Merge e5914409d3 into 4064062e7d

* add color type and color to rgb int node

* review fix for allowing output

---------

Co-authored-by: Jedrzej Kosinski <kosinkadink1@gmail.com>

.github/workflows/windows_release_dependencies.yml

@@ -29,7 +29,7 @@ on:
         description: 'python patch version'
         required: true
         type: string
-        default: "9"
+        default: "11"
 #  push:
 #    branches:
 #      - master

comfy/model_management.py

@@ -138,6 +138,12 @@ try:
 except:
     ixuca_available = False
 
+try:
+    import torchada  # noqa: F401
+    musa_available = hasattr(torch, "musa") and torch.musa.is_available()
+except:
+    musa_available = False
+
 if args.cpu:
     cpu_state = CPUState.CPU
 
@@ -145,27 +151,24 @@ def is_intel_xpu():
     global cpu_state
     global xpu_available
     if cpu_state == CPUState.GPU:
-        if xpu_available:
-            return True
+        return xpu_available
     return False
 
 def is_ascend_npu():
     global npu_available
-    if npu_available:
-        return True
-    return False
+    return npu_available
 
 def is_mlu():
     global mlu_available
-    if mlu_available:
-        return True
-    return False
+    return mlu_available
 
 def is_ixuca():
     global ixuca_available
-    if ixuca_available:
-        return True
-    return False
+    return ixuca_available
+
+def is_musa():
+    global musa_available
+    return musa_available
 
 def get_torch_device():
     global directml_enabled
@@ -310,7 +313,7 @@ def amd_min_version(device=None, min_rdna_version=0):
     return False
 
 MIN_WEIGHT_MEMORY_RATIO = 0.4
-if is_nvidia():
+if is_nvidia() or is_musa():
     MIN_WEIGHT_MEMORY_RATIO = 0.0
 
 ENABLE_PYTORCH_ATTENTION = False
@@ -319,7 +322,7 @@ if args.use_pytorch_cross_attention:
     XFORMERS_IS_AVAILABLE = False
 
 try:
-    if is_nvidia():
+    if is_nvidia() or is_musa():
         if torch_version_numeric[0] >= 2:
             if ENABLE_PYTORCH_ATTENTION == False and args.use_split_cross_attention == False and args.use_quad_cross_attention == False:
                 ENABLE_PYTORCH_ATTENTION = True
@@ -386,7 +389,7 @@ if ENABLE_PYTORCH_ATTENTION:
 
 PRIORITIZE_FP16 = False  # TODO: remove and replace with something that shows exactly which dtype is faster than the other
 try:
-    if (is_nvidia() or is_amd()) and PerformanceFeature.Fp16Accumulation in args.fast:
+    if (is_nvidia() or is_amd() or is_musa()) and PerformanceFeature.Fp16Accumulation in args.fast:
         torch.backends.cuda.matmul.allow_fp16_accumulation = True
         PRIORITIZE_FP16 = True  # TODO: limit to cards where it actually boosts performance
         logging.info("Enabled fp16 accumulation.")
@@ -1031,7 +1034,7 @@ if args.async_offload is not None:
     NUM_STREAMS = args.async_offload
 else:
     #  Enable by default on Nvidia and AMD
-    if is_nvidia() or is_amd():
+    if is_nvidia() or is_amd() or is_musa():
         NUM_STREAMS = 2
 
 if args.disable_async_offload:
@@ -1128,7 +1131,7 @@ PINNED_MEMORY = {}
 TOTAL_PINNED_MEMORY = 0
 MAX_PINNED_MEMORY = -1
 if not args.disable_pinned_memory:
-    if is_nvidia() or is_amd():
+    if is_nvidia() or is_amd() or is_musa():
         if WINDOWS:
             MAX_PINNED_MEMORY = get_total_memory(torch.device("cpu")) * 0.45  # Windows limit is apparently 50%
         else:
@@ -1272,6 +1275,8 @@ def pytorch_attention_flash_attention():
             return True #if you have pytorch attention enabled on AMD it probably supports at least mem efficient attention
         if is_ixuca():
             return True
+        if is_musa():
+            return True
     return False
 
 def force_upcast_attention_dtype():
@@ -1403,6 +1408,9 @@ def should_use_fp16(device=None, model_params=0, prioritize_performance=True, ma
     if torch.version.hip:
         return True
 
+    if is_musa():
+        return True
+
     props = torch.cuda.get_device_properties(device)
     if props.major >= 8:
         return True
@@ -1473,6 +1481,9 @@ def should_use_bf16(device=None, model_params=0, prioritize_performance=True, ma
                 return True
             return False
 
+    if is_musa():
+        return True
+
     props = torch.cuda.get_device_properties(device)
 
     if is_mlu():
@@ -1495,25 +1506,27 @@ def supports_fp8_compute(device=None):
     if SUPPORT_FP8_OPS:
         return True
 
-    if not is_nvidia():
-        return False
-
     props = torch.cuda.get_device_properties(device)
-    if props.major >= 9:
-        return True
-    if props.major < 8:
-        return False
-    if props.minor < 9:
-        return False
-
-    if torch_version_numeric < (2, 3):
-        return False
-
-    if WINDOWS:
-        if torch_version_numeric < (2, 4):
+    if is_nvidia():
+        if props.major >= 9:
+            return True
+        if props.major < 8:
+            return False
+        if props.minor < 9:
             return False
 
-    return True
+        if torch_version_numeric < (2, 3):
+            return False
+
+        if WINDOWS:
+            if torch_version_numeric < (2, 4):
+                return False
+
+    elif is_musa():
+        if props.major >= 3:
+            return True
+
+    return False
 
 def supports_nvfp4_compute(device=None):
     if not is_nvidia():
@@ -1564,7 +1577,7 @@ def unload_all_models():
     free_memory(1e30, get_torch_device())
 
 def debug_memory_summary():
-    if is_amd() or is_nvidia():
+    if is_amd() or is_nvidia() or is_musa():
         return torch.cuda.memory.memory_summary()
     return ""
 

comfy_api/latest/_io.py

@@ -1146,6 +1146,20 @@ class ImageCompare(ComfyTypeI):
       def as_dict(self):
           return super().as_dict()
 
+
+@comfytype(io_type="COLOR")
+class Color(ComfyTypeIO):
+  Type = str
+
+  class Input(WidgetInput):
+      def __init__(self, id: str, display_name: str=None, optional=False, tooltip: str=None,
+                   socketless: bool=True, advanced: bool=None, default: str="#ffffff"):
+          super().__init__(id, display_name, optional, tooltip, None, default, socketless, None, None, None, None, advanced)
+          self.default: str
+
+      def as_dict(self):
+          return super().as_dict()
+
 DYNAMIC_INPUT_LOOKUP: dict[str, Callable[[dict[str, Any], dict[str, Any], tuple[str, dict[str, Any]], str, list[str] | None], None]] = {}
 def register_dynamic_input_func(io_type: str, func: Callable[[dict[str, Any], dict[str, Any], tuple[str, dict[str, Any]], str, list[str] | None], None]):
     DYNAMIC_INPUT_LOOKUP[io_type] = func
@@ -1252,23 +1266,6 @@ class NodeInfoV1:
     price_badge: dict | None = None
     search_aliases: list[str]=None
 
-@dataclass
-class NodeInfoV3:
-    input: dict=None
-    output: dict=None
-    hidden: list[str]=None
-    name: str=None
-    display_name: str=None
-    description: str=None
-    python_module: Any = None
-    category: str=None
-    output_node: bool=None
-    deprecated: bool=None
-    experimental: bool=None
-    dev_only: bool=None
-    api_node: bool=None
-    price_badge: dict | None = None
-
 
 @dataclass
 class PriceBadgeDepends:
@@ -1497,40 +1494,6 @@ class Schema:
         )
         return info
 
-
-    def get_v3_info(self, cls) -> NodeInfoV3:
-        input_dict = {}
-        output_dict = {}
-        hidden_list = []
-        # TODO: make sure dynamic types will be handled correctly
-        if self.inputs:
-            for input in self.inputs:
-                add_to_dict_v3(input, input_dict)
-        if self.outputs:
-            for output in self.outputs:
-                add_to_dict_v3(output, output_dict)
-        if self.hidden:
-            for hidden in self.hidden:
-                hidden_list.append(hidden.value)
-
-        info = NodeInfoV3(
-            input=input_dict,
-            output=output_dict,
-            hidden=hidden_list,
-            name=self.node_id,
-            display_name=self.display_name,
-            description=self.description,
-            category=self.category,
-            output_node=self.is_output_node,
-            deprecated=self.is_deprecated,
-            experimental=self.is_experimental,
-            dev_only=self.is_dev_only,
-            api_node=self.is_api_node,
-            python_module=getattr(cls, "RELATIVE_PYTHON_MODULE", "nodes"),
-            price_badge=self.price_badge.as_dict(self.inputs) if self.price_badge is not None else None,
-        )
-        return info
-
 def get_finalized_class_inputs(d: dict[str, Any], live_inputs: dict[str, Any], include_hidden=False) -> tuple[dict[str, Any], V3Data]:
     out_dict = {
         "required": {},
@@ -1585,9 +1548,6 @@ def add_to_dict_v1(i: Input, d: dict):
     as_dict.pop("optional", None)
     d.setdefault(key, {})[i.id] = (i.get_io_type(), as_dict)
 
-def add_to_dict_v3(io: Input | Output, d: dict):
-    d[io.id] = (io.get_io_type(), io.as_dict())
-
 class DynamicPathsDefaultValue:
     EMPTY_DICT = "empty_dict"
 
@@ -1748,13 +1708,6 @@ class _ComfyNodeBaseInternal(_ComfyNodeInternal):
         # set hidden
         type_clone.hidden = HiddenHolder.from_v3_data(v3_data)
         return type_clone
-
-    @final
-    @classmethod
-    def GET_NODE_INFO_V3(cls) -> dict[str, Any]:
-        schema = cls.GET_SCHEMA()
-        info = schema.get_v3_info(cls)
-        return asdict(info)
     #############################################
     # V1 Backwards Compatibility code
     #--------------------------------------------
@@ -2099,6 +2052,7 @@ __all__ = [
     "AnyType",
     "MultiType",
     "Tracks",
+    "Color",
     # Dynamic Types
     "MatchType",
     "DynamicCombo",
@@ -2107,12 +2061,10 @@ __all__ = [
     "HiddenHolder",
     "Hidden",
     "NodeInfoV1",
-    "NodeInfoV3",
     "Schema",
     "ComfyNode",
     "NodeOutput",
     "add_to_dict_v1",
-    "add_to_dict_v3",
     "V3Data",
     "ImageCompare",
     "PriceBadgeDepends",

comfy_api_nodes/apis/recraft.py

@@ -1,11 +1,8 @@
 from __future__ import annotations
 
-
-
 from enum import Enum
-from typing import Optional
 
-from pydantic import BaseModel, Field, conint, confloat
+from pydantic import BaseModel, Field
 
 
 class RecraftColor:
@@ -229,24 +226,24 @@ class RecraftColorObject(BaseModel):
 
 
 class RecraftControlsObject(BaseModel):
-    colors: Optional[list[RecraftColorObject]] = Field(None, description='An array of preferable colors')
-    background_color: Optional[RecraftColorObject] = Field(None, description='Use given color as a desired background color')
-    no_text: Optional[bool] = Field(None, description='Do not embed text layouts')
-    artistic_level: Optional[conint(ge=0, le=5)] = Field(None, description='Defines artistic tone of your image. At a simple level, the person looks straight at the camera in a static and clean style. Dynamic and eccentric levels introduce movement and creativity. The value should be in range [0..5].')
+    colors: list[RecraftColorObject] | None = Field(None, description='An array of preferable colors')
+    background_color: RecraftColorObject | None = Field(None, description='Use given color as a desired background color')
+    no_text: bool | None = Field(None, description='Do not embed text layouts')
+    artistic_level: int | None = Field(None, description='Defines artistic tone of your image. At a simple level, the person looks straight at the camera in a static and clean style. Dynamic and eccentric levels introduce movement and creativity. The value should be in range [0..5].')
 
 
 class RecraftImageGenerationRequest(BaseModel):
     prompt: str = Field(..., description='The text prompt describing the image to generate')
-    size: Optional[RecraftImageSize] = Field(None, description='The size of the generated image (e.g., "1024x1024")')
-    n: conint(ge=1, le=6) = Field(..., description='The number of images to generate')
-    negative_prompt: Optional[str] = Field(None, description='A text description of undesired elements on an image')
-    model: Optional[RecraftModel] = Field(RecraftModel.recraftv3, description='The model to use for generation (e.g., "recraftv3")')
-    style: Optional[str] = Field(None, description='The style to apply to the generated image (e.g., "digital_illustration")')
-    substyle: Optional[str] = Field(None, description='The substyle to apply to the generated image, depending on the style input')
-    controls: Optional[RecraftControlsObject] = Field(None, description='A set of custom parameters to tweak generation process')
-    style_id: Optional[str] = Field(None, description='Use a previously uploaded style as a reference; UUID')
-    strength: Optional[confloat(ge=0.0, le=1.0)] = Field(None, description='Defines the difference with the original image, should lie in [0, 1], where 0 means almost identical, and 1 means miserable similarity')
-    random_seed: Optional[int] = Field(None, description="Seed for video generation")
+    size: RecraftImageSize | None = Field(None, description='The size of the generated image (e.g., "1024x1024")')
+    n: int = Field(..., description='The number of images to generate')
+    negative_prompt: str | None = Field(None, description='A text description of undesired elements on an image')
+    model: RecraftModel | None = Field(RecraftModel.recraftv3, description='The model to use for generation (e.g., "recraftv3")')
+    style: str | None = Field(None, description='The style to apply to the generated image (e.g., "digital_illustration")')
+    substyle: str | None = Field(None, description='The substyle to apply to the generated image, depending on the style input')
+    controls: RecraftControlsObject | None = Field(None, description='A set of custom parameters to tweak generation process')
+    style_id: str | None = Field(None, description='Use a previously uploaded style as a reference; UUID')
+    strength: float | None = Field(None, description='Defines the difference with the original image, should lie in [0, 1], where 0 means almost identical, and 1 means miserable similarity')
+    random_seed: int | None = Field(None, description="Seed for video generation")
     # text_layout
 
 
@@ -258,5 +255,13 @@ class RecraftReturnedObject(BaseModel):
 class RecraftImageGenerationResponse(BaseModel):
     created: int = Field(..., description='Unix timestamp when the generation was created')
     credits: int = Field(..., description='Number of credits used for the generation')
-    data: Optional[list[RecraftReturnedObject]] = Field(None, description='Array of generated image information')
-    image: Optional[RecraftReturnedObject] = Field(None, description='Single generated image')
+    data: list[RecraftReturnedObject] | None = Field(None, description='Array of generated image information')
+    image: RecraftReturnedObject | None = Field(None, description='Single generated image')
+
+
+class RecraftCreateStyleRequest(BaseModel):
+    style: str = Field(..., description="realistic_image, digital_illustration, vector_illustration, or icon")
+
+
+class RecraftCreateStyleResponse(BaseModel):
+    id: str = Field(..., description="UUID of the created style")

comfy_api_nodes/nodes_recraft.py

@@ -12,6 +12,8 @@ from comfy_api_nodes.apis.recraft import (
     RecraftColor,
     RecraftColorChain,
     RecraftControls,
+    RecraftCreateStyleRequest,
+    RecraftCreateStyleResponse,
     RecraftImageGenerationRequest,
     RecraftImageGenerationResponse,
     RecraftImageSize,
@@ -323,6 +325,75 @@ class RecraftStyleInfiniteStyleLibrary(IO.ComfyNode):
         return IO.NodeOutput(RecraftStyle(style_id=style_id))
 
 
+class RecraftCreateStyleNode(IO.ComfyNode):
+    @classmethod
+    def define_schema(cls):
+        return IO.Schema(
+            node_id="RecraftCreateStyleNode",
+            display_name="Recraft Create Style",
+            category="api node/image/Recraft",
+            description="Create a custom style from reference images. "
+            "Upload 1-5 images to use as style references. "
+            "Total size of all images is limited to 5 MB.",
+            inputs=[
+                IO.Combo.Input(
+                    "style",
+                    options=["realistic_image", "digital_illustration"],
+                    tooltip="The base style of the generated images.",
+                ),
+                IO.Autogrow.Input(
+                    "images",
+                    template=IO.Autogrow.TemplatePrefix(
+                        IO.Image.Input("image"),
+                        prefix="image",
+                        min=1,
+                        max=5,
+                    ),
+                ),
+            ],
+            outputs=[
+                IO.String.Output(display_name="style_id"),
+            ],
+            hidden=[
+                IO.Hidden.auth_token_comfy_org,
+                IO.Hidden.api_key_comfy_org,
+                IO.Hidden.unique_id,
+            ],
+            is_api_node=True,
+            price_badge=IO.PriceBadge(
+                expr="""{"type":"usd","usd": 0.04}""",
+            ),
+        )
+
+    @classmethod
+    async def execute(
+        cls,
+        style: str,
+        images: IO.Autogrow.Type,
+    ) -> IO.NodeOutput:
+        files = []
+        total_size = 0
+        max_total_size = 5 * 1024 * 1024  # 5 MB limit
+        for i, img in enumerate(list(images.values())):
+            file_bytes = tensor_to_bytesio(img, total_pixels=2048 * 2048, mime_type="image/webp").read()
+            total_size += len(file_bytes)
+            if total_size > max_total_size:
+                raise Exception("Total size of all images exceeds 5 MB limit.")
+            files.append((f"file{i + 1}", file_bytes))
+
+        response = await sync_op(
+            cls,
+            endpoint=ApiEndpoint(path="/proxy/recraft/styles", method="POST"),
+            response_model=RecraftCreateStyleResponse,
+            files=files,
+            data=RecraftCreateStyleRequest(style=style),
+            content_type="multipart/form-data",
+            max_retries=1,
+        )
+
+        return IO.NodeOutput(response.id)
+
+
 class RecraftTextToImageNode(IO.ComfyNode):
     @classmethod
     def define_schema(cls):
@@ -395,7 +466,7 @@ class RecraftTextToImageNode(IO.ComfyNode):
         negative_prompt: str = None,
         recraft_controls: RecraftControls = None,
     ) -> IO.NodeOutput:
-        validate_string(prompt, strip_whitespace=False, max_length=1000)
+        validate_string(prompt, strip_whitespace=False, min_length=1, max_length=1000)
         default_style = RecraftStyle(RecraftStyleV3.realistic_image)
         if recraft_style is None:
             recraft_style = default_style
@@ -1024,6 +1095,7 @@ class RecraftExtension(ComfyExtension):
             RecraftStyleV3DigitalIllustrationNode,
             RecraftStyleV3LogoRasterNode,
             RecraftStyleInfiniteStyleLibrary,
+            RecraftCreateStyleNode,
             RecraftColorRGBNode,
             RecraftControlsNode,
         ]

comfy_extras/nodes_color.py

@@ -0,0 +1,42 @@
+from typing_extensions import override
+from comfy_api.latest import ComfyExtension, io
+
+
+class ColorToRGBInt(io.ComfyNode):
+    @classmethod
+    def define_schema(cls) -> io.Schema:
+        return io.Schema(
+            node_id="ColorToRGBInt",
+            display_name="Color to RGB Int",
+            category="utils",
+            description="Convert a color to a RGB integer value.",
+            inputs=[
+                io.Color.Input("color"),
+            ],
+            outputs=[
+                io.Int.Output(display_name="rgb_int"),
+            ],
+        )
+
+    @classmethod
+    def execute(
+        cls,
+        color: str,
+    ) -> io.NodeOutput:
+        # expect format #RRGGBB
+        if len(color) != 7 or color[0] != "#":
+            raise ValueError("Color must be in format #RRGGBB")
+        r = int(color[1:3], 16)
+        g = int(color[3:5], 16)
+        b = int(color[5:7], 16)
+        return io.NodeOutput(r * 256 * 256 + g * 256 + b)
+
+
+class ColorExtension(ComfyExtension):
+    @override
+    async def get_node_list(self) -> list[type[io.ComfyNode]]:
+        return [ColorToRGBInt]
+
+
+async def comfy_entrypoint() -> ColorExtension:
+    return ColorExtension()

comfy_extras/nodes_hunyuan.py

@@ -56,7 +56,7 @@ class EmptyHunyuanLatentVideo(io.ComfyNode):
     @classmethod
     def execute(cls, width, height, length, batch_size=1) -> io.NodeOutput:
         latent = torch.zeros([batch_size, 16, ((length - 1) // 4) + 1, height // 8, width // 8], device=comfy.model_management.intermediate_device())
-        return io.NodeOutput({"samples":latent})
+        return io.NodeOutput({"samples": latent, "downscale_ratio_spacial": 8})
 
     generate = execute  # TODO: remove
 
@@ -73,7 +73,7 @@ class EmptyHunyuanVideo15Latent(EmptyHunyuanLatentVideo):
     def execute(cls, width, height, length, batch_size=1) -> io.NodeOutput:
         # Using scale factor of 16 instead of 8
         latent = torch.zeros([batch_size, 32, ((length - 1) // 4) + 1, height // 16, width // 16], device=comfy.model_management.intermediate_device())
-        return io.NodeOutput({"samples": latent})
+        return io.NodeOutput({"samples": latent, "downscale_ratio_spacial": 16})
 
 
 class HunyuanVideo15ImageToVideo(io.ComfyNode):

main.py

@@ -37,11 +37,13 @@ if __name__ == "__main__":
         devices = ','.join(map(str, devices))
         os.environ['CUDA_VISIBLE_DEVICES'] = str(devices)
         os.environ['HIP_VISIBLE_DEVICES'] = str(devices)
+        os.environ['MUSA_VISIBLE_DEVICES'] = str(devices)
 
     if args.cuda_device is not None:
         os.environ['CUDA_VISIBLE_DEVICES'] = str(args.cuda_device)
         os.environ['HIP_VISIBLE_DEVICES'] = str(args.cuda_device)
         os.environ["ASCEND_RT_VISIBLE_DEVICES"] = str(args.cuda_device)
+        os.environ['MUSA_VISIBLE_DEVICES'] = str(args.cuda_device)
         logging.info("Set cuda device to: {}".format(args.cuda_device))
 
     if args.oneapi_device_selector is not None:

nodes.py

@@ -2432,7 +2432,8 @@ async def init_builtin_extra_nodes():
         "nodes_wanmove.py",
         "nodes_image_compare.py",
         "nodes_zimage.py",
-        "nodes_lora_debug.py"
+        "nodes_lora_debug.py",
+        "nodes_color.py"
     ]
 
     import_failed = []

requirements.txt

@@ -29,3 +29,4 @@ kornia>=0.7.1
 spandrel
 pydantic~=2.0
 pydantic-settings~=2.0
+torchada>=0.1.22