tensor] = None
@@ -97,7 +93,6 @@ class AddsMaskGuidance:
     mask_latents: torch.FloatTensor
     scheduler: SchedulerMixin
     noise: torch.Tensor
-    _debug: Optional[Callable] = None
 
     def __call__(self, step_output: Union[BaseOutput, SchedulerOutput], t: torch.Tensor, conditioning) -> BaseOutput:
         output_class = step_output.__class__  # We'll create a new one with masked data.
@@ -134,8 +129,6 @@ class AddsMaskGuidance:
         # mask_latents = self.scheduler.scale_model_input(mask_latents, t)
         mask_latents = einops.repeat(mask_latents, "b c h w -> (repeat b) c h w", repeat=batch_size)
         masked_input = torch.lerp(mask_latents.to(dtype=latents.dtype), latents, mask.to(dtype=latents.dtype))
-        if self._debug:
-            self._debug(masked_input, f"t={t} lerped")
         return masked_input
 
 
@@ -167,33 +160,6 @@ def is_inpainting_model(unet: UNet2DConditionModel):
     return unet.conv_in.in_channels == 9
 
 
-CallbackType = TypeVar("CallbackType")
-ReturnType = TypeVar("ReturnType")
-ParamType = ParamSpec("ParamType")
-
-
-@dataclass(frozen=True)
-class GeneratorToCallbackinator(Generic[ParamType, ReturnType, CallbackType]):
-    """Convert a generator to a function with a callback and a return value."""
-
-    generator_method: Callable[ParamType, ReturnType]
-    callback_arg_type: Type[CallbackType]
-
-    def __call__(
-        self,
-        *args: ParamType.args,
-        callback: Callable[[CallbackType], Any] = None,
-        **kwargs: ParamType.kwargs,
-    ) -> ReturnType:
-        result = None
-        for result in self.generator_method(*args, **kwargs):
-            if callback is not None and isinstance(result, self.callback_arg_type):
-                callback(result)
-        if result is None:
-            raise AssertionError("why was that an empty generator?")
-        return result
-
-
 @dataclass
 class ControlNetData:
     model: ControlNetModel = Field(default=None)
@@ -207,8 +173,8 @@ class ControlNetData:
 
 @dataclass
 class ConditioningData:
-    unconditioned_embeddings: torch.Tensor
-    text_embeddings: torch.Tensor
+    unconditioned_embeddings: BasicConditioningInfo
+    text_embeddings: BasicConditioningInfo
     guidance_scale: Union[float, List[float]]
     """
     Guidance scale as defined in [Classifier-Free Diffusion Guidance](https://arxiv.org/abs/2207.12598).
@@ -284,7 +250,6 @@ class StableDiffusionGeneratorPipeline(StableDiffusionPipeline):
         feature_extractor ([`CLIPFeatureExtractor`]):
             Model that extracts features from generated images to be used as inputs for the `safety_checker`.
     """
-    ID_LENGTH = 8
 
     def __init__(
         self,
@@ -328,33 +293,41 @@ class StableDiffusionGeneratorPipeline(StableDiffusionPipeline):
         if xformers is available, use it, otherwise use sliced attention.
         """
         config = InvokeAIAppConfig.get_config()
-        if torch.cuda.is_available() and is_xformers_available() and not config.disable_xformers:
-            self.enable_xformers_memory_efficient_attention()
+        if self.unet.device.type == "cuda":
+            if is_xformers_available() and not config.disable_xformers:
+                self.enable_xformers_memory_efficient_attention()
+                return
+            elif hasattr(torch.nn.functional, "scaled_dot_product_attention"):
+                # diffusers enable sdp automatically
+                return
+
+        if self.unet.device.type == "cpu" or self.unet.device.type == "mps":
+            mem_free = psutil.virtual_memory().free
+        elif self.unet.device.type == "cuda":
+            mem_free, _ = torch.cuda.mem_get_info(normalize_device(self.unet.device))
         else:
-            if self.device.type == "cpu" or self.device.type == "mps":
-                mem_free = psutil.virtual_memory().free
-            elif self.device.type == "cuda":
-                mem_free, _ = torch.cuda.mem_get_info(normalize_device(self.device))
-            else:
-                raise ValueError(f"unrecognized device {self.device}")
-            # input tensor of [1, 4, h/8, w/8]
-            # output tensor of [16, (h/8 * w/8), (h/8 * w/8)]
-            bytes_per_element_needed_for_baddbmm_duplication = latents.element_size() + 4
-            max_size_required_for_baddbmm = (
-                16
-                * latents.size(dim=2)
-                * latents.size(dim=3)
-                * latents.size(dim=2)
-                * latents.size(dim=3)
-                * bytes_per_element_needed_for_baddbmm_duplication
-            )
-            if max_size_required_for_baddbmm > (mem_free * 3.0 / 4.0):  # 3.3 / 4.0 is from old Invoke code
-                self.enable_attention_slicing(slice_size="max")
-            elif torch.backends.mps.is_available():
-                # diffusers recommends always enabling for mps
-                self.enable_attention_slicing(slice_size="max")
-            else:
-                self.disable_attention_slicing()
+            raise ValueError(f"unrecognized device {self.unet.device}")
+        # input tensor of [1, 4, h/8, w/8]
+        # output tensor of [16, (h/8 * w/8), (h/8 * w/8)]
+        bytes_per_element_needed_for_baddbmm_duplication = latents.element_size() + 4
+        max_size_required_for_baddbmm = (
+            16
+            * latents.size(dim=2)
+            * latents.size(dim=3)
+            * latents.size(dim=2)
+            * latents.size(dim=3)
+            * bytes_per_element_needed_for_baddbmm_duplication
+        )
+        if max_size_required_for_baddbmm > (mem_free * 3.0 / 4.0):  # 3.3 / 4.0 is from old Invoke code
+            self.enable_attention_slicing(slice_size="max")
+        elif torch.backends.mps.is_available():
+            # diffusers recommends always enabling for mps
+            self.enable_attention_slicing(slice_size="max")
+        else:
+            self.disable_attention_slicing()
+
+    def to(self, torch_device: Optional[Union[str, torch.device]] = None, silence_dtype_warnings=False):
+        raise Exception("Should not be called")
 
     def latents_from_embeddings(
         self,
@@ -362,35 +335,72 @@ class StableDiffusionGeneratorPipeline(StableDiffusionPipeline):
         num_inference_steps: int,
         conditioning_data: ConditioningData,
         *,
-        noise: torch.Tensor,
-        timesteps=None,
+        noise: Optional[torch.Tensor],
+        timesteps: torch.Tensor,
+        init_timestep: torch.Tensor,
         additional_guidance: List[Callable] = None,
-        run_id=None,
         callback: Callable[[PipelineIntermediateState], None] = None,
         control_data: List[ControlNetData] = None,
+        mask: Optional[torch.Tensor] = None,
+        seed: Optional[int] = None,
     ) -> tuple[torch.Tensor, Optional[AttentionMapSaver]]:
-        if self.scheduler.config.get("cpu_only", False):
-            scheduler_device = torch.device("cpu")
-        else:
-            scheduler_device = self.unet.device
+        if init_timestep.shape[0] == 0:
+            return latents, None
 
-        if timesteps is None:
-            self.scheduler.set_timesteps(num_inference_steps, device=scheduler_device)
-            timesteps = self.scheduler.timesteps
-        infer_latents_from_embeddings = GeneratorToCallbackinator(
-            self.generate_latents_from_embeddings, PipelineIntermediateState
-        )
-        result: PipelineIntermediateState = infer_latents_from_embeddings(
-            latents,
-            timesteps,
-            conditioning_data,
-            noise=noise,
-            run_id=run_id,
-            additional_guidance=additional_guidance,
-            control_data=control_data,
-            callback=callback,
-        )
-        return result.latents, result.attention_map_saver
+        if additional_guidance is None:
+            additional_guidance = []
+
+        orig_latents = latents.clone()
+
+        batch_size = latents.shape[0]
+        batched_t = init_timestep.expand(batch_size)
+
+        if noise is not None:
+            # latents = noise * self.scheduler.init_noise_sigma # it's like in t2l according to diffusers
+            latents = self.scheduler.add_noise(latents, noise, batched_t)
+
+        if mask is not None:
+            if is_inpainting_model(self.unet):
+                # You'd think the inpainting model wouldn't be paying attention to the area it is going to repaint
+                # (that's why there's a mask!) but it seems to really want that blanked out.
+                # masked_latents = latents * torch.where(mask < 0.5, 1, 0) TODO: inpaint/outpaint/infill
+
+                # TODO: we should probably pass this in so we don't have to try/finally around setting it.
+                self.invokeai_diffuser.model_forward_callback = AddsMaskLatents(self._unet_forward, mask, orig_latents)
+            else:
+                # if no noise provided, noisify unmasked area based on seed(or 0 as fallback)
+                if noise is None:
+                    noise = torch.randn(
+                        orig_latents.shape,
+                        dtype=torch.float32,
+                        device="cpu",
+                        generator=torch.Generator(device="cpu").manual_seed(seed or 0),
+                    ).to(device=orig_latents.device, dtype=orig_latents.dtype)
+
+                    latents = self.scheduler.add_noise(latents, noise, batched_t)
+                    latents = torch.lerp(
+                        orig_latents, latents.to(dtype=orig_latents.dtype), mask.to(dtype=orig_latents.dtype)
+                    )
+
+                additional_guidance.append(AddsMaskGuidance(mask, orig_latents, self.scheduler, noise))
+
+        try:
+            latents, attention_map_saver = self.generate_latents_from_embeddings(
+                latents,
+                timesteps,
+                conditioning_data,
+                additional_guidance=additional_guidance,
+                control_data=control_data,
+                callback=callback,
+            )
+        finally:
+            self.invokeai_diffuser.model_forward_callback = self._unet_forward
+
+        # restore unmasked part
+        if mask is not None:
+            latents = torch.lerp(orig_latents, latents.to(dtype=orig_latents.dtype), mask.to(dtype=orig_latents.dtype))
+
+        return latents, attention_map_saver
 
     def generate_latents_from_embeddings(
         self,
@@ -398,42 +408,40 @@ class StableDiffusionGeneratorPipeline(StableDiffusionPipeline):
         timesteps,
         conditioning_data: ConditioningData,
         *,
-        noise: torch.Tensor,
-        run_id: str = None,
         additional_guidance: List[Callable] = None,
         control_data: List[ControlNetData] = None,
+        callback: Callable[[PipelineIntermediateState], None] = None,
     ):
         self._adjust_memory_efficient_attention(latents)
-        if run_id is None:
-            run_id = secrets.token_urlsafe(self.ID_LENGTH)
         if additional_guidance is None:
             additional_guidance = []
+
+        batch_size = latents.shape[0]
+        attention_map_saver: Optional[AttentionMapSaver] = None
+
+        if timesteps.shape[0] == 0:
+            return latents, attention_map_saver
+
         extra_conditioning_info = conditioning_data.extra
         with self.invokeai_diffuser.custom_attention_context(
             self.invokeai_diffuser.model,
             extra_conditioning_info=extra_conditioning_info,
             step_count=len(self.scheduler.timesteps),
         ):
-            yield PipelineIntermediateState(
-                run_id=run_id,
-                step=-1,
-                timestep=self.scheduler.config.num_train_timesteps,
-                latents=latents,
-            )
+            if callback is not None:
+                callback(
+                    PipelineIntermediateState(
+                        step=-1,
+                        order=self.scheduler.order,
+                        total_steps=len(timesteps),
+                        timestep=self.scheduler.config.num_train_timesteps,
+                        latents=latents,
+                    )
+                )
 
-            batch_size = latents.shape[0]
-            batched_t = torch.full(
-                (batch_size,),
-                timesteps[0],
-                dtype=timesteps.dtype,
-                device=self.unet.device,
-            )
-            latents = self.scheduler.add_noise(latents, noise, batched_t)
-
-            attention_map_saver: Optional[AttentionMapSaver] = None
             # print("timesteps:", timesteps)
             for i, t in enumerate(self.progress_bar(timesteps)):
-                batched_t.fill_(t)
+                batched_t = t.expand(batch_size)
                 step_output = self.step(
                     batched_t,
                     latents,
@@ -462,14 +470,18 @@ class StableDiffusionGeneratorPipeline(StableDiffusionPipeline):
                 #    attention_map_saver = AttentionMapSaver(token_ids=attention_map_token_ids, latents_shape=latents.shape[-2:])
                 #    self.invokeai_diffuser.setup_attention_map_saving(attention_map_saver)
 
-                yield PipelineIntermediateState(
-                    run_id=run_id,
-                    step=i,
-                    timestep=int(t),
-                    latents=latents,
-                    predicted_original=predicted_original,
-                    attention_map_saver=attention_map_saver,
-                )
+                if callback is not None:
+                    callback(
+                        PipelineIntermediateState(
+                            step=i,
+                            order=self.scheduler.order,
+                            total_steps=len(timesteps),
+                            timestep=int(t),
+                            latents=latents,
+                            predicted_original=predicted_original,
+                            attention_map_saver=attention_map_saver,
+                        )
+                    )
 
             return latents, attention_map_saver
 
@@ -491,95 +503,39 @@ class StableDiffusionGeneratorPipeline(StableDiffusionPipeline):
 
         # TODO: should this scaling happen here or inside self._unet_forward?
         #     i.e. before or after passing it to InvokeAIDiffuserComponent
-        unet_latent_input = self.scheduler.scale_model_input(latents, timestep)
+        latent_model_input = self.scheduler.scale_model_input(latents, timestep)
 
         # default is no controlnet, so set controlnet processing output to None
-        down_block_res_samples, mid_block_res_sample = None, None
-
+        controlnet_down_block_samples, controlnet_mid_block_sample = None, None
         if control_data is not None:
-            # control_data should be type List[ControlNetData]
-            # this loop covers both ControlNet (one ControlNetData in list)
-            #      and MultiControlNet (multiple ControlNetData in list)
-            for i, control_datum in enumerate(control_data):
-                control_mode = control_datum.control_mode
-                # soft_injection and cfg_injection are the two ControlNet control_mode booleans
-                #     that are combined at higher level to make control_mode enum
-                #  soft_injection determines whether to do per-layer re-weighting adjustment (if True)
-                #     or default weighting (if False)
-                soft_injection = control_mode == "more_prompt" or control_mode == "more_control"
-                #  cfg_injection = determines whether to apply ControlNet to only the conditional (if True)
-                #      or the default both conditional and unconditional (if False)
-                cfg_injection = control_mode == "more_control" or control_mode == "unbalanced"
+            controlnet_down_block_samples, controlnet_mid_block_sample = self.invokeai_diffuser.do_controlnet_step(
+                control_data=control_data,
+                sample=latent_model_input,
+                timestep=timestep,
+                step_index=step_index,
+                total_step_count=total_step_count,
+                conditioning_data=conditioning_data,
+            )
 
-                first_control_step = math.floor(control_datum.begin_step_percent * total_step_count)
-                last_control_step = math.ceil(control_datum.end_step_percent * total_step_count)
-                # only apply controlnet if current step is within the controlnet's begin/end step range
-                if step_index >= first_control_step and step_index <= last_control_step:
-                    if cfg_injection:
-                        control_latent_input = unet_latent_input
-                    else:
-                        # expand the latents input to control model if doing classifier free guidance
-                        #    (which I think for now is always true, there is conditional elsewhere that stops execution if
-                        #     classifier_free_guidance is <= 1.0 ?)
-                        control_latent_input = torch.cat([unet_latent_input] * 2)
-
-                    if cfg_injection:  # only applying ControlNet to conditional instead of in unconditioned
-                        encoder_hidden_states = conditioning_data.text_embeddings
-                        encoder_attention_mask = None
-                    else:
-                        (
-                            encoder_hidden_states,
-                            encoder_attention_mask,
-                        ) = self.invokeai_diffuser._concat_conditionings_for_batch(
-                            conditioning_data.unconditioned_embeddings,
-                            conditioning_data.text_embeddings,
-                        )
-                    if isinstance(control_datum.weight, list):
-                        # if controlnet has multiple weights, use the weight for the current step
-                        controlnet_weight = control_datum.weight[step_index]
-                    else:
-                        # if controlnet has a single weight, use it for all steps
-                        controlnet_weight = control_datum.weight
-
-                    # controlnet(s) inference
-                    down_samples, mid_sample = control_datum.model(
-                        sample=control_latent_input,
-                        timestep=timestep,
-                        encoder_hidden_states=encoder_hidden_states,
-                        controlnet_cond=control_datum.image_tensor,
-                        conditioning_scale=controlnet_weight,  # controlnet specific, NOT the guidance scale
-                        encoder_attention_mask=encoder_attention_mask,
-                        guess_mode=soft_injection,  # this is still called guess_mode in diffusers ControlNetModel
-                        return_dict=False,
-                    )
-                    if cfg_injection:
-                        # Inferred ControlNet only for the conditional batch.
-                        # To apply the output of ControlNet to both the unconditional and conditional batches,
-                        #    prepend zeros for unconditional batch
-                        down_samples = [torch.cat([torch.zeros_like(d), d]) for d in down_samples]
-                        mid_sample = torch.cat([torch.zeros_like(mid_sample), mid_sample])
-
-                    if down_block_res_samples is None and mid_block_res_sample is None:
-                        down_block_res_samples, mid_block_res_sample = down_samples, mid_sample
-                    else:
-                        # add controlnet outputs together if have multiple controlnets
-                        down_block_res_samples = [
-                            samples_prev + samples_curr
-                            for samples_prev, samples_curr in zip(down_block_res_samples, down_samples)
-                        ]
-                        mid_block_res_sample += mid_sample
-
-        # predict the noise residual
-        noise_pred = self.invokeai_diffuser.do_diffusion_step(
-            x=unet_latent_input,
-            sigma=t,
-            unconditioning=conditioning_data.unconditioned_embeddings,
-            conditioning=conditioning_data.text_embeddings,
-            unconditional_guidance_scale=conditioning_data.guidance_scale,
+        uc_noise_pred, c_noise_pred = self.invokeai_diffuser.do_unet_step(
+            sample=latent_model_input,
+            timestep=t,  # TODO: debug how handled batched and non batched timesteps
             step_index=step_index,
             total_step_count=total_step_count,
-            down_block_additional_residuals=down_block_res_samples,  # from controlnet(s)
-            mid_block_additional_residual=mid_block_res_sample,  # from controlnet(s)
+            conditioning_data=conditioning_data,
+            # extra:
+            down_block_additional_residuals=controlnet_down_block_samples,  # from controlnet(s)
+            mid_block_additional_residual=controlnet_mid_block_sample,  # from controlnet(s)
+        )
+
+        guidance_scale = conditioning_data.guidance_scale
+        if isinstance(guidance_scale, list):
+            guidance_scale = guidance_scale[step_index]
+
+        noise_pred = self.invokeai_diffuser._combine(
+            uc_noise_pred,
+            c_noise_pred,
+            guidance_scale,
         )
 
         # compute the previous noisy sample x_t -> x_t-1
@@ -621,126 +577,3 @@ class StableDiffusionGeneratorPipeline(StableDiffusionPipeline):
             cross_attention_kwargs=cross_attention_kwargs,
             **kwargs,
         ).sample
-
-    def get_img2img_timesteps(self, num_inference_steps: int, strength: float, device=None) -> (torch.Tensor, int):
-        img2img_pipeline = StableDiffusionImg2ImgPipeline(**self.components)
-        assert img2img_pipeline.scheduler is self.scheduler
-
-        if self.scheduler.config.get("cpu_only", False):
-            scheduler_device = torch.device("cpu")
-        else:
-            scheduler_device = self.unet.device
-
-        img2img_pipeline.scheduler.set_timesteps(num_inference_steps, device=scheduler_device)
-        timesteps, adjusted_steps = img2img_pipeline.get_timesteps(
-            num_inference_steps, strength, device=scheduler_device
-        )
-        # Workaround for low strength resulting in zero timesteps.
-        # TODO: submit upstream fix for zero-step img2img
-        if timesteps.numel() == 0:
-            timesteps = self.scheduler.timesteps[-1:]
-            adjusted_steps = timesteps.numel()
-        return timesteps, adjusted_steps
-
-    def inpaint_from_embeddings(
-        self,
-        init_image: torch.FloatTensor,
-        mask: torch.FloatTensor,
-        strength: float,
-        num_inference_steps: int,
-        conditioning_data: ConditioningData,
-        *,
-        callback: Callable[[PipelineIntermediateState], None] = None,
-        run_id=None,
-        noise_func=None,
-        seed=None,
-    ) -> InvokeAIStableDiffusionPipelineOutput:
-        device = self.unet.device
-        latents_dtype = self.unet.dtype
-
-        if isinstance(init_image, PIL.Image.Image):
-            init_image = image_resized_to_grid_as_tensor(init_image.convert("RGB"))
-
-        init_image = init_image.to(device=device, dtype=latents_dtype)
-        mask = mask.to(device=device, dtype=latents_dtype)
-
-        if init_image.dim() == 3:
-            init_image = init_image.unsqueeze(0)
-
-        timesteps, _ = self.get_img2img_timesteps(num_inference_steps, strength)
-
-        # 6. Prepare latent variables
-        # can't quite use upstream StableDiffusionImg2ImgPipeline.prepare_latents
-        # because we have our own noise function
-        init_image_latents = self.non_noised_latents_from_image(init_image, device=device, dtype=latents_dtype)
-        if seed is not None:
-            set_seed(seed)
-        noise = noise_func(init_image_latents)
-
-        if mask.dim() == 3:
-            mask = mask.unsqueeze(0)
-        latent_mask = tv_resize(mask, init_image_latents.shape[-2:], T.InterpolationMode.BILINEAR).to(
-            device=device, dtype=latents_dtype
-        )
-
-        guidance: List[Callable] = []
-
-        if is_inpainting_model(self.unet):
-            # You'd think the inpainting model wouldn't be paying attention to the area it is going to repaint
-            # (that's why there's a mask!) but it seems to really want that blanked out.
-            masked_init_image = init_image * torch.where(mask < 0.5, 1, 0)
-            masked_latents = self.non_noised_latents_from_image(masked_init_image, device=device, dtype=latents_dtype)
-
-            # TODO: we should probably pass this in so we don't have to try/finally around setting it.
-            self.invokeai_diffuser.model_forward_callback = AddsMaskLatents(
-                self._unet_forward, latent_mask, masked_latents
-            )
-        else:
-            guidance.append(AddsMaskGuidance(latent_mask, init_image_latents, self.scheduler, noise))
-
-        try:
-            result_latents, result_attention_maps = self.latents_from_embeddings(
-                latents=init_image_latents
-                if strength < 1.0
-                else torch.zeros_like(
-                    init_image_latents, device=init_image_latents.device, dtype=init_image_latents.dtype
-                ),
-                num_inference_steps=num_inference_steps,
-                conditioning_data=conditioning_data,
-                noise=noise,
-                timesteps=timesteps,
-                additional_guidance=guidance,
-                run_id=run_id,
-                callback=callback,
-            )
-        finally:
-            self.invokeai_diffuser.model_forward_callback = self._unet_forward
-
-        # https://discuss.huggingface.co/t/memory-usage-by-later-pipeline-stages/23699
-        torch.cuda.empty_cache()
-
-        with torch.inference_mode():
-            image = self.decode_latents(result_latents)
-            output = InvokeAIStableDiffusionPipelineOutput(
-                images=image,
-                nsfw_content_detected=[],
-                attention_map_saver=result_attention_maps,
-            )
-            return output
-
-    def non_noised_latents_from_image(self, init_image, *, device: torch.device, dtype):
-        init_image = init_image.to(device=device, dtype=dtype)
-        with torch.inference_mode():
-            init_latent_dist = self.vae.encode(init_image).latent_dist
-            init_latents = init_latent_dist.sample().to(dtype=dtype)  # FIXME: uses torch.randn. make reproducible!
-
-        init_latents = 0.18215 * init_latents
-        return init_latents
-
-    def debug_latents(self, latents, msg):
-        from invokeai.backend.image_util import debug_image
-
-        with torch.inference_mode():
-            decoded = self.numpy_to_pil(self.decode_latents(latents))
-        for i, img in enumerate(decoded):
-            debug_image(img, f"latents {msg} {i+1}/{len(decoded)}", debug_status=True)

invokeai/backend/stable_diffusion/diffusion/__init__.py

@@ -3,4 +3,9 @@ Initialization file for invokeai.models.diffusion
 """
 from .cross_attention_control import InvokeAICrossAttentionMixin
 from .cross_attention_map_saving import AttentionMapSaver
-from .shared_invokeai_diffusion import InvokeAIDiffuserComponent, PostprocessingSettings
+from .shared_invokeai_diffusion import (
+    InvokeAIDiffuserComponent,
+    PostprocessingSettings,
+    BasicConditioningInfo,
+    SDXLConditioningInfo,
+)

invokeai/backend/stable_diffusion/diffusion/shared_invokeai_diffusion.py

@@ -1,6 +1,8 @@
+from __future__ import annotations
+
 from contextlib import contextmanager
 from dataclasses import dataclass
-from math import ceil
+import math
 from typing import Any, Callable, Dict, Optional, Union, List
 
 import numpy as np
@@ -32,6 +34,29 @@ ModelForwardCallback: TypeAlias = Union[
 ]
 
 
+@dataclass
+class BasicConditioningInfo:
+    embeds: torch.Tensor
+    extra_conditioning: Optional[InvokeAIDiffuserComponent.ExtraConditioningInfo]
+    # weight: float
+    # mode: ConditioningAlgo
+
+    def to(self, device, dtype=None):
+        self.embeds = self.embeds.to(device=device, dtype=dtype)
+        return self
+
+
+@dataclass
+class SDXLConditioningInfo(BasicConditioningInfo):
+    pooled_embeds: torch.Tensor
+    add_time_ids: torch.Tensor
+
+    def to(self, device, dtype=None):
+        self.pooled_embeds = self.pooled_embeds.to(device=device, dtype=dtype)
+        self.add_time_ids = self.add_time_ids.to(device=device, dtype=dtype)
+        return super().to(device=device, dtype=dtype)
+
+
 @dataclass(frozen=True)
 class PostprocessingSettings:
     threshold: float
@@ -127,33 +152,126 @@ class InvokeAIDiffuserComponent:
         for _, module in tokens_cross_attention_modules:
             module.set_attention_slice_calculated_callback(None)
 
-    def do_diffusion_step(
+    def do_controlnet_step(
         self,
-        x: torch.Tensor,
-        sigma: torch.Tensor,
-        unconditioning: Union[torch.Tensor, dict],
-        conditioning: Union[torch.Tensor, dict],
-        # unconditional_guidance_scale: float,
-        unconditional_guidance_scale: Union[float, List[float]],
-        step_index: Optional[int] = None,
-        total_step_count: Optional[int] = None,
+        control_data,
+        sample: torch.Tensor,
+        timestep: torch.Tensor,
+        step_index: int,
+        total_step_count: int,
+        conditioning_data,
+    ):
+        down_block_res_samples, mid_block_res_sample = None, None
+
+        # control_data should be type List[ControlNetData]
+        # this loop covers both ControlNet (one ControlNetData in list)
+        #      and MultiControlNet (multiple ControlNetData in list)
+        for i, control_datum in enumerate(control_data):
+            control_mode = control_datum.control_mode
+            # soft_injection and cfg_injection are the two ControlNet control_mode booleans
+            #     that are combined at higher level to make control_mode enum
+            #  soft_injection determines whether to do per-layer re-weighting adjustment (if True)
+            #     or default weighting (if False)
+            soft_injection = control_mode == "more_prompt" or control_mode == "more_control"
+            #  cfg_injection = determines whether to apply ControlNet to only the conditional (if True)
+            #      or the default both conditional and unconditional (if False)
+            cfg_injection = control_mode == "more_control" or control_mode == "unbalanced"
+
+            first_control_step = math.floor(control_datum.begin_step_percent * total_step_count)
+            last_control_step = math.ceil(control_datum.end_step_percent * total_step_count)
+            # only apply controlnet if current step is within the controlnet's begin/end step range
+            if step_index >= first_control_step and step_index <= last_control_step:
+                if cfg_injection:
+                    sample_model_input = sample
+                else:
+                    # expand the latents input to control model if doing classifier free guidance
+                    #    (which I think for now is always true, there is conditional elsewhere that stops execution if
+                    #     classifier_free_guidance is <= 1.0 ?)
+                    sample_model_input = torch.cat([sample] * 2)
+
+                added_cond_kwargs = None
+
+                if cfg_injection:  # only applying ControlNet to conditional instead of in unconditioned
+                    if type(conditioning_data.text_embeddings) is SDXLConditioningInfo:
+                        added_cond_kwargs = {
+                            "text_embeds": conditioning_data.text_embeddings.pooled_embeds,
+                            "time_ids": conditioning_data.text_embeddings.add_time_ids,
+                        }
+                    encoder_hidden_states = conditioning_data.text_embeddings.embeds
+                    encoder_attention_mask = None
+                else:
+                    if type(conditioning_data.text_embeddings) is SDXLConditioningInfo:
+                        added_cond_kwargs = {
+                            "text_embeds": torch.cat(
+                                [
+                                    # TODO: how to pad? just by zeros? or even truncate?
+                                    conditioning_data.unconditioned_embeddings.pooled_embeds,
+                                    conditioning_data.text_embeddings.pooled_embeds,
+                                ],
+                                dim=0,
+                            ),
+                            "time_ids": torch.cat(
+                                [
+                                    conditioning_data.unconditioned_embeddings.add_time_ids,
+                                    conditioning_data.text_embeddings.add_time_ids,
+                                ],
+                                dim=0,
+                            ),
+                        }
+                    (
+                        encoder_hidden_states,
+                        encoder_attention_mask,
+                    ) = self._concat_conditionings_for_batch(
+                        conditioning_data.unconditioned_embeddings.embeds,
+                        conditioning_data.text_embeddings.embeds,
+                    )
+                if isinstance(control_datum.weight, list):
+                    # if controlnet has multiple weights, use the weight for the current step
+                    controlnet_weight = control_datum.weight[step_index]
+                else:
+                    # if controlnet has a single weight, use it for all steps
+                    controlnet_weight = control_datum.weight
+
+                # controlnet(s) inference
+                down_samples, mid_sample = control_datum.model(
+                    sample=sample_model_input,
+                    timestep=timestep,
+                    encoder_hidden_states=encoder_hidden_states,
+                    controlnet_cond=control_datum.image_tensor,
+                    conditioning_scale=controlnet_weight,  # controlnet specific, NOT the guidance scale
+                    encoder_attention_mask=encoder_attention_mask,
+                    added_cond_kwargs=added_cond_kwargs,
+                    guess_mode=soft_injection,  # this is still called guess_mode in diffusers ControlNetModel
+                    return_dict=False,
+                )
+                if cfg_injection:
+                    # Inferred ControlNet only for the conditional batch.
+                    # To apply the output of ControlNet to both the unconditional and conditional batches,
+                    #    prepend zeros for unconditional batch
+                    down_samples = [torch.cat([torch.zeros_like(d), d]) for d in down_samples]
+                    mid_sample = torch.cat([torch.zeros_like(mid_sample), mid_sample])
+
+                if down_block_res_samples is None and mid_block_res_sample is None:
+                    down_block_res_samples, mid_block_res_sample = down_samples, mid_sample
+                else:
+                    # add controlnet outputs together if have multiple controlnets
+                    down_block_res_samples = [
+                        samples_prev + samples_curr
+                        for samples_prev, samples_curr in zip(down_block_res_samples, down_samples)
+                    ]
+                    mid_block_res_sample += mid_sample
+
+        return down_block_res_samples, mid_block_res_sample
+
+    def do_unet_step(
+        self,
+        sample: torch.Tensor,
+        timestep: torch.Tensor,
+        conditioning_data,  # TODO: type
+        step_index: int,
+        total_step_count: int,
         **kwargs,
     ):
-        """
-        :param x: current latents
-        :param sigma: aka t, passed to the internal model to control how much denoising will occur
-        :param unconditioning: embeddings for unconditioned output. for hybrid conditioning this is a dict of tensors [B x 77 x 768], otherwise a single tensor [B x 77 x 768]
-        :param conditioning: embeddings for conditioned output. for hybrid conditioning this is a dict of tensors [B x 77 x 768], otherwise a single tensor [B x 77 x 768]
-        :param unconditional_guidance_scale: aka CFG scale, controls how much effect the conditioning tensor has
-        :param step_index: counts upwards from 0 to (step_count-1) (as passed to setup_cross_attention_control, if using). May be called multiple times for a single step, therefore do not assume that its value will monotically increase. If None, will be estimated by comparing sigma against self.model.sigmas .
-        :return: the new latents after applying the model to x using unscaled unconditioning and CFG-scaled conditioning.
-        """
-
-        if isinstance(unconditional_guidance_scale, list):
-            guidance_scale = unconditional_guidance_scale[step_index]
-        else:
-            guidance_scale = unconditional_guidance_scale
-
         cross_attention_control_types_to_do = []
         context: Context = self.cross_attention_control_context
         if self.cross_attention_control_context is not None:
@@ -163,25 +281,15 @@ class InvokeAIDiffuserComponent:
             )
 
         wants_cross_attention_control = len(cross_attention_control_types_to_do) > 0
-        wants_hybrid_conditioning = isinstance(conditioning, dict)
 
-        if wants_hybrid_conditioning:
-            unconditioned_next_x, conditioned_next_x = self._apply_hybrid_conditioning(
-                x,
-                sigma,
-                unconditioning,
-                conditioning,
-                **kwargs,
-            )
-        elif wants_cross_attention_control:
+        if wants_cross_attention_control:
             (
                 unconditioned_next_x,
                 conditioned_next_x,
             ) = self._apply_cross_attention_controlled_conditioning(
-                x,
-                sigma,
-                unconditioning,
-                conditioning,
+                sample,
+                timestep,
+                conditioning_data,
                 cross_attention_control_types_to_do,
                 **kwargs,
             )
@@ -190,10 +298,9 @@ class InvokeAIDiffuserComponent:
                 unconditioned_next_x,
                 conditioned_next_x,
             ) = self._apply_standard_conditioning_sequentially(
-                x,
-                sigma,
-                unconditioning,
-                conditioning,
+                sample,
+                timestep,
+                conditioning_data,
                 **kwargs,
             )
 
@@ -202,21 +309,13 @@ class InvokeAIDiffuserComponent:
                 unconditioned_next_x,
                 conditioned_next_x,
             ) = self._apply_standard_conditioning(
-                x,
-                sigma,
-                unconditioning,
-                conditioning,
+                sample,
+                timestep,
+                conditioning_data,
                 **kwargs,
             )
 
-        combined_next_x = self._combine(
-            # unconditioned_next_x, conditioned_next_x, unconditional_guidance_scale
-            unconditioned_next_x,
-            conditioned_next_x,
-            guidance_scale,
-        )
-
-        return combined_next_x
+        return unconditioned_next_x, conditioned_next_x
 
     def do_latent_postprocessing(
         self,
@@ -228,7 +327,6 @@ class InvokeAIDiffuserComponent:
     ) -> torch.Tensor:
         if postprocessing_settings is not None:
             percent_through = step_index / total_step_count
-            latents = self.apply_threshold(postprocessing_settings, latents, percent_through)
             latents = self.apply_symmetry(postprocessing_settings, latents, percent_through)
         return latents
 
@@ -281,17 +379,40 @@ class InvokeAIDiffuserComponent:
 
     # methods below are called from do_diffusion_step and should be considered private to this class.
 
-    def _apply_standard_conditioning(self, x, sigma, unconditioning, conditioning, **kwargs):
+    def _apply_standard_conditioning(self, x, sigma, conditioning_data, **kwargs):
         # fast batched path
         x_twice = torch.cat([x] * 2)
         sigma_twice = torch.cat([sigma] * 2)
 
-        both_conditionings, encoder_attention_mask = self._concat_conditionings_for_batch(unconditioning, conditioning)
+        added_cond_kwargs = None
+        if type(conditioning_data.text_embeddings) is SDXLConditioningInfo:
+            added_cond_kwargs = {
+                "text_embeds": torch.cat(
+                    [
+                        # TODO: how to pad? just by zeros? or even truncate?
+                        conditioning_data.unconditioned_embeddings.pooled_embeds,
+                        conditioning_data.text_embeddings.pooled_embeds,
+                    ],
+                    dim=0,
+                ),
+                "time_ids": torch.cat(
+                    [
+                        conditioning_data.unconditioned_embeddings.add_time_ids,
+                        conditioning_data.text_embeddings.add_time_ids,
+                    ],
+                    dim=0,
+                ),
+            }
+
+        both_conditionings, encoder_attention_mask = self._concat_conditionings_for_batch(
+            conditioning_data.unconditioned_embeddings.embeds, conditioning_data.text_embeddings.embeds
+        )
         both_results = self.model_forward_callback(
             x_twice,
             sigma_twice,
             both_conditionings,
             encoder_attention_mask=encoder_attention_mask,
+            added_cond_kwargs=added_cond_kwargs,
             **kwargs,
         )
         unconditioned_next_x, conditioned_next_x = both_results.chunk(2)
@@ -301,8 +422,7 @@ class InvokeAIDiffuserComponent:
         self,
         x: torch.Tensor,
         sigma,
-        unconditioning: torch.Tensor,
-        conditioning: torch.Tensor,
+        conditioning_data,
         **kwargs,
     ):
         # low-memory sequential path
@@ -320,52 +440,46 @@ class InvokeAIDiffuserComponent:
         if mid_block_additional_residual is not None:
             uncond_mid_block, cond_mid_block = mid_block_additional_residual.chunk(2)
 
+        added_cond_kwargs = None
+        is_sdxl = type(conditioning_data.text_embeddings) is SDXLConditioningInfo
+        if is_sdxl:
+            added_cond_kwargs = {
+                "text_embeds": conditioning_data.unconditioned_embeddings.pooled_embeds,
+                "time_ids": conditioning_data.unconditioned_embeddings.add_time_ids,
+            }
+
         unconditioned_next_x = self.model_forward_callback(
             x,
             sigma,
-            unconditioning,
+            conditioning_data.unconditioned_embeddings.embeds,
             down_block_additional_residuals=uncond_down_block,
             mid_block_additional_residual=uncond_mid_block,
+            added_cond_kwargs=added_cond_kwargs,
             **kwargs,
         )
+
+        if is_sdxl:
+            added_cond_kwargs = {
+                "text_embeds": conditioning_data.text_embeddings.pooled_embeds,
+                "time_ids": conditioning_data.text_embeddings.add_time_ids,
+            }
+
         conditioned_next_x = self.model_forward_callback(
             x,
             sigma,
-            conditioning,
+            conditioning_data.text_embeddings.embeds,
             down_block_additional_residuals=cond_down_block,
             mid_block_additional_residual=cond_mid_block,
+            added_cond_kwargs=added_cond_kwargs,
             **kwargs,
         )
         return unconditioned_next_x, conditioned_next_x
 
-    # TODO: looks unused
-    def _apply_hybrid_conditioning(self, x, sigma, unconditioning, conditioning, **kwargs):
-        assert isinstance(conditioning, dict)
-        assert isinstance(unconditioning, dict)
-        x_twice = torch.cat([x] * 2)
-        sigma_twice = torch.cat([sigma] * 2)
-        both_conditionings = dict()
-        for k in conditioning:
-            if isinstance(conditioning[k], list):
-                both_conditionings[k] = [
-                    torch.cat([unconditioning[k][i], conditioning[k][i]]) for i in range(len(conditioning[k]))
-                ]
-            else:
-                both_conditionings[k] = torch.cat([unconditioning[k], conditioning[k]])
-        unconditioned_next_x, conditioned_next_x = self.model_forward_callback(
-            x_twice,
-            sigma_twice,
-            both_conditionings,
-            **kwargs,
-        ).chunk(2)
-        return unconditioned_next_x, conditioned_next_x
-
     def _apply_cross_attention_controlled_conditioning(
         self,
         x: torch.Tensor,
         sigma,
-        unconditioning,
-        conditioning,
+        conditioning_data,
         cross_attention_control_types_to_do,
         **kwargs,
     ):
@@ -391,26 +505,43 @@ class InvokeAIDiffuserComponent:
             mask=context.cross_attention_mask,
             cross_attention_types_to_do=[],
         )
+
+        added_cond_kwargs = None
+        is_sdxl = type(conditioning_data.text_embeddings) is SDXLConditioningInfo
+        if is_sdxl:
+            added_cond_kwargs = {
+                "text_embeds": conditioning_data.unconditioned_embeddings.pooled_embeds,
+                "time_ids": conditioning_data.unconditioned_embeddings.add_time_ids,
+            }
+
         # no cross attention for unconditioning (negative prompt)
         unconditioned_next_x = self.model_forward_callback(
             x,
             sigma,
-            unconditioning,
+            conditioning_data.unconditioned_embeddings.embeds,
             {"swap_cross_attn_context": cross_attn_processor_context},
             down_block_additional_residuals=uncond_down_block,
             mid_block_additional_residual=uncond_mid_block,
+            added_cond_kwargs=added_cond_kwargs,
             **kwargs,
         )
 
+        if is_sdxl:
+            added_cond_kwargs = {
+                "text_embeds": conditioning_data.text_embeddings.pooled_embeds,
+                "time_ids": conditioning_data.text_embeddings.add_time_ids,
+            }
+
         # do requested cross attention types for conditioning (positive prompt)
         cross_attn_processor_context.cross_attention_types_to_do = cross_attention_control_types_to_do
         conditioned_next_x = self.model_forward_callback(
             x,
             sigma,
-            conditioning,
+            conditioning_data.text_embeddings.embeds,
             {"swap_cross_attn_context": cross_attn_processor_context},
             down_block_additional_residuals=cond_down_block,
             mid_block_additional_residual=cond_mid_block,
+            added_cond_kwargs=added_cond_kwargs,
             **kwargs,
         )
         return unconditioned_next_x, conditioned_next_x
@@ -421,63 +552,6 @@ class InvokeAIDiffuserComponent:
         combined_next_x = unconditioned_next_x + scaled_delta
         return combined_next_x
 
-    def apply_threshold(
-        self,
-        postprocessing_settings: PostprocessingSettings,
-        latents: torch.Tensor,
-        percent_through: float,
-    ) -> torch.Tensor:
-        if postprocessing_settings.threshold is None or postprocessing_settings.threshold == 0.0:
-            return latents
-
-        threshold = postprocessing_settings.threshold
-        warmup = postprocessing_settings.warmup
-
-        if percent_through < warmup:
-            current_threshold = threshold + threshold * 5 * (1 - (percent_through / warmup))
-        else:
-            current_threshold = threshold
-
-        if current_threshold <= 0:
-            return latents
-
-        maxval = latents.max().item()
-        minval = latents.min().item()
-
-        scale = 0.7  # default value from #395
-
-        if self.debug_thresholding:
-            std, mean = [i.item() for i in torch.std_mean(latents)]
-            outside = torch.count_nonzero((latents < -current_threshold) | (latents > current_threshold))
-            logger.info(f"Threshold: %={percent_through} threshold={current_threshold:.3f} (of {threshold:.3f})")
-            logger.debug(f"min, mean, max = {minval:.3f}, {mean:.3f}, {maxval:.3f}\tstd={std}")
-            logger.debug(f"{outside / latents.numel() * 100:.2f}% values outside threshold")
-
-        if maxval < current_threshold and minval > -current_threshold:
-            return latents
-
-        num_altered = 0
-
-        # MPS torch.rand_like is fine because torch.rand_like is wrapped in generate.py!
-
-        if maxval > current_threshold:
-            latents = torch.clone(latents)
-            maxval = np.clip(maxval * scale, 1, current_threshold)
-            num_altered += torch.count_nonzero(latents > maxval)
-            latents[latents > maxval] = torch.rand_like(latents[latents > maxval]) * maxval
-
-        if minval < -current_threshold:
-            latents = torch.clone(latents)
-            minval = np.clip(minval * scale, -current_threshold, -1)
-            num_altered += torch.count_nonzero(latents < minval)
-            latents[latents < minval] = torch.rand_like(latents[latents < minval]) * minval
-
-        if self.debug_thresholding:
-            logger.debug(f"min,     , max = {minval:.3f},        , {maxval:.3f}\t(scaled by {scale})")
-            logger.debug(f"{num_altered / latents.numel() * 100:.2f}% values altered")
-
-        return latents
-
     def apply_symmetry(
         self,
         postprocessing_settings: PostprocessingSettings,
@@ -539,18 +613,6 @@ class InvokeAIDiffuserComponent:
         self.last_percent_through = percent_through
         return latents.to(device=dev)
 
-    def estimate_percent_through(self, step_index, sigma):
-        if step_index is not None and self.cross_attention_control_context is not None:
-            # percent_through will never reach 1.0 (but this is intended)
-            return float(step_index) / float(self.cross_attention_control_context.step_count)
-        # find the best possible index of the current sigma in the sigma sequence
-        smaller_sigmas = torch.nonzero(self.model.sigmas <= sigma)
-        sigma_index = smaller_sigmas[-1].item() if smaller_sigmas.shape[0] > 0 else 0
-        # flip because sigmas[0] is for the fully denoised image
-        # percent_through must be <1
-        return 1.0 - float(sigma_index + 1) / float(self.model.sigmas.shape[0])
-        # print('estimated percent_through', percent_through, 'from sigma', sigma.item())
-
     # todo: make this work
     @classmethod
     def apply_conjunction(cls, x, t, forward_func, uc, c_or_weighted_c_list, global_guidance_scale):
@@ -564,7 +626,7 @@ class InvokeAIDiffuserComponent:
         # below is fugly omg
         conditionings = [uc] + [c for c, weight in weighted_cond_list]
         weights = [1] + [weight for c, weight in weighted_cond_list]
-        chunk_count = ceil(len(conditionings) / 2)
+        chunk_count = math.ceil(len(conditionings) / 2)
         deltas = None
         for chunk_index in range(chunk_count):
             offset = chunk_index * 2

invokeai/backend/util/hotfixes.py

@@ -4,8 +4,15 @@ import torch
 from torch import nn
 
 from diffusers.configuration_utils import ConfigMixin, register_to_config
+from diffusers.loaders import FromOriginalControlnetMixin
 from diffusers.models.attention_processor import AttentionProcessor, AttnProcessor
-from diffusers.models.embeddings import TimestepEmbedding, Timesteps
+from diffusers.models.embeddings import (
+    TextImageProjection,
+    TextImageTimeEmbedding,
+    TextTimeEmbedding,
+    TimestepEmbedding,
+    Timesteps,
+)
 from diffusers.models.modeling_utils import ModelMixin
 from diffusers.models.unet_2d_blocks import (
     CrossAttnDownBlock2D,
@@ -18,10 +25,11 @@ from diffusers.models.unet_2d_condition import UNet2DConditionModel
 import diffusers
 from diffusers.models.controlnet import ControlNetConditioningEmbedding, ControlNetOutput, zero_module
 
+# TODO: create PR to diffusers
 # Modified ControlNetModel with encoder_attention_mask argument added
 
 
-class ControlNetModel(ModelMixin, ConfigMixin):
+class ControlNetModel(ModelMixin, ConfigMixin, FromOriginalControlnetMixin):
     """
     A ControlNet model.
 
@@ -52,12 +60,25 @@ class ControlNetModel(ModelMixin, ConfigMixin):
             The epsilon to use for the normalization.
         cross_attention_dim (`int`, defaults to 1280):
             The dimension of the cross attention features.
+        transformer_layers_per_block (`int` or `Tuple[int]`, *optional*, defaults to 1):
+            The number of transformer blocks of type [`~models.attention.BasicTransformerBlock`]. Only relevant for
+            [`~models.unet_2d_blocks.CrossAttnDownBlock2D`], [`~models.unet_2d_blocks.CrossAttnUpBlock2D`],
+            [`~models.unet_2d_blocks.UNetMidBlock2DCrossAttn`].
+        encoder_hid_dim (`int`, *optional*, defaults to None):
+            If `encoder_hid_dim_type` is defined, `encoder_hidden_states` will be projected from `encoder_hid_dim`
+            dimension to `cross_attention_dim`.
+        encoder_hid_dim_type (`str`, *optional*, defaults to `None`):
+            If given, the `encoder_hidden_states` and potentially other embeddings are down-projected to text
+            embeddings of dimension `cross_attention` according to `encoder_hid_dim_type`.
         attention_head_dim (`Union[int, Tuple[int]]`, defaults to 8):
             The dimension of the attention heads.
         use_linear_projection (`bool`, defaults to `False`):
         class_embed_type (`str`, *optional*, defaults to `None`):
             The type of class embedding to use which is ultimately summed with the time embeddings. Choose from None,
             `"timestep"`, `"identity"`, `"projection"`, or `"simple_projection"`.
+        addition_embed_type (`str`, *optional*, defaults to `None`):
+            Configures an optional embedding which will be summed with the time embeddings. Choose from `None` or
+            "text". "text" will use the `TextTimeEmbedding` layer.
         num_class_embeds (`int`, *optional*, defaults to 0):
             Input dimension of the learnable embedding matrix to be projected to `time_embed_dim`, when performing
             class conditioning with `class_embed_type` equal to `None`.
@@ -98,10 +119,15 @@ class ControlNetModel(ModelMixin, ConfigMixin):
         norm_num_groups: Optional[int] = 32,
         norm_eps: float = 1e-5,
         cross_attention_dim: int = 1280,
+        transformer_layers_per_block: Union[int, Tuple[int]] = 1,
+        encoder_hid_dim: Optional[int] = None,
+        encoder_hid_dim_type: Optional[str] = None,
         attention_head_dim: Union[int, Tuple[int]] = 8,
         num_attention_heads: Optional[Union[int, Tuple[int]]] = None,
         use_linear_projection: bool = False,
         class_embed_type: Optional[str] = None,
+        addition_embed_type: Optional[str] = None,
+        addition_time_embed_dim: Optional[int] = None,
         num_class_embeds: Optional[int] = None,
         upcast_attention: bool = False,
         resnet_time_scale_shift: str = "default",
@@ -109,6 +135,7 @@ class ControlNetModel(ModelMixin, ConfigMixin):
         controlnet_conditioning_channel_order: str = "rgb",
         conditioning_embedding_out_channels: Optional[Tuple[int]] = (16, 32, 96, 256),
         global_pool_conditions: bool = False,
+        addition_embed_type_num_heads=64,
     ):
         super().__init__()
 
@@ -136,6 +163,9 @@ class ControlNetModel(ModelMixin, ConfigMixin):
                 f"Must provide the same number of `num_attention_heads` as `down_block_types`. `num_attention_heads`: {num_attention_heads}. `down_block_types`: {down_block_types}."
             )
 
+        if isinstance(transformer_layers_per_block, int):
+            transformer_layers_per_block = [transformer_layers_per_block] * len(down_block_types)
+
         # input
         conv_in_kernel = 3
         conv_in_padding = (conv_in_kernel - 1) // 2
@@ -145,16 +175,43 @@ class ControlNetModel(ModelMixin, ConfigMixin):
 
         # time
         time_embed_dim = block_out_channels[0] * 4
-
         self.time_proj = Timesteps(block_out_channels[0], flip_sin_to_cos, freq_shift)
         timestep_input_dim = block_out_channels[0]
-
         self.time_embedding = TimestepEmbedding(
             timestep_input_dim,
             time_embed_dim,
             act_fn=act_fn,
         )
 
+        if encoder_hid_dim_type is None and encoder_hid_dim is not None:
+            encoder_hid_dim_type = "text_proj"
+            self.register_to_config(encoder_hid_dim_type=encoder_hid_dim_type)
+            logger.info("encoder_hid_dim_type defaults to 'text_proj' as `encoder_hid_dim` is defined.")
+
+        if encoder_hid_dim is None and encoder_hid_dim_type is not None:
+            raise ValueError(
+                f"`encoder_hid_dim` has to be defined when `encoder_hid_dim_type` is set to {encoder_hid_dim_type}."
+            )
+
+        if encoder_hid_dim_type == "text_proj":
+            self.encoder_hid_proj = nn.Linear(encoder_hid_dim, cross_attention_dim)
+        elif encoder_hid_dim_type == "text_image_proj":
+            # image_embed_dim DOESN'T have to be `cross_attention_dim`. To not clutter the __init__ too much
+            # they are set to `cross_attention_dim` here as this is exactly the required dimension for the currently only use
+            # case when `addition_embed_type == "text_image_proj"` (Kadinsky 2.1)`
+            self.encoder_hid_proj = TextImageProjection(
+                text_embed_dim=encoder_hid_dim,
+                image_embed_dim=cross_attention_dim,
+                cross_attention_dim=cross_attention_dim,
+            )
+
+        elif encoder_hid_dim_type is not None:
+            raise ValueError(
+                f"encoder_hid_dim_type: {encoder_hid_dim_type} must be None, 'text_proj' or 'text_image_proj'."
+            )
+        else:
+            self.encoder_hid_proj = None
+
         # class embedding
         if class_embed_type is None and num_class_embeds is not None:
             self.class_embedding = nn.Embedding(num_class_embeds, time_embed_dim)
@@ -178,6 +235,29 @@ class ControlNetModel(ModelMixin, ConfigMixin):
         else:
             self.class_embedding = None
 
+        if addition_embed_type == "text":
+            if encoder_hid_dim is not None:
+                text_time_embedding_from_dim = encoder_hid_dim
+            else:
+                text_time_embedding_from_dim = cross_attention_dim
+
+            self.add_embedding = TextTimeEmbedding(
+                text_time_embedding_from_dim, time_embed_dim, num_heads=addition_embed_type_num_heads
+            )
+        elif addition_embed_type == "text_image":
+            # text_embed_dim and image_embed_dim DON'T have to be `cross_attention_dim`. To not clutter the __init__ too much
+            # they are set to `cross_attention_dim` here as this is exactly the required dimension for the currently only use
+            # case when `addition_embed_type == "text_image"` (Kadinsky 2.1)`
+            self.add_embedding = TextImageTimeEmbedding(
+                text_embed_dim=cross_attention_dim, image_embed_dim=cross_attention_dim, time_embed_dim=time_embed_dim
+            )
+        elif addition_embed_type == "text_time":
+            self.add_time_proj = Timesteps(addition_time_embed_dim, flip_sin_to_cos, freq_shift)
+            self.add_embedding = TimestepEmbedding(projection_class_embeddings_input_dim, time_embed_dim)
+
+        elif addition_embed_type is not None:
+            raise ValueError(f"addition_embed_type: {addition_embed_type} must be None, 'text' or 'text_image'.")
+
         # control net conditioning embedding
         self.controlnet_cond_embedding = ControlNetConditioningEmbedding(
             conditioning_embedding_channels=block_out_channels[0],
@@ -212,6 +292,7 @@ class ControlNetModel(ModelMixin, ConfigMixin):
             down_block = get_down_block(
                 down_block_type,
                 num_layers=layers_per_block,
+                transformer_layers_per_block=transformer_layers_per_block[i],
                 in_channels=input_channel,
                 out_channels=output_channel,
                 temb_channels=time_embed_dim,
@@ -248,6 +329,7 @@ class ControlNetModel(ModelMixin, ConfigMixin):
         self.controlnet_mid_block = controlnet_block
 
         self.mid_block = UNetMidBlock2DCrossAttn(
+            transformer_layers_per_block=transformer_layers_per_block[-1],
             in_channels=mid_block_channel,
             temb_channels=time_embed_dim,
             resnet_eps=norm_eps,
@@ -277,7 +359,22 @@ class ControlNetModel(ModelMixin, ConfigMixin):
                 The UNet model weights to copy to the [`ControlNetModel`]. All configuration options are also copied
                 where applicable.
         """
+        transformer_layers_per_block = (
+            unet.config.transformer_layers_per_block if "transformer_layers_per_block" in unet.config else 1
+        )
+        encoder_hid_dim = unet.config.encoder_hid_dim if "encoder_hid_dim" in unet.config else None
+        encoder_hid_dim_type = unet.config.encoder_hid_dim_type if "encoder_hid_dim_type" in unet.config else None
+        addition_embed_type = unet.config.addition_embed_type if "addition_embed_type" in unet.config else None
+        addition_time_embed_dim = (
+            unet.config.addition_time_embed_dim if "addition_time_embed_dim" in unet.config else None
+        )
+
         controlnet = cls(
+            encoder_hid_dim=encoder_hid_dim,
+            encoder_hid_dim_type=encoder_hid_dim_type,
+            addition_embed_type=addition_embed_type,
+            addition_time_embed_dim=addition_time_embed_dim,
+            transformer_layers_per_block=transformer_layers_per_block,
             in_channels=unet.config.in_channels,
             flip_sin_to_cos=unet.config.flip_sin_to_cos,
             freq_shift=unet.config.freq_shift,
@@ -463,6 +560,7 @@ class ControlNetModel(ModelMixin, ConfigMixin):
         class_labels: Optional[torch.Tensor] = None,
         timestep_cond: Optional[torch.Tensor] = None,
         attention_mask: Optional[torch.Tensor] = None,
+        added_cond_kwargs: Optional[Dict[str, torch.Tensor]] = None,
         cross_attention_kwargs: Optional[Dict[str, Any]] = None,
         encoder_attention_mask: Optional[torch.Tensor] = None,
         guess_mode: bool = False,
@@ -486,7 +584,9 @@ class ControlNetModel(ModelMixin, ConfigMixin):
                 Optional class labels for conditioning. Their embeddings will be summed with the timestep embeddings.
             timestep_cond (`torch.Tensor`, *optional*, defaults to `None`):
             attention_mask (`torch.Tensor`, *optional*, defaults to `None`):
-            cross_attention_kwargs(`dict[str]`, *optional*, defaults to `None`):
+            added_cond_kwargs (`dict`):
+                Additional conditions for the Stable Diffusion XL UNet.
+            cross_attention_kwargs (`dict[str]`, *optional*, defaults to `None`):
                 A kwargs dictionary that if specified is passed along to the `AttnProcessor`.
             encoder_attention_mask (`torch.Tensor`):
                 A cross-attention mask of shape `(batch, sequence_length)` is applied to `encoder_hidden_states`. If
@@ -549,6 +649,7 @@ class ControlNetModel(ModelMixin, ConfigMixin):
         t_emb = t_emb.to(dtype=sample.dtype)
 
         emb = self.time_embedding(t_emb, timestep_cond)
+        aug_emb = None
 
         if self.class_embedding is not None:
             if class_labels is None:
@@ -560,11 +661,34 @@ class ControlNetModel(ModelMixin, ConfigMixin):
             class_emb = self.class_embedding(class_labels).to(dtype=self.dtype)
             emb = emb + class_emb
 
+        if "addition_embed_type" in self.config:
+            if self.config.addition_embed_type == "text":
+                aug_emb = self.add_embedding(encoder_hidden_states)
+
+            elif self.config.addition_embed_type == "text_time":
+                if "text_embeds" not in added_cond_kwargs:
+                    raise ValueError(
+                        f"{self.__class__} has the config param `addition_embed_type` set to 'text_time' which requires the keyword argument `text_embeds` to be passed in `added_cond_kwargs`"
+                    )
+                text_embeds = added_cond_kwargs.get("text_embeds")
+                if "time_ids" not in added_cond_kwargs:
+                    raise ValueError(
+                        f"{self.__class__} has the config param `addition_embed_type` set to 'text_time' which requires the keyword argument `time_ids` to be passed in `added_cond_kwargs`"
+                    )
+                time_ids = added_cond_kwargs.get("time_ids")
+                time_embeds = self.add_time_proj(time_ids.flatten())
+                time_embeds = time_embeds.reshape((text_embeds.shape[0], -1))
+
+                add_embeds = torch.concat([text_embeds, time_embeds], dim=-1)
+                add_embeds = add_embeds.to(emb.dtype)
+                aug_emb = self.add_embedding(add_embeds)
+
+        emb = emb + aug_emb if aug_emb is not None else emb
+
         # 2. pre-process
         sample = self.conv_in(sample)
 
         controlnet_cond = self.controlnet_cond_embedding(controlnet_cond)
-
         sample = sample + controlnet_cond
 
         # 3. down

invokeai/backend/web/__init__.py

@@ -1,4 +0,0 @@
-"""
-Initialization file for the web backend.
-"""
-from .invoke_ai_web_server import InvokeAIWebServer

invokeai/backend/web/modules/create_cmd_parser.py

@@ -1,56 +0,0 @@
-import argparse
-import os
-
-from ...args import PRECISION_CHOICES
-
-
-def create_cmd_parser():
-    parser = argparse.ArgumentParser(description="InvokeAI web UI")
-    parser.add_argument(
-        "--host",
-        type=str,
-        help="The host to serve on",
-        default="localhost",
-    )
-    parser.add_argument("--port", type=int, help="The port to serve on", default=9090)
-    parser.add_argument(
-        "--cors",
-        nargs="*",
-        type=str,
-        help="Additional allowed origins, comma-separated",
-    )
-    parser.add_argument(
-        "--embedding_path",
-        type=str,
-        help="Path to a pre-trained embedding manager checkpoint - can only be set on command line",
-    )
-    # TODO: Can't get flask to serve images from any dir (saving to the dir does work when specified)
-    # parser.add_argument(
-    #     "--output_dir",
-    #     default="outputs/",
-    #     type=str,
-    #     help="Directory for output images",
-    # )
-    parser.add_argument(
-        "-v",
-        "--verbose",
-        action="store_true",
-        help="Enables verbose logging",
-    )
-    parser.add_argument(
-        "--precision",
-        dest="precision",
-        type=str,
-        choices=PRECISION_CHOICES,
-        metavar="PRECISION",
-        help=f'Set model precision. Defaults to auto selected based on device. Options: {", ".join(PRECISION_CHOICES)}',
-        default="auto",
-    )
-    parser.add_argument(
-        "--free_gpu_mem",
-        dest="free_gpu_mem",
-        action="store_true",
-        help="Force free gpu memory before final decoding",
-    )
-
-    return parser

invokeai/backend/web/modules/get_canvas_generation_mode.py

@@ -1,113 +0,0 @@
-from typing import Literal, Union
-
-from PIL import Image, ImageChops
-from PIL.Image import Image as ImageType
-
-
-# https://stackoverflow.com/questions/43864101/python-pil-check-if-image-is-transparent
-def check_for_any_transparency(img: Union[ImageType, str]) -> bool:
-    if type(img) is str:
-        img = Image.open(str)
-
-    if img.info.get("transparency", None) is not None:
-        return True
-    if img.mode == "P":
-        transparent = img.info.get("transparency", -1)
-        for _, index in img.getcolors():
-            if index == transparent:
-                return True
-    elif img.mode == "RGBA":
-        extrema = img.getextrema()
-        if extrema[3][0] < 255:
-            return True
-    return False
-
-
-def get_canvas_generation_mode(
-    init_img: Union[ImageType, str], init_mask: Union[ImageType, str]
-) -> Literal["txt2img", "outpainting", "inpainting", "img2img",]:
-    if type(init_img) is str:
-        init_img = Image.open(init_img)
-
-    if type(init_mask) is str:
-        init_mask = Image.open(init_mask)
-
-    init_img = init_img.convert("RGBA")
-
-    # Get alpha from init_img
-    init_img_alpha = init_img.split()[-1]
-    init_img_alpha_mask = init_img_alpha.convert("L")
-    init_img_has_transparency = check_for_any_transparency(init_img)
-
-    if init_img_has_transparency:
-        init_img_is_fully_transparent = True if init_img_alpha_mask.getbbox() is None else False
-
-    """
-    Mask images are white in areas where no change should be made, black where changes
-    should be made.
-    """
-
-    # Fit the mask to init_img's size and convert it to greyscale
-    init_mask = init_mask.resize(init_img.size).convert("L")
-
-    """
-    PIL.Image.getbbox() returns the bounding box of non-zero areas of the image, so we first
-    invert the mask image so that masked areas are white and other areas black == zero.
-    getbbox() now tells us if the are any masked areas.
-    """
-    init_mask_bbox = ImageChops.invert(init_mask).getbbox()
-    init_mask_exists = False if init_mask_bbox is None else True
-
-    if init_img_has_transparency:
-        if init_img_is_fully_transparent:
-            return "txt2img"
-        else:
-            return "outpainting"
-    else:
-        if init_mask_exists:
-            return "inpainting"
-        else:
-            return "img2img"
-
-
-def main():
-    # Testing
-    init_img_opaque = "test_images/init-img_opaque.png"
-    init_img_partial_transparency = "test_images/init-img_partial_transparency.png"
-    init_img_full_transparency = "test_images/init-img_full_transparency.png"
-    init_mask_no_mask = "test_images/init-mask_no_mask.png"
-    init_mask_has_mask = "test_images/init-mask_has_mask.png"
-
-    print(
-        "OPAQUE IMAGE, NO MASK, expect img2img, got ",
-        get_canvas_generation_mode(init_img_opaque, init_mask_no_mask),
-    )
-
-    print(
-        "IMAGE WITH TRANSPARENCY, NO MASK, expect outpainting, got ",
-        get_canvas_generation_mode(init_img_partial_transparency, init_mask_no_mask),
-    )
-
-    print(
-        "FULLY TRANSPARENT IMAGE NO MASK, expect txt2img, got ",
-        get_canvas_generation_mode(init_img_full_transparency, init_mask_no_mask),
-    )
-
-    print(
-        "OPAQUE IMAGE, WITH MASK, expect inpainting, got ",
-        get_canvas_generation_mode(init_img_opaque, init_mask_has_mask),
-    )
-
-    print(
-        "IMAGE WITH TRANSPARENCY, WITH MASK, expect outpainting, got ",
-        get_canvas_generation_mode(init_img_partial_transparency, init_mask_has_mask),
-    )
-
-    print(
-        "FULLY TRANSPARENT IMAGE WITH MASK, expect txt2img, got ",
-        get_canvas_generation_mode(init_img_full_transparency, init_mask_has_mask),
-    )
-
-
-if __name__ == "__main__":
-    main()

invokeai/backend/web/modules/parameters.py

@@ -1,82 +0,0 @@
-import argparse
-
-from .parse_seed_weights import parse_seed_weights
-
-SAMPLER_CHOICES = [
-    "ddim",
-    "ddpm",
-    "deis",
-    "lms",
-    "lms_k",
-    "pndm",
-    "heun",
-    "heun_k",
-    "euler",
-    "euler_k",
-    "euler_a",
-    "kdpm_2",
-    "kdpm_2_a",
-    "dpmpp_2s",
-    "dpmpp_2s_k",
-    "dpmpp_2m",
-    "dpmpp_2m_k",
-    "dpmpp_2m_sde",
-    "dpmpp_2m_sde_k",
-    "dpmpp_sde",
-    "dpmpp_sde_k",
-    "unipc",
-]
-
-
-def parameters_to_command(params):
-    """
-    Converts dict of parameters into a `invoke.py` REPL command.
-    """
-
-    switches = list()
-
-    if "prompt" in params:
-        switches.append(f'"{params["prompt"]}"')
-    if "steps" in params:
-        switches.append(f'-s {params["steps"]}')
-    if "seed" in params:
-        switches.append(f'-S {params["seed"]}')
-    if "width" in params:
-        switches.append(f'-W {params["width"]}')
-    if "height" in params:
-        switches.append(f'-H {params["height"]}')
-    if "cfg_scale" in params:
-        switches.append(f'-C {params["cfg_scale"]}')
-    if "sampler_name" in params:
-        switches.append(f'-A {params["sampler_name"]}')
-    if "seamless" in params and params["seamless"] == True:
-        switches.append(f"--seamless")
-    if "hires_fix" in params and params["hires_fix"] == True:
-        switches.append(f"--hires")
-    if "init_img" in params and len(params["init_img"]) > 0:
-        switches.append(f'-I {params["init_img"]}')
-    if "init_mask" in params and len(params["init_mask"]) > 0:
-        switches.append(f'-M {params["init_mask"]}')
-    if "init_color" in params and len(params["init_color"]) > 0:
-        switches.append(f'--init_color {params["init_color"]}')
-    if "strength" in params and "init_img" in params:
-        switches.append(f'-f {params["strength"]}')
-        if "fit" in params and params["fit"] == True:
-            switches.append(f"--fit")
-    if "facetool" in params:
-        switches.append(f'-ft {params["facetool"]}')
-    if "facetool_strength" in params and params["facetool_strength"]:
-        switches.append(f'-G {params["facetool_strength"]}')
-    elif "gfpgan_strength" in params and params["gfpgan_strength"]:
-        switches.append(f'-G {params["gfpgan_strength"]}')
-    if "codeformer_fidelity" in params:
-        switches.append(f'-cf {params["codeformer_fidelity"]}')
-    if "upscale" in params and params["upscale"]:
-        switches.append(f'-U {params["upscale"][0]} {params["upscale"][1]}')
-    if "variation_amount" in params and params["variation_amount"] > 0:
-        switches.append(f'-v {params["variation_amount"]}')
-        if "with_variations" in params:
-            seed_weight_pairs = ",".join(f"{seed}:{weight}" for seed, weight in params["with_variations"])
-            switches.append(f"-V {seed_weight_pairs}")
-
-    return " ".join(switches)

invokeai/backend/web/modules/parse_seed_weights.py

@@ -1,47 +0,0 @@
-def parse_seed_weights(seed_weights):
-    """
-    Accepts seed weights as string in "12345:0.1,23456:0.2,3456:0.3" format
-    Validates them
-    If valid: returns as [[12345, 0.1], [23456, 0.2], [3456, 0.3]]
-    If invalid: returns False
-    """
-
-    # Must be a string
-    if not isinstance(seed_weights, str):
-        return False
-    # String must not be empty
-    if len(seed_weights) == 0:
-        return False
-
-    pairs = []
-
-    for pair in seed_weights.split(","):
-        split_values = pair.split(":")
-
-        # Seed and weight are required
-        if len(split_values) != 2:
-            return False
-
-        if len(split_values[0]) == 0 or len(split_values[1]) == 1:
-            return False
-
-        # Try casting the seed to int and weight to float
-        try:
-            seed = int(split_values[0])
-            weight = float(split_values[1])
-        except ValueError:
-            return False
-
-        # Seed must be 0 or above
-        if not seed >= 0:
-            return False
-
-        # Weight must be between 0 and 1
-        if not (weight >= 0 and weight <= 1):
-            return False
-
-        # This pair is valid
-        pairs.append([seed, weight])
-
-    # All pairs are valid
-    return pairs

invokeai/frontend/install/model_install.py

@@ -775,7 +775,7 @@ def main():
 
     if not config.model_conf_path.exists():
         logger.info("Your InvokeAI root directory is not set up. Calling invokeai-configure.")
-        from invokeai.frontend.install import invokeai_configure
+        from invokeai.frontend.install.invokeai_configure import invokeai_configure
 
         invokeai_configure()
         sys.exit(0)

invokeai/frontend/web/dist/assets/ThemeLocaleProvider-bc3e6f20.js

@@ -1,4 +1,4 @@
-import{B as m,g7 as Je,A as y,a5 as Ka,g8 as Xa,af as va,aj as d,g9 as b,ga as t,gb as Ya,gc as h,gd as ua,ge as Ja,gf as Qa,aL as Za,gg as et,ad as rt,gh as at}from"./index-deaa1f26.js";import{s as fa,n as o,t as tt,o as ha,p as ot,q as ma,v as ga,w as ya,x as it,y as Sa,z as pa,A as xr,B as nt,D as lt,E as st,F as xa,G as $a,H as ka,J as dt,K as _a,L as ct,M as bt,N as vt,O as ut,Q as wa,R as ft,S as ht,T as mt,U as gt,V as yt,W as St,e as pt,X as xt}from"./menu-b4489359.js";var za=String.raw,Ca=za`
+import{B as m,g7 as Je,A as y,a5 as Ka,g8 as Xa,af as va,aj as d,g9 as b,ga as t,gb as Ya,gc as h,gd as ua,ge as Ja,gf as Qa,aL as Za,gg as et,ad as rt,gh as at}from"./index-2c171c8f.js";import{s as fa,n as o,t as tt,o as ha,p as ot,q as ma,v as ga,w as ya,x as it,y as Sa,z as pa,A as xr,B as nt,D as lt,E as st,F as xa,G as $a,H as ka,J as dt,K as _a,L as ct,M as bt,N as vt,O as ut,Q as wa,R as ft,S as ht,T as mt,U as gt,V as yt,W as St,e as pt,X as xt}from"./menu-971c0572.js";var za=String.raw,Ca=za`
   :root,
   :host {
     --chakra-vh: 100vh;

invokeai/frontend/web/dist/index.html

@@ -12,7 +12,7 @@
         margin: 0;
       }
     </style>
-    <script type="module" crossorigin src="./assets/index-deaa1f26.js"></script>
+    <script type="module" crossorigin src="./assets/index-2c171c8f.js"></script>
   </head>
 
   <body dir="ltr">

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

@@ -503,6 +503,9 @@
         "hiresStrength": "High Res Strength",
         "imageFit": "Fit Initial Image To Output Size",
         "codeformerFidelity": "Fidelity",
+        "maskAdjustmentsHeader": "Mask Adjustments",
+        "maskBlur": "Mask Blur",
+        "maskBlurMethod": "Mask Blur Method",
         "seamSize": "Seam Size",
         "seamBlur": "Seam Blur",
         "seamStrength": "Seam Strength",

invokeai/frontend/web/package.json

@@ -61,6 +61,7 @@
     "@dagrejs/graphlib": "^2.1.13",
     "@dnd-kit/core": "^6.0.8",
     "@dnd-kit/modifiers": "^6.0.1",
+    "@dnd-kit/utilities": "^3.2.1",
     "@emotion/react": "^11.11.1",
     "@emotion/styled": "^11.11.0",
     "@floating-ui/react-dom": "^2.0.1",

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

@@ -503,10 +503,17 @@
         "hiresStrength": "High Res Strength",
         "imageFit": "Fit Initial Image To Output Size",
         "codeformerFidelity": "Fidelity",
+        "maskAdjustmentsHeader": "Mask Adjustments",
+        "maskBlur": "Mask Blur",
+        "maskBlurMethod": "Mask Blur Method",
+        "seamPaintingHeader": "Seam Painting",
         "seamSize": "Seam Size",
         "seamBlur": "Seam Blur",
-        "seamStrength": "Seam Strength",
         "seamSteps": "Seam Steps",
+        "seamStrength": "Seam Strength",
+        "seamThreshold": "Seam Threshold",
+        "seamLowThreshold": "Low",
+        "seamHighThreshold": "High",
         "scaleBeforeProcessing": "Scale Before Processing",
         "scaledWidth": "Scaled W",
         "scaledHeight": "Scaled H",

invokeai/frontend/web/scripts/colors.js

@@ -0,0 +1,34 @@
+export const COLORS = {
+  reset: '\x1b[0m',
+  bright: '\x1b[1m',
+  dim: '\x1b[2m',
+  underscore: '\x1b[4m',
+  blink: '\x1b[5m',
+  reverse: '\x1b[7m',
+  hidden: '\x1b[8m',
+
+  fg: {
+    black: '\x1b[30m',
+    red: '\x1b[31m',
+    green: '\x1b[32m',
+    yellow: '\x1b[33m',
+    blue: '\x1b[34m',
+    magenta: '\x1b[35m',
+    cyan: '\x1b[36m',
+    white: '\x1b[37m',
+    gray: '\x1b[90m',
+    crimson: '\x1b[38m',
+  },
+  bg: {
+    black: '\x1b[40m',
+    red: '\x1b[41m',
+    green: '\x1b[42m',
+    yellow: '\x1b[43m',
+    blue: '\x1b[44m',
+    magenta: '\x1b[45m',
+    cyan: '\x1b[46m',
+    white: '\x1b[47m',
+    gray: '\x1b[100m',
+    crimson: '\x1b[48m',
+  },
+};

invokeai/frontend/web/scripts/typegen.js

@@ -1,23 +1,83 @@
 import fs from 'node:fs';
 import openapiTS from 'openapi-typescript';
+import { COLORS } from './colors.js';
 
-const OPENAPI_URL = 'http://localhost:9090/openapi.json';
+const OPENAPI_URL = 'http://127.0.0.1:9090/openapi.json';
 const OUTPUT_FILE = 'src/services/api/schema.d.ts';
 
 async function main() {
   process.stdout.write(
-    `Generating types "${OPENAPI_URL}" --> "${OUTPUT_FILE}"...`
+    `Generating types "${OPENAPI_URL}" --> "${OUTPUT_FILE}"...\n\n`
   );
   const types = await openapiTS(OPENAPI_URL, {
     exportType: true,
-    transform: (schemaObject) => {
+    transform: (schemaObject, metadata) => {
       if ('format' in schemaObject && schemaObject.format === 'binary') {
         return schemaObject.nullable ? 'Blob | null' : 'Blob';
       }
+
+      /**
+       * Because invocations may have required fields that accept connection input, the generated
+       * types may be incorrect.
+       *
+       * For example, the ImageResizeInvocation has a required `image` field, but because it accepts
+       * connection input, it should be optional on instantiation of the field.
+       *
+       * To handle this, the schema exposes an `input` property that can be used to determine if the
+       * field accepts connection input. If it does, we can make the field optional.
+       */
+
+      // Check if we are generating types for an invocation
+      const isInvocationPath = metadata.path.match(
+        /^#\/components\/schemas\/\w*Invocation$/
+      );
+
+      const hasInvocationProperties =
+        schemaObject.properties &&
+        ['id', 'is_intermediate', 'type'].every(
+          (prop) => prop in schemaObject.properties
+        );
+
+      if (isInvocationPath && hasInvocationProperties) {
+        // We only want to make fields optional if they are required
+        if (!Array.isArray(schemaObject?.required)) {
+          schemaObject.required = ['id', 'type'];
+          return;
+        }
+
+        schemaObject.required.forEach((prop) => {
+          const acceptsConnection = ['any', 'connection'].includes(
+            schemaObject.properties?.[prop]?.['input']
+          );
+
+          if (acceptsConnection) {
+            // remove this prop from the required array
+            const invocationName = metadata.path.split('/').pop();
+            console.log(
+              `Making connectable field optional: ${COLORS.fg.green}${invocationName}.${COLORS.fg.cyan}${prop}${COLORS.reset}`
+            );
+            schemaObject.required = schemaObject.required.filter(
+              (r) => r !== prop
+            );
+          }
+        });
+
+        schemaObject.required = [
+          ...new Set(schemaObject.required.concat(['id', 'type'])),
+        ];
+
+        return;
+      }
+      // if (
+      //   'input' in schemaObject &&
+      //   (schemaObject.input === 'any' || schemaObject.input === 'connection')
+      // ) {
+      //   schemaObject.required = false;
+      // }
     },
   });
   fs.writeFileSync(OUTPUT_FILE, types);
-  process.stdout.write(` OK!\r\n`);
+  process.stdout.write(`\nOK!\r\n`);
 }
 
 main();

invokeai/frontend/web/src/app/components/GlobalHotkeys.ts

@@ -1,8 +1,12 @@
 import { createSelector } from '@reduxjs/toolkit';
-import { RootState } from 'app/store/store';
+import { stateSelector } from 'app/store/store';
 import { useAppDispatch, useAppSelector } from 'app/store/storeHooks';
 import { requestCanvasRescale } from 'features/canvas/store/thunks/requestCanvasScale';
-import { shiftKeyPressed } from 'features/ui/store/hotkeysSlice';
+import {
+  ctrlKeyPressed,
+  metaKeyPressed,
+  shiftKeyPressed,
+} from 'features/ui/store/hotkeysSlice';
 import { activeTabNameSelector } from 'features/ui/store/uiSelectors';
 import {
   setActiveTab,
@@ -16,11 +20,11 @@ import React, { memo } from 'react';
 import { isHotkeyPressed, useHotkeys } from 'react-hotkeys-hook';
 
 const globalHotkeysSelector = createSelector(
-  [(state: RootState) => state.hotkeys, (state: RootState) => state.ui],
-  (hotkeys, ui) => {
-    const { shift } = hotkeys;
+  [stateSelector],
+  ({ hotkeys, ui }) => {
+    const { shift, ctrl, meta } = hotkeys;
     const { shouldPinParametersPanel, shouldPinGallery } = ui;
-    return { shift, shouldPinGallery, shouldPinParametersPanel };
+    return { shift, ctrl, meta, shouldPinGallery, shouldPinParametersPanel };
   },
   {
     memoizeOptions: {
@@ -37,9 +41,8 @@ const globalHotkeysSelector = createSelector(
  */
 const GlobalHotkeys: React.FC = () => {
   const dispatch = useAppDispatch();
-  const { shift, shouldPinParametersPanel, shouldPinGallery } = useAppSelector(
-    globalHotkeysSelector
-  );
+  const { shift, ctrl, meta, shouldPinParametersPanel, shouldPinGallery } =
+    useAppSelector(globalHotkeysSelector);
   const activeTabName = useAppSelector(activeTabNameSelector);
 
   useHotkeys(
@@ -50,9 +53,19 @@ const GlobalHotkeys: React.FC = () => {
       } else {
         shift && dispatch(shiftKeyPressed(false));
       }
+      if (isHotkeyPressed('ctrl')) {
+        !ctrl && dispatch(ctrlKeyPressed(true));
+      } else {
+        ctrl && dispatch(ctrlKeyPressed(false));
+      }
+      if (isHotkeyPressed('meta')) {
+        !meta && dispatch(metaKeyPressed(true));
+      } else {
+        meta && dispatch(metaKeyPressed(false));
+      }
     },
     { keyup: true, keydown: true },
-    [shift]
+    [shift, ctrl, meta]
   );
 
   useHotkeys('o', () => {

invokeai/frontend/web/src/app/components/InvokeAIUI.tsx

@@ -14,7 +14,7 @@ import { $authToken, $baseUrl, $projectId } from 'services/api/client';
 import { socketMiddleware } from 'services/events/middleware';
 import Loading from '../../common/components/Loading/Loading';
 import '../../i18n';
-import ImageDndContext from './ImageDnd/ImageDndContext';
+import AppDndContext from '../../features/dnd/components/AppDndContext';
 
 const App = lazy(() => import('./App'));
 const ThemeLocaleProvider = lazy(() => import('./ThemeLocaleProvider'));
@@ -80,9 +80,9 @@ const InvokeAIUI = ({
       <Provider store={store}>
         <React.Suspense fallback={<Loading />}>
           <ThemeLocaleProvider>
-            <ImageDndContext>
+            <AppDndContext>
               <App config={config} headerComponent={headerComponent} />
-            </ImageDndContext>
+            </AppDndContext>
           </ThemeLocaleProvider>
         </React.Suspense>
       </Provider>

invokeai/frontend/web/src/app/logging/logger.ts

@@ -19,7 +19,8 @@ type LoggerNamespace =
   | 'nodes'
   | 'system'
   | 'socketio'
-  | 'session';
+  | 'session'
+  | 'dnd';
 
 export const logger = (namespace: LoggerNamespace) =>
   $logger.get().child({ namespace });

invokeai/frontend/web/src/app/store/middleware/devtools/actionsDenylist.ts

@@ -15,7 +15,7 @@ export const actionsDenylist = [
   'socket/socketGeneratorProgress',
   'socket/appSocketGeneratorProgress',
   // every time user presses shift
-  'hotkeys/shiftKeyPressed',
+  // 'hotkeys/shiftKeyPressed',
   // this happens after every state change
   '@@REMEMBER_PERSISTED',
 ];

invokeai/frontend/web/src/app/store/middleware/listenerMiddleware/index.ts

@@ -15,6 +15,7 @@ import { addDeleteBoardAndImagesFulfilledListener } from './listeners/boardAndIm
 import { addBoardIdSelectedListener } from './listeners/boardIdSelected';
 import { addCanvasCopiedToClipboardListener } from './listeners/canvasCopiedToClipboard';
 import { addCanvasDownloadedAsImageListener } from './listeners/canvasDownloadedAsImage';
+import { addCanvasMaskSavedToGalleryListener } from './listeners/canvasMaskSavedToGallery';
 import { addCanvasMergedListener } from './listeners/canvasMerged';
 import { addCanvasSavedToGalleryListener } from './listeners/canvasSavedToGallery';
 import { addControlNetAutoProcessListener } from './listeners/controlNetAutoProcess';
@@ -27,8 +28,8 @@ import {
   addImageDeletedFulfilledListener,
   addImageDeletedPendingListener,
   addImageDeletedRejectedListener,
-  addRequestedSingleImageDeletionListener,
   addRequestedMultipleImageDeletionListener,
+  addRequestedSingleImageDeletionListener,
 } from './listeners/imageDeleted';
 import { addImageDroppedListener } from './listeners/imageDropped';
 import {
@@ -79,6 +80,8 @@ import { addUserInvokedCanvasListener } from './listeners/userInvokedCanvas';
 import { addUserInvokedImageToImageListener } from './listeners/userInvokedImageToImage';
 import { addUserInvokedNodesListener } from './listeners/userInvokedNodes';
 import { addUserInvokedTextToImageListener } from './listeners/userInvokedTextToImage';
+import { addImagesStarredListener } from './listeners/imagesStarred';
+import { addImagesUnstarredListener } from './listeners/imagesUnstarred';
 
 export const listenerMiddleware = createListenerMiddleware();
 
@@ -120,6 +123,10 @@ addImageDeletedRejectedListener();
 addDeleteBoardAndImagesFulfilledListener();
 addImageToDeleteSelectedListener();
 
+// Image starred
+addImagesStarredListener();
+addImagesUnstarredListener();
+
 // User Invoked
 addUserInvokedCanvasListener();
 addUserInvokedNodesListener();
@@ -129,6 +136,7 @@ addSessionReadyToInvokeListener();
 
 // Canvas actions
 addCanvasSavedToGalleryListener();
+addCanvasMaskSavedToGalleryListener();
 addCanvasDownloadedAsImageListener();
 addCanvasCopiedToClipboardListener();
 addCanvasMergedListener();

invokeai/frontend/web/src/app/store/middleware/listenerMiddleware/listeners/canvasMaskSavedToGallery.ts

@@ -0,0 +1,60 @@
+import { logger } from 'app/logging/logger';
+import { canvasMaskSavedToGallery } from 'features/canvas/store/actions';
+import { getCanvasData } from 'features/canvas/util/getCanvasData';
+import { addToast } from 'features/system/store/systemSlice';
+import { imagesApi } from 'services/api/endpoints/images';
+import { startAppListening } from '..';
+
+export const addCanvasMaskSavedToGalleryListener = () => {
+  startAppListening({
+    actionCreator: canvasMaskSavedToGallery,
+    effect: async (action, { dispatch, getState }) => {
+      const log = logger('canvas');
+      const state = getState();
+
+      const canvasBlobsAndImageData = await getCanvasData(
+        state.canvas.layerState,
+        state.canvas.boundingBoxCoordinates,
+        state.canvas.boundingBoxDimensions,
+        state.canvas.isMaskEnabled,
+        state.canvas.shouldPreserveMaskedArea
+      );
+
+      if (!canvasBlobsAndImageData) {
+        return;
+      }
+
+      const { maskBlob } = canvasBlobsAndImageData;
+
+      if (!maskBlob) {
+        log.error('Problem getting mask layer blob');
+        dispatch(
+          addToast({
+            title: 'Problem Saving Mask',
+            description: 'Unable to export mask',
+            status: 'error',
+          })
+        );
+        return;
+      }
+
+      const { autoAddBoardId } = state.gallery;
+
+      dispatch(
+        imagesApi.endpoints.uploadImage.initiate({
+          file: new File([maskBlob], 'canvasMaskImage.png', {
+            type: 'image/png',
+          }),
+          image_category: 'mask',
+          is_intermediate: false,
+          board_id: autoAddBoardId === 'none' ? undefined : autoAddBoardId,
+          crop_visible: true,
+          postUploadAction: {
+            type: 'TOAST',
+            toastOptions: { title: 'Mask Saved to Assets' },
+          },
+        })
+      );
+    },
+  });
+};

invokeai/frontend/web/src/app/store/middleware/listenerMiddleware/listeners/imageDeleted.ts

@@ -121,7 +121,7 @@ export const addRequestedMultipleImageDeletionListener = () => {
     effect: async (action, { dispatch, getState }) => {
       const { imageDTOs, imagesUsage } = action.payload;
 
-      if (imageDTOs.length < 1 || imagesUsage.length < 1) {
+      if (imageDTOs.length <= 1 || imagesUsage.length <= 1) {
         // handle singles in separate listener
         return;
       }

invokeai/frontend/web/src/app/store/middleware/listenerMiddleware/listeners/imageDropped.ts

@@ -1,16 +1,20 @@
 import { createAction } from '@reduxjs/toolkit';
-import {
-  TypesafeDraggableData,
-  TypesafeDroppableData,
-} from 'app/components/ImageDnd/typesafeDnd';
 import { logger } from 'app/logging/logger';
 import { setInitialCanvasImage } from 'features/canvas/store/canvasSlice';
 import { controlNetImageChanged } from 'features/controlNet/store/controlNetSlice';
+import {
+  TypesafeDraggableData,
+  TypesafeDroppableData,
+} from 'features/dnd/types';
 import { imageSelected } from 'features/gallery/store/gallerySlice';
-import { fieldValueChanged } from 'features/nodes/store/nodesSlice';
+import {
+  fieldImageValueChanged,
+  workflowExposedFieldAdded,
+} from 'features/nodes/store/nodesSlice';
 import { initialImageChanged } from 'features/parameters/store/generationSlice';
 import { imagesApi } from 'services/api/endpoints/images';
 import { startAppListening } from '../';
+import { parseify } from 'common/util/serialize';
 
 export const dndDropped = createAction<{
   overData: TypesafeDroppableData;
@@ -21,7 +25,7 @@ export const addImageDroppedListener = () => {
   startAppListening({
     actionCreator: dndDropped,
     effect: async (action, { dispatch }) => {
-      const log = logger('images');
+      const log = logger('dnd');
       const { activeData, overData } = action.payload;
 
       if (activeData.payloadType === 'IMAGE_DTO') {
@@ -31,10 +35,28 @@ export const addImageDroppedListener = () => {
           { activeData, overData },
           `Images (${activeData.payload.imageDTOs.length}) dropped`
         );
+      } else if (activeData.payloadType === 'NODE_FIELD') {
+        log.debug(
+          { activeData: parseify(activeData), overData: parseify(overData) },
+          'Node field dropped'
+        );
       } else {
         log.debug({ activeData, overData }, `Unknown payload dropped`);
       }
 
+      if (
+        overData.actionType === 'ADD_FIELD_TO_LINEAR' &&
+        activeData.payloadType === 'NODE_FIELD'
+      ) {
+        const { nodeId, field } = activeData.payload;
+        dispatch(
+          workflowExposedFieldAdded({
+            nodeId,
+            fieldName: field.name,
+          })
+        );
+      }
+
       /**
        * Image dropped on current image
        */
@@ -99,7 +121,7 @@ export const addImageDroppedListener = () => {
       ) {
         const { fieldName, nodeId } = overData.context;
         dispatch(
-          fieldValueChanged({
+          fieldImageValueChanged({
             nodeId,
             fieldName,
             value: activeData.payload.imageDTO,

invokeai/frontend/web/src/app/store/middleware/listenerMiddleware/listeners/imageUploaded.ts

@@ -2,7 +2,7 @@ import { UseToastOptions } from '@chakra-ui/react';
 import { logger } from 'app/logging/logger';
 import { setInitialCanvasImage } from 'features/canvas/store/canvasSlice';
 import { controlNetImageChanged } from 'features/controlNet/store/controlNetSlice';
-import { fieldValueChanged } from 'features/nodes/store/nodesSlice';
+import { fieldImageValueChanged } from 'features/nodes/store/nodesSlice';
 import { initialImageChanged } from 'features/parameters/store/generationSlice';
 import { addToast } from 'features/system/store/systemSlice';
 import { omit } from 'lodash-es';
@@ -111,7 +111,9 @@ export const addImageUploadedFulfilledListener = () => {
 
       if (postUploadAction?.type === 'SET_NODES_IMAGE') {
         const { nodeId, fieldName } = postUploadAction;
-        dispatch(fieldValueChanged({ nodeId, fieldName, value: imageDTO }));
+        dispatch(
+          fieldImageValueChanged({ nodeId, fieldName, value: imageDTO })
+        );
         dispatch(
           addToast({
             ...DEFAULT_UPLOADED_TOAST,

invokeai/frontend/web/src/app/store/middleware/listenerMiddleware/listeners/imagesStarred.ts

@@ -0,0 +1,30 @@
+import { imagesApi } from 'services/api/endpoints/images';
+import { startAppListening } from '..';
+import { selectionChanged } from '../../../../../features/gallery/store/gallerySlice';
+import { ImageDTO } from '../../../../../services/api/types';
+
+export const addImagesStarredListener = () => {
+  startAppListening({
+    matcher: imagesApi.endpoints.starImages.matchFulfilled,
+    effect: async (action, { dispatch, getState }) => {
+      const { updated_image_names: starredImages } = action.payload;
+
+      const state = getState();
+
+      const { selection } = state.gallery;
+      const updatedSelection: ImageDTO[] = [];
+
+      selection.forEach((selectedImageDTO) => {
+        if (starredImages.includes(selectedImageDTO.image_name)) {
+          updatedSelection.push({
+            ...selectedImageDTO,
+            starred: true,
+          });
+        } else {
+          updatedSelection.push(selectedImageDTO);
+        }
+      });
+      dispatch(selectionChanged(updatedSelection));
+    },
+  });
+};

invokeai/frontend/web/src/app/store/middleware/listenerMiddleware/listeners/imagesUnstarred.ts

@@ -0,0 +1,30 @@
+import { imagesApi } from 'services/api/endpoints/images';
+import { startAppListening } from '..';
+import { selectionChanged } from '../../../../../features/gallery/store/gallerySlice';
+import { ImageDTO } from '../../../../../services/api/types';
+
+export const addImagesUnstarredListener = () => {
+  startAppListening({
+    matcher: imagesApi.endpoints.unstarImages.matchFulfilled,
+    effect: async (action, { dispatch, getState }) => {
+      const { updated_image_names: unstarredImages } = action.payload;
+
+      const state = getState();
+
+      const { selection } = state.gallery;
+      const updatedSelection: ImageDTO[] = [];
+
+      selection.forEach((selectedImageDTO) => {
+        if (unstarredImages.includes(selectedImageDTO.image_name)) {
+          updatedSelection.push({
+            ...selectedImageDTO,
+            starred: false,
+          });
+        } else {
+          updatedSelection.push(selectedImageDTO);
+        }
+      });
+      dispatch(selectionChanged(updatedSelection));
+    },
+  });
+};

invokeai/frontend/web/src/app/store/middleware/listenerMiddleware/listeners/modelsLoaded.ts

@@ -15,12 +15,21 @@ import {
   setShouldUseSDXLRefiner,
 } from 'features/sdxl/store/sdxlSlice';
 import { forEach, some } from 'lodash-es';
-import { modelsApi, vaeModelsAdapter } from 'services/api/endpoints/models';
+import {
+  mainModelsAdapter,
+  modelsApi,
+  vaeModelsAdapter,
+} from 'services/api/endpoints/models';
+import { TypeGuardFor } from 'services/api/types';
 import { startAppListening } from '..';
 
 export const addModelsLoadedListener = () => {
   startAppListening({
-    predicate: (state, action) =>
+    predicate: (
+      action
+    ): action is TypeGuardFor<
+      typeof modelsApi.endpoints.getMainModels.matchFulfilled
+    > =>
       modelsApi.endpoints.getMainModels.matchFulfilled(action) &&
       !action.meta.arg.originalArgs.includes('sdxl-refiner'),
     effect: async (action, { getState, dispatch }) => {
@@ -32,29 +41,28 @@ export const addModelsLoadedListener = () => {
       );
 
       const currentModel = getState().generation.model;
+      const models = mainModelsAdapter.getSelectors().selectAll(action.payload);
 
-      const isCurrentModelAvailable = some(
-        action.payload.entities,
-        (m) =>
-          m?.model_name === currentModel?.model_name &&
-          m?.base_model === currentModel?.base_model &&
-          m?.model_type === currentModel?.model_type
-      );
-
-      if (isCurrentModelAvailable) {
-        return;
-      }
-
-      const firstModelId = action.payload.ids[0];
-      const firstModel = action.payload.entities[firstModelId];
-
-      if (!firstModel) {
+      if (models.length === 0) {
         // No models loaded at all
         dispatch(modelChanged(null));
         return;
       }
 
-      const result = zMainOrOnnxModel.safeParse(firstModel);
+      const isCurrentModelAvailable = currentModel
+        ? models.some(
+            (m) =>
+              m.model_name === currentModel.model_name &&
+              m.base_model === currentModel.base_model &&
+              m.model_type === currentModel.model_type
+          )
+        : false;
+
+      if (isCurrentModelAvailable) {
+        return;
+      }
+
+      const result = zMainOrOnnxModel.safeParse(models[0]);
 
       if (!result.success) {
         log.error(
@@ -68,7 +76,11 @@ export const addModelsLoadedListener = () => {
     },
   });
   startAppListening({
-    predicate: (state, action) =>
+    predicate: (
+      action
+    ): action is TypeGuardFor<
+      typeof modelsApi.endpoints.getMainModels.matchFulfilled
+    > =>
       modelsApi.endpoints.getMainModels.matchFulfilled(action) &&
       action.meta.arg.originalArgs.includes('sdxl-refiner'),
     effect: async (action, { getState, dispatch }) => {
@@ -80,30 +92,29 @@ export const addModelsLoadedListener = () => {
       );
 
       const currentModel = getState().sdxl.refinerModel;
+      const models = mainModelsAdapter.getSelectors().selectAll(action.payload);
 
-      const isCurrentModelAvailable = some(
-        action.payload.entities,
-        (m) =>
-          m?.model_name === currentModel?.model_name &&
-          m?.base_model === currentModel?.base_model &&
-          m?.model_type === currentModel?.model_type
-      );
-
-      if (isCurrentModelAvailable) {
-        return;
-      }
-
-      const firstModelId = action.payload.ids[0];
-      const firstModel = action.payload.entities[firstModelId];
-
-      if (!firstModel) {
+      if (models.length === 0) {
         // No models loaded at all
         dispatch(refinerModelChanged(null));
         dispatch(setShouldUseSDXLRefiner(false));
         return;
       }
 
-      const result = zSDXLRefinerModel.safeParse(firstModel);
+      const isCurrentModelAvailable = currentModel
+        ? models.some(
+            (m) =>
+              m.model_name === currentModel.model_name &&
+              m.base_model === currentModel.base_model &&
+              m.model_type === currentModel.model_type
+          )
+        : false;
+
+      if (isCurrentModelAvailable) {
+        return;
+      }
+
+      const result = zSDXLRefinerModel.safeParse(models[0]);
 
       if (!result.success) {
         log.error(

invokeai/frontend/web/src/app/store/middleware/listenerMiddleware/listeners/receivedOpenAPISchema.ts

@@ -13,7 +13,7 @@ export const addReceivedOpenAPISchemaListener = () => {
       const log = logger('system');
       const schemaJSON = action.payload;
 
-      log.debug({ schemaJSON }, 'Dereferenced OpenAPI schema');
+      log.debug({ schemaJSON }, 'Received OpenAPI schema');
 
       const nodeTemplates = parseSchema(schemaJSON);
 
@@ -28,9 +28,12 @@ export const addReceivedOpenAPISchemaListener = () => {
 
   startAppListening({
     actionCreator: receivedOpenAPISchema.rejected,
-    effect: () => {
+    effect: (action) => {
       const log = logger('system');
-      log.error('Problem dereferencing OpenAPI Schema');
+      log.error(
+        { error: parseify(action.error) },
+        'Problem retrieving OpenAPI Schema'
+      );
     },
   });
 };

invokeai/frontend/web/src/app/store/middleware/listenerMiddleware/listeners/socketio/socketInvocationComplete.ts

@@ -7,6 +7,7 @@ import {
   imageSelected,
 } from 'features/gallery/store/gallerySlice';
 import { IMAGE_CATEGORIES } from 'features/gallery/store/types';
+import { CANVAS_OUTPUT } from 'features/nodes/util/graphBuilders/constants';
 import { progressImageSet } from 'features/system/store/systemSlice';
 import { imagesApi } from 'services/api/endpoints/images';
 import { isImageOutput } from 'services/api/guards';
@@ -18,7 +19,7 @@ import {
 } from 'services/events/actions';
 import { startAppListening } from '../..';
 
-const nodeDenylist = ['dataURL_image'];
+const nodeDenylist = ['load_image'];
 
 export const addInvocationCompleteEventListener = () => {
   startAppListening({
@@ -52,7 +53,9 @@ export const addInvocationCompleteEventListener = () => {
 
         // Add canvas images to the staging area
         if (
-          graph_execution_state_id === canvas.layerState.stagingArea.sessionId
+          graph_execution_state_id ===
+            canvas.layerState.stagingArea.sessionId &&
+          [CANVAS_OUTPUT].includes(data.source_node_id)
         ) {
           dispatch(addImageToStagingArea(imageDTO));
         }

invokeai/frontend/web/src/app/store/middleware/listenerMiddleware/listeners/tabChanged.ts

@@ -12,7 +12,10 @@ export const addTabChangedListener = () => {
       if (activeTabName === 'unifiedCanvas') {
         const currentBaseModel = getState().generation.model?.base_model;
 
-        if (currentBaseModel && ['sd-1', 'sd-2'].includes(currentBaseModel)) {
+        if (
+          currentBaseModel &&
+          ['sd-1', 'sd-2', 'sdxl'].includes(currentBaseModel)
+        ) {
           // if we're already on a valid model, no change needed
           return;
         }
@@ -36,7 +39,9 @@ export const addTabChangedListener = () => {
           const validCanvasModels = mainModelsAdapter
             .getSelectors()
             .selectAll(models)
-            .filter((model) => ['sd-1', 'sd-2'].includes(model.base_model));
+            .filter((model) =>
+              ['sd-1', 'sd-2', 'sxdl'].includes(model.base_model)
+            );
 
           const firstValidCanvasModel = validCanvasModels[0];
 

invokeai/frontend/web/src/app/store/middleware/listenerMiddleware/listeners/userInvokedCanvas.ts

@@ -1,6 +1,7 @@
 import { logger } from 'app/logging/logger';
 import { userInvoked } from 'app/store/actions';
 import openBase64ImageInTab from 'common/util/openBase64ImageInTab';
+import { parseify } from 'common/util/serialize';
 import {
   canvasSessionIdChanged,
   stagingAreaInitialized,
@@ -15,7 +16,6 @@ import { imagesApi } from 'services/api/endpoints/images';
 import { sessionCreated } from 'services/api/thunks/session';
 import { ImageDTO } from 'services/api/types';
 import { startAppListening } from '..';
-import { parseify } from 'common/util/serialize';
 
 /**
  * This listener is responsible invoking the canvas. This involves a number of steps:

invokeai/frontend/web/src/app/store/middleware/listenerMiddleware/listeners/userInvokedNodes.ts

@@ -15,7 +15,7 @@ export const addUserInvokedNodesListener = () => {
       const log = logger('session');
       const state = getState();
 
-      const graph = buildNodesGraph(state);
+      const graph = buildNodesGraph(state.nodes);
       dispatch(nodesGraphBuilt(graph));
       log.debug({ graph: parseify(graph) }, 'Nodes graph built');
 

invokeai/frontend/web/src/app/types/invokeai.ts

@@ -1,86 +1,7 @@
-import {
-  // CONTROLNET_MODELS,
-  CONTROLNET_PROCESSORS,
-} from 'features/controlNet/store/constants';
+import { CONTROLNET_PROCESSORS } from 'features/controlNet/store/constants';
 import { InvokeTabName } from 'features/ui/store/tabMap';
 import { O } from 'ts-toolbelt';
 
-// These are old types from the model management UI
-
-// export type ModelStatus = 'active' | 'cached' | 'not loaded';
-
-// export type Model = {
-//   status: ModelStatus;
-//   description: string;
-//   weights: string;
-//   config?: string;
-//   vae?: string;
-//   width?: number;
-//   height?: number;
-//   default?: boolean;
-//   format?: string;
-// };
-
-// export type DiffusersModel = {
-//   status: ModelStatus;
-//   description: string;
-//   repo_id?: string;
-//   path?: string;
-//   vae?: {
-//     repo_id?: string;
-//     path?: string;
-//   };
-//   format?: string;
-//   default?: boolean;
-// };
-
-// export type ModelList = Record<string, Model & DiffusersModel>;
-
-// export type FoundModel = {
-//   name: string;
-//   location: string;
-// };
-
-// export type InvokeModelConfigProps = {
-//   name: string | undefined;
-//   description: string | undefined;
-//   config: string | undefined;
-//   weights: string | undefined;
-//   vae: string | undefined;
-//   width: number | undefined;
-//   height: number | undefined;
-//   default: boolean | undefined;
-//   format: string | undefined;
-// };
-
-// export type InvokeDiffusersModelConfigProps = {
-//   name: string | undefined;
-//   description: string | undefined;
-//   repo_id: string | undefined;
-//   path: string | undefined;
-//   default: boolean | undefined;
-//   format: string | undefined;
-//   vae: {
-//     repo_id: string | undefined;
-//     path: string | undefined;
-//   };
-// };
-
-// export type InvokeModelConversionProps = {
-//   model_name: string;
-//   save_location: string;
-//   custom_location: string | null;
-// };
-
-// export type InvokeModelMergingProps = {
-//   models_to_merge: string[];
-//   alpha: number;
-//   interp: 'weighted_sum' | 'sigmoid' | 'inv_sigmoid' | 'add_difference';
-//   force: boolean;
-//   merged_model_name: string;
-//   model_merge_save_path: string | null;
-// };
-
 /**
  * A disable-able application feature
  */

invokeai/frontend/web/src/common/components/IAIContextMenu.tsx

@@ -0,0 +1,126 @@
+/**
+ * This is a copy-paste of https://github.com/lukasbach/chakra-ui-contextmenu with a small change.
+ *
+ * The reactflow background element somehow prevents the chakra `useOutsideClick()` hook from working.
+ * With a menu open, clicking on the reactflow background element doesn't close the menu.
+ *
+ * Reactflow does provide an `onPaneClick` to handle clicks on the background element, but it is not
+ * straightforward to programatically close the menu.
+ *
+ * As a (hopefully temporary) workaround, we will use a dirty hack:
+ * - create `globalContextMenuCloseTrigger: number` in `ui` slice
+ * - increment it in `onPaneClick`
+ * - `useEffect()` to close the menu when `globalContextMenuCloseTrigger` changes
+ */
+
+import {
+  Menu,
+  MenuButton,
+  MenuButtonProps,
+  MenuProps,
+  Portal,
+  PortalProps,
+  useEventListener,
+} from '@chakra-ui/react';
+import { useAppSelector } from 'app/store/storeHooks';
+import * as React from 'react';
+import {
+  MutableRefObject,
+  useCallback,
+  useEffect,
+  useRef,
+  useState,
+} from 'react';
+
+export interface IAIContextMenuProps<T extends HTMLElement> {
+  renderMenu: () => JSX.Element | null;
+  children: (ref: MutableRefObject<T | null>) => JSX.Element | null;
+  menuProps?: Omit<MenuProps, 'children'> & { children?: React.ReactNode };
+  portalProps?: Omit<PortalProps, 'children'> & { children?: React.ReactNode };
+  menuButtonProps?: MenuButtonProps;
+}
+
+export function IAIContextMenu<T extends HTMLElement = HTMLElement>(
+  props: IAIContextMenuProps<T>
+) {
+  const [isOpen, setIsOpen] = useState(false);
+  const [isRendered, setIsRendered] = useState(false);
+  const [isDeferredOpen, setIsDeferredOpen] = useState(false);
+  const [position, setPosition] = useState<[number, number]>([0, 0]);
+  const targetRef = useRef<T>(null);
+
+  const globalContextMenuCloseTrigger = useAppSelector(
+    (state) => state.ui.globalContextMenuCloseTrigger
+  );
+
+  useEffect(() => {
+    if (isOpen) {
+      setTimeout(() => {
+        setIsRendered(true);
+        setTimeout(() => {
+          setIsDeferredOpen(true);
+        });
+      });
+    } else {
+      setIsDeferredOpen(false);
+      const timeout = setTimeout(() => {
+        setIsRendered(isOpen);
+      }, 1000);
+      return () => clearTimeout(timeout);
+    }
+  }, [isOpen]);
+
+  useEffect(() => {
+    setIsOpen(false);
+    setIsDeferredOpen(false);
+    setIsRendered(false);
+  }, [globalContextMenuCloseTrigger]);
+
+  useEventListener('contextmenu', (e) => {
+    if (
+      targetRef.current?.contains(e.target as HTMLElement) ||
+      e.target === targetRef.current
+    ) {
+      e.preventDefault();
+      setIsOpen(true);
+      setPosition([e.pageX, e.pageY]);
+    } else {
+      setIsOpen(false);
+    }
+  });
+
+  const onCloseHandler = useCallback(() => {
+    props.menuProps?.onClose?.();
+    setIsOpen(false);
+  }, [props.menuProps]);
+
+  return (
+    <>
+      {props.children(targetRef)}
+      {isRendered && (
+        <Portal {...props.portalProps}>
+          <Menu
+            isOpen={isDeferredOpen}
+            gutter={0}
+            {...props.menuProps}
+            onClose={onCloseHandler}
+          >
+            <MenuButton
+              aria-hidden={true}
+              w={1}
+              h={1}
+              style={{
+                position: 'absolute',
+                left: position[0],
+                top: position[1],
+                cursor: 'default',
+              }}
+              {...props.menuButtonProps}
+            />
+            {props.renderMenu()}
+          </Menu>
+        </Portal>
+      )}
+    </>
+  );
+}

invokeai/frontend/web/src/common/components/IAIDndImage.tsx

@@ -1,16 +1,11 @@
 import {
   ChakraProps,
   Flex,
+  FlexProps,
   Icon,
   Image,
   useColorMode,
-  useColorModeValue,
 } from '@chakra-ui/react';
-import {
-  TypesafeDraggableData,
-  TypesafeDroppableData,
-} from 'app/components/ImageDnd/typesafeDnd';
-import IAIIconButton from 'common/components/IAIIconButton';
 import {
   IAILoadingImageFallback,
   IAINoContentFallback,
@@ -21,27 +16,39 @@ import ImageContextMenu from 'features/gallery/components/ImageContextMenu/Image
 import {
   MouseEvent,
   ReactElement,
+  ReactNode,
   SyntheticEvent,
   memo,
   useCallback,
   useState,
 } from 'react';
-import { FaImage, FaUndo, FaUpload } from 'react-icons/fa';
+import { FaImage, FaUpload } from 'react-icons/fa';
 import { ImageDTO, PostUploadAction } from 'services/api/types';
 import { mode } from 'theme/util/mode';
 import IAIDraggable from './IAIDraggable';
 import IAIDroppable from './IAIDroppable';
 import SelectionOverlay from './SelectionOverlay';
+import {
+  TypesafeDraggableData,
+  TypesafeDroppableData,
+} from 'features/dnd/types';
 
-type IAIDndImageProps = {
+const defaultUploadElement = (
+  <Icon
+    as={FaUpload}
+    sx={{
+      boxSize: 16,
+    }}
+  />
+);
+
+const defaultNoContentFallback = <IAINoContentFallback icon={FaImage} />;
+
+type IAIDndImageProps = FlexProps & {
   imageDTO: ImageDTO | undefined;
   onError?: (event: SyntheticEvent<HTMLImageElement>) => void;
   onLoad?: (event: SyntheticEvent<HTMLImageElement>) => void;
   onClick?: (event: MouseEvent<HTMLDivElement>) => void;
-  onClickReset?: (event: MouseEvent<HTMLButtonElement>) => void;
-  withResetIcon?: boolean;
-  resetIcon?: ReactElement;
-  resetTooltip?: string;
   withMetadataOverlay?: boolean;
   isDragDisabled?: boolean;
   isDropDisabled?: boolean;
@@ -52,21 +59,21 @@ type IAIDndImageProps = {
   fitContainer?: boolean;
   droppableData?: TypesafeDroppableData;
   draggableData?: TypesafeDraggableData;
-  dropLabel?: string;
+  dropLabel?: ReactNode;
   isSelected?: boolean;
   thumbnail?: boolean;
   noContentFallback?: ReactElement;
   useThumbailFallback?: boolean;
   withHoverOverlay?: boolean;
+  children?: JSX.Element;
+  uploadElement?: ReactNode;
 };
 
 const IAIDndImage = (props: IAIDndImageProps) => {
   const {
     imageDTO,
-    onClickReset,
     onError,
     onClick,
-    withResetIcon = false,
     withMetadataOverlay = false,
     isDropDisabled = false,
     isDragDisabled = false,
@@ -80,32 +87,37 @@ const IAIDndImage = (props: IAIDndImageProps) => {
     dropLabel,
     isSelected = false,
     thumbnail = false,
-    resetTooltip = 'Reset',
-    resetIcon = <FaUndo />,
-    noContentFallback = <IAINoContentFallback icon={FaImage} />,
+    noContentFallback = defaultNoContentFallback,
+    uploadElement = defaultUploadElement,
     useThumbailFallback,
     withHoverOverlay = false,
+    children,
+    onMouseOver,
+    onMouseOut,
   } = props;
 
   const { colorMode } = useColorMode();
   const [isHovered, setIsHovered] = useState(false);
-  const handleMouseOver = useCallback(() => {
-    setIsHovered(true);
-  }, []);
-  const handleMouseOut = useCallback(() => {
-    setIsHovered(false);
-  }, []);
+  const handleMouseOver = useCallback(
+    (e: MouseEvent<HTMLDivElement>) => {
+      if (onMouseOver) onMouseOver(e);
+      setIsHovered(true);
+    },
+    [onMouseOver]
+  );
+  const handleMouseOut = useCallback(
+    (e: MouseEvent<HTMLDivElement>) => {
+      if (onMouseOut) onMouseOut(e);
+      setIsHovered(false);
+    },
+    [onMouseOut]
+  );
 
   const { getUploadButtonProps, getUploadInputProps } = useImageUploadButton({
     postUploadAction,
     isDisabled: isUploadDisabled,
   });
 
-  const resetIconShadow = useColorModeValue(
-    `drop-shadow(0px 0px 0.1rem var(--invokeai-colors-base-600))`,
-    `drop-shadow(0px 0px 0.1rem var(--invokeai-colors-base-800))`
-  );
-
   const uploadButtonStyles = isUploadDisabled
     ? {}
     : {
@@ -157,11 +169,10 @@ const IAIDndImage = (props: IAIDndImageProps) => {
                     <IAILoadingImageFallback image={imageDTO} />
                   )
                 }
-                width={imageDTO.width}
-                height={imageDTO.height}
                 onError={onError}
                 draggable={false}
                 sx={{
+                  w: imageDTO.width,
                   objectFit: 'contain',
                   maxW: 'full',
                   maxH: 'full',
@@ -196,12 +207,7 @@ const IAIDndImage = (props: IAIDndImageProps) => {
                 {...getUploadButtonProps()}
               >
                 <input {...getUploadInputProps()} />
-                <Icon
-                  as={FaUpload}
-                  sx={{
-                    boxSize: 16,
-                  }}
-                />
+                {uploadElement}
               </Flex>
             </>
           )}
@@ -213,6 +219,7 @@ const IAIDndImage = (props: IAIDndImageProps) => {
               onClick={onClick}
             />
           )}
+          {children}
           {!isDropDisabled && (
             <IAIDroppable
               data={droppableData}
@@ -220,30 +227,6 @@ const IAIDndImage = (props: IAIDndImageProps) => {
               dropLabel={dropLabel}
             />
           )}
-          {onClickReset && withResetIcon && imageDTO && (
-            <IAIIconButton
-              onClick={onClickReset}
-              aria-label={resetTooltip}
-              tooltip={resetTooltip}
-              icon={resetIcon}
-              size="sm"
-              variant="link"
-              sx={{
-                position: 'absolute',
-                top: 1,
-                insetInlineEnd: 1,
-                p: 0,
-                minW: 0,
-                svg: {
-                  transitionProperty: 'common',
-                  transitionDuration: 'normal',
-                  fill: 'base.100',
-                  _hover: { fill: 'base.50' },
-                  filter: resetIconShadow,
-                },
-              }}
-            />
-          )}
         </Flex>
       )}
     </ImageContextMenu>

invokeai/frontend/web/src/common/components/IAIDndImageIcon.tsx

@@ -0,0 +1,46 @@
+import { SystemStyleObject, useColorModeValue } from '@chakra-ui/react';
+import { MouseEvent, ReactElement, memo } from 'react';
+import IAIIconButton from './IAIIconButton';
+
+type Props = {
+  onClick: (event: MouseEvent<HTMLButtonElement>) => void;
+  tooltip: string;
+  icon?: ReactElement;
+  styleOverrides?: SystemStyleObject;
+};
+
+const IAIDndImageIcon = (props: Props) => {
+  const { onClick, tooltip, icon, styleOverrides } = props;
+
+  const resetIconShadow = useColorModeValue(
+    `drop-shadow(0px 0px 0.1rem var(--invokeai-colors-base-600))`,
+    `drop-shadow(0px 0px 0.1rem var(--invokeai-colors-base-800))`
+  );
+  return (
+    <IAIIconButton
+      onClick={onClick}
+      aria-label={tooltip}
+      tooltip={tooltip}
+      icon={icon}
+      size="sm"
+      variant="link"
+      sx={{
+        position: 'absolute',
+        top: 1,
+        insetInlineEnd: 1,
+        p: 0,
+        minW: 0,
+        svg: {
+          transitionProperty: 'common',
+          transitionDuration: 'normal',
+          fill: 'base.100',
+          _hover: { fill: 'base.50' },
+          filter: resetIconShadow,
+        },
+        ...styleOverrides,
+      }}
+    />
+  );
+};
+
+export default memo(IAIDndImageIcon);

invokeai/frontend/web/src/common/components/IAIDraggable.tsx

@@ -1,22 +1,19 @@
-import { Box } from '@chakra-ui/react';
-import {
-  TypesafeDraggableData,
-  useDraggable,
-} from 'app/components/ImageDnd/typesafeDnd';
-import { MouseEvent, memo, useRef } from 'react';
+import { Box, BoxProps } from '@chakra-ui/react';
+import { useDraggableTypesafe } from 'features/dnd/hooks/typesafeHooks';
+import { TypesafeDraggableData } from 'features/dnd/types';
+import { memo, useRef } from 'react';
 import { v4 as uuidv4 } from 'uuid';
 
-type IAIDraggableProps = {
+type IAIDraggableProps = BoxProps & {
   disabled?: boolean;
   data?: TypesafeDraggableData;
-  onClick?: (event: MouseEvent<HTMLDivElement>) => void;
 };
 
 const IAIDraggable = (props: IAIDraggableProps) => {
-  const { data, disabled, onClick } = props;
+  const { data, disabled, ...rest } = props;
   const dndId = useRef(uuidv4());
 
-  const { attributes, listeners, setNodeRef } = useDraggable({
+  const { attributes, listeners, setNodeRef } = useDraggableTypesafe({
     id: dndId.current,
     disabled,
     data,
@@ -24,7 +21,6 @@ const IAIDraggable = (props: IAIDraggableProps) => {
 
   return (
     <Box
-      onClick={onClick}
       ref={setNodeRef}
       position="absolute"
       w="full"
@@ -33,6 +29,7 @@ const IAIDraggable = (props: IAIDraggableProps) => {
       insetInlineStart={0}
       {...attributes}
       {...listeners}
+      {...rest}
     />
   );
 };

invokeai/frontend/web/src/common/components/IAIDroppable.tsx

@@ -1,9 +1,7 @@
 import { Box } from '@chakra-ui/react';
-import {
-  TypesafeDroppableData,
-  isValidDrop,
-  useDroppable,
-} from 'app/components/ImageDnd/typesafeDnd';
+import { useDroppableTypesafe } from 'features/dnd/hooks/typesafeHooks';
+import { TypesafeDroppableData } from 'features/dnd/types';
+import { isValidDrop } from 'features/dnd/util/isValidDrop';
 import { AnimatePresence } from 'framer-motion';
 import { ReactNode, memo, useRef } from 'react';
 import { v4 as uuidv4 } from 'uuid';
@@ -19,7 +17,7 @@ const IAIDroppable = (props: IAIDroppableProps) => {
   const { dropLabel, data, disabled } = props;
   const dndId = useRef(uuidv4());
 
-  const { isOver, setNodeRef, active } = useDroppable({
+  const { isOver, setNodeRef, active } = useDroppableTypesafe({
     id: dndId.current,
     disabled,
     data,

invokeai/frontend/web/src/common/components/IAIImageFallback.tsx

@@ -49,7 +49,7 @@ export const IAILoadingImageFallback = (props: Props) => {
 
 type IAINoImageFallbackProps = {
   label?: string;
-  icon?: As;
+  icon?: As | null;
   boxSize?: StyleProps['boxSize'];
   sx?: ChakraProps['sx'];
 };
@@ -76,7 +76,7 @@ export const IAINoContentFallback = (props: IAINoImageFallbackProps) => {
         ...props.sx,
       }}
     >
-      <Icon as={icon} boxSize={boxSize} opacity={0.7} />
+      {icon && <Icon as={icon} boxSize={boxSize} opacity={0.7} />}
       {props.label && <Text textAlign="center">{props.label}</Text>}
     </Flex>
   );