Merge branch 'master' into dr-support-pip-cm

.gitattributes

@@ -1,2 +1,3 @@
 /web/assets/** linguist-generated
 /web/** linguist-vendored
+comfy_api_nodes/apis/__init__.py linguist-generated

README.md

@@ -66,6 +66,7 @@ See what ComfyUI can do with the [example workflows](https://comfyanonymous.gith
    - [Lumina Image 2.0](https://comfyanonymous.github.io/ComfyUI_examples/lumina2/)
    - [HiDream](https://comfyanonymous.github.io/ComfyUI_examples/hidream/)
    - [Cosmos Predict2](https://comfyanonymous.github.io/ComfyUI_examples/cosmos_predict2/)
+   - [Qwen Image](https://comfyanonymous.github.io/ComfyUI_examples/qwen_image/)
 - Image Editing Models
    - [Omnigen 2](https://comfyanonymous.github.io/ComfyUI_examples/omnigen/)
    - [Flux Kontext](https://comfyanonymous.github.io/ComfyUI_examples/flux/#flux-kontext-image-editing-model)

comfy/ldm/qwen_image/model.py

@@ -8,7 +8,7 @@ from einops import repeat
 from comfy.ldm.lightricks.model import TimestepEmbedding, Timesteps
 from comfy.ldm.modules.attention import optimized_attention_masked
 from comfy.ldm.flux.layers import EmbedND
-
+import comfy.ldm.common_dit
 
 class GELU(nn.Module):
     def __init__(self, dim_in: int, dim_out: int, approximate: str = "none", bias: bool = True, dtype=None, device=None, operations=None):
@@ -364,8 +364,9 @@ class QwenImageTransformer2DModel(nn.Module):
 
         image_rotary_emb = self.pos_embeds(x, context)
 
-        orig_shape = x.shape
+        hidden_states = comfy.ldm.common_dit.pad_to_patch_size(x, (1, self.patch_size, self.patch_size))
-        hidden_states = x.view(orig_shape[0], orig_shape[1], orig_shape[-2] // 2, 2, orig_shape[-1] // 2, 2)
+        orig_shape = hidden_states.shape
+        hidden_states = hidden_states.view(orig_shape[0], orig_shape[1], orig_shape[-2] // 2, 2, orig_shape[-1] // 2, 2)
         hidden_states = hidden_states.permute(0, 2, 4, 1, 3, 5)
         hidden_states = hidden_states.reshape(orig_shape[0], (orig_shape[-2] // 2) * (orig_shape[-1] // 2), orig_shape[1] * 4)
 
@@ -396,4 +397,4 @@ class QwenImageTransformer2DModel(nn.Module):
 
         hidden_states = hidden_states.view(orig_shape[0], orig_shape[-2] // 2, orig_shape[-1] // 2, orig_shape[1], 2, 2)
         hidden_states = hidden_states.permute(0, 3, 1, 4, 2, 5)
-        return hidden_states.reshape(orig_shape)
+        return hidden_states.reshape(orig_shape)[:, :, :, :x.shape[-2], :x.shape[-1]]

comfy/lora.py

@@ -293,6 +293,15 @@ def model_lora_keys_unet(model, key_map={}):
                 key_lora = k[len("diffusion_model."):-len(".weight")]
                 key_map["{}".format(key_lora)] = k
 
+    if isinstance(model, comfy.model_base.QwenImage):
+        for k in sdk:
+            if k.startswith("diffusion_model.") and k.endswith(".weight"): #QwenImage lora format
+                key_lora = k[len("diffusion_model."):-len(".weight")]
+                # Direct mapping for transformer_blocks format (QwenImage LoRA format)
+                key_map["{}".format(key_lora)] = k
+                # Support transformer prefix format
+                key_map["transformer.{}".format(key_lora)] = k
+
     return key_map
 
 

comfy/supported_models.py

@@ -1237,7 +1237,7 @@ class QwenImage(supported_models_base.BASE):
 
     sampling_settings = {
         "multiplier": 1.0,
-        "shift": 2.6,
+        "shift": 1.15,
     }
 
     memory_usage_factor = 1.8 #TODO

comfy/weight_adapter/lora.py

@@ -96,6 +96,7 @@ class LoRAAdapter(WeightAdapterBase):
         diffusers3_lora = "{}.lora.up.weight".format(x)
         mochi_lora = "{}.lora_B".format(x)
         transformers_lora = "{}.lora_linear_layer.up.weight".format(x)
+        qwen_default_lora = "{}.lora_B.default.weight".format(x)
         A_name = None
 
         if regular_lora in lora.keys():
@@ -122,6 +123,10 @@ class LoRAAdapter(WeightAdapterBase):
             A_name = transformers_lora
             B_name = "{}.lora_linear_layer.down.weight".format(x)
             mid_name = None
+        elif qwen_default_lora in lora.keys():
+            A_name = qwen_default_lora
+            B_name = "{}.lora_A.default.weight".format(x)
+            mid_name = None
 
         if A_name is not None:
             mid = None

comfy_api_nodes/apis/tripo_api.py

@@ -127,7 +127,7 @@ class TripoTextToModelRequest(BaseModel):
     type: TripoTaskType = Field(TripoTaskType.TEXT_TO_MODEL, description='Type of task')
     prompt: str = Field(..., description='The text prompt describing the model to generate', max_length=1024)
     negative_prompt: Optional[str] = Field(None, description='The negative text prompt', max_length=1024)
-    model_version: Optional[TripoModelVersion] = TripoModelVersion.V2_5
+    model_version: Optional[TripoModelVersion] = TripoModelVersion.v2_5_20250123
     face_limit: Optional[int] = Field(None, description='The number of faces to limit the generation to')
     texture: Optional[bool] = Field(True, description='Whether to apply texture to the generated model')
     pbr: Optional[bool] = Field(True, description='Whether to apply PBR to the generated model')

comfy_api_nodes/nodes_veo2.py

@@ -8,10 +8,10 @@ from typing import Optional
 from comfy.comfy_types.node_typing import IO, ComfyNodeABC
 from comfy_api.input_impl.video_types import VideoFromFile
 from comfy_api_nodes.apis import (
-    Veo2GenVidRequest,
+    VeoGenVidRequest,
-    Veo2GenVidResponse,
+    VeoGenVidResponse,
-    Veo2GenVidPollRequest,
+    VeoGenVidPollRequest,
-    Veo2GenVidPollResponse
+    VeoGenVidPollResponse
 )
 from comfy_api_nodes.apis.client import (
     ApiEndpoint,
@@ -35,7 +35,7 @@ def convert_image_to_base64(image: torch.Tensor):
     return tensor_to_base64_string(scaled_image)
 
 
-def get_video_url_from_response(poll_response: Veo2GenVidPollResponse) -> Optional[str]:
+def get_video_url_from_response(poll_response: VeoGenVidPollResponse) -> Optional[str]:
     if (
         poll_response.response
         and hasattr(poll_response.response, "videos")
@@ -130,6 +130,14 @@ class VeoVideoGenerationNode(ComfyNodeABC):
                     "default": None,
                     "tooltip": "Optional reference image to guide video generation",
                 }),
+                "model": (
+                    IO.COMBO,
+                    {
+                        "options": ["veo-2.0-generate-001"],
+                        "default": "veo-2.0-generate-001",
+                        "tooltip": "Veo 2 model to use for video generation",
+                    },
+                ),
             },
             "hidden": {
                 "auth_token": "AUTH_TOKEN_COMFY_ORG",
@@ -141,7 +149,7 @@ class VeoVideoGenerationNode(ComfyNodeABC):
     RETURN_TYPES = (IO.VIDEO,)
     FUNCTION = "generate_video"
     CATEGORY = "api node/video/Veo"
-    DESCRIPTION = "Generates videos from text prompts using Google's Veo API"
+    DESCRIPTION = "Generates videos from text prompts using Google's Veo 2 API"
     API_NODE = True
 
     def generate_video(
@@ -154,6 +162,8 @@ class VeoVideoGenerationNode(ComfyNodeABC):
         person_generation="ALLOW",
         seed=0,
         image=None,
+        model="veo-2.0-generate-001",
+        generate_audio=False,
         unique_id: Optional[str] = None,
         **kwargs,
     ):
@@ -188,16 +198,19 @@ class VeoVideoGenerationNode(ComfyNodeABC):
             parameters["negativePrompt"] = negative_prompt
         if seed > 0:
             parameters["seed"] = seed
+        # Only add generateAudio for Veo 3 models
+        if "veo-3.0" in model:
+            parameters["generateAudio"] = generate_audio
 
         # Initial request to start video generation
         initial_operation = SynchronousOperation(
             endpoint=ApiEndpoint(
-                path="/proxy/veo/generate",
+                path=f"/proxy/veo/{model}/generate",
                 method=HttpMethod.POST,
-                request_model=Veo2GenVidRequest,
+                request_model=VeoGenVidRequest,
-                response_model=Veo2GenVidResponse
+                response_model=VeoGenVidResponse
             ),
-            request=Veo2GenVidRequest(
+            request=VeoGenVidRequest(
                 instances=instances,
                 parameters=parameters
             ),
@@ -223,16 +236,16 @@ class VeoVideoGenerationNode(ComfyNodeABC):
         # Define the polling operation
         poll_operation = PollingOperation(
             poll_endpoint=ApiEndpoint(
-                path="/proxy/veo/poll",
+                path=f"/proxy/veo/{model}/poll",
                 method=HttpMethod.POST,
-                request_model=Veo2GenVidPollRequest,
+                request_model=VeoGenVidPollRequest,
-                response_model=Veo2GenVidPollResponse
+                response_model=VeoGenVidPollResponse
             ),
             completed_statuses=["completed"],
             failed_statuses=[],  # No failed statuses, we'll handle errors after polling
             status_extractor=status_extractor,
             progress_extractor=progress_extractor,
-            request=Veo2GenVidPollRequest(
+            request=VeoGenVidPollRequest(
                 operationName=operation_name
             ),
             auth_kwargs=kwargs,
@@ -298,11 +311,64 @@ class VeoVideoGenerationNode(ComfyNodeABC):
         return (VideoFromFile(video_io),)
 
 
-# Register the node
+class Veo3VideoGenerationNode(VeoVideoGenerationNode):
+    """
+    Generates videos from text prompts using Google's Veo 3 API.
+
+    Supported models:
+    - veo-3.0-generate-001
+    - veo-3.0-fast-generate-001
+
+    This node extends the base Veo node with Veo 3 specific features including
+    audio generation and fixed 8-second duration.
+    """
+
+    @classmethod
+    def INPUT_TYPES(s):
+        parent_input = super().INPUT_TYPES()
+
+        # Update model options for Veo 3
+        parent_input["optional"]["model"] = (
+            IO.COMBO,
+            {
+                "options": ["veo-3.0-generate-001", "veo-3.0-fast-generate-001"],
+                "default": "veo-3.0-generate-001",
+                "tooltip": "Veo 3 model to use for video generation",
+            },
+        )
+
+        # Add generateAudio parameter
+        parent_input["optional"]["generate_audio"] = (
+            IO.BOOLEAN,
+            {
+                "default": False,
+                "tooltip": "Generate audio for the video. Supported by all Veo 3 models.",
+            }
+        )
+
+        # Update duration constraints for Veo 3 (only 8 seconds supported)
+        parent_input["optional"]["duration_seconds"] = (
+            IO.INT,
+            {
+                "default": 8,
+                "min": 8,
+                "max": 8,
+                "step": 1,
+                "display": "number",
+                "tooltip": "Duration of the output video in seconds (Veo 3 only supports 8 seconds)",
+            },
+        )
+
+        return parent_input
+
+
+# Register the nodes
 NODE_CLASS_MAPPINGS = {
     "VeoVideoGenerationNode": VeoVideoGenerationNode,
+    "Veo3VideoGenerationNode": Veo3VideoGenerationNode,
 }
 
 NODE_DISPLAY_NAME_MAPPINGS = {
-    "VeoVideoGenerationNode": "Google Veo2 Video Generation",
+    "VeoVideoGenerationNode": "Google Veo 2 Video Generation",
+    "Veo3VideoGenerationNode": "Google Veo 3 Video Generation",
 }

comfy_extras/nodes_model_merging_model_specific.py

@@ -314,6 +314,29 @@ class ModelMergeCosmosPredict2_14B(comfy_extras.nodes_model_merging.ModelMergeBl
 
         return {"required": arg_dict}
 
+class ModelMergeQwenImage(comfy_extras.nodes_model_merging.ModelMergeBlocks):
+    CATEGORY = "advanced/model_merging/model_specific"
+
+    @classmethod
+    def INPUT_TYPES(s):
+        arg_dict = { "model1": ("MODEL",),
+                              "model2": ("MODEL",)}
+
+        argument = ("FLOAT", {"default": 1.0, "min": 0.0, "max": 1.0, "step": 0.01})
+
+        arg_dict["pos_embeds."] = argument
+        arg_dict["img_in."] = argument
+        arg_dict["txt_norm."] = argument
+        arg_dict["txt_in."] = argument
+        arg_dict["time_text_embed."] = argument
+
+        for i in range(60):
+            arg_dict["transformer_blocks.{}.".format(i)] = argument
+
+        arg_dict["proj_out."] = argument
+
+        return {"required": arg_dict}
+
 NODE_CLASS_MAPPINGS = {
     "ModelMergeSD1": ModelMergeSD1,
     "ModelMergeSD2": ModelMergeSD1, #SD1 and SD2 have the same blocks
@@ -329,4 +352,5 @@ NODE_CLASS_MAPPINGS = {
     "ModelMergeWAN2_1": ModelMergeWAN2_1,
     "ModelMergeCosmosPredict2_2B": ModelMergeCosmosPredict2_2B,
     "ModelMergeCosmosPredict2_14B": ModelMergeCosmosPredict2_14B,
+    "ModelMergeQwenImage": ModelMergeQwenImage,
 }

comfyui_version.py

@@ -1,3 +1,3 @@
 # This file is automatically generated by the build process when version is
 # updated in pyproject.toml.
-__version__ = "0.3.48"
+__version__ = "0.3.49"

pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "ComfyUI"
-version = "0.3.48"
+version = "0.3.49"
 readme = "README.md"
 license = { file = "LICENSE" }
 requires-python = ">=3.9"

requirements.txt

@@ -1,5 +1,5 @@
 comfyui-frontend-package==1.23.4
-comfyui-workflow-templates==0.1.47
+comfyui-workflow-templates==0.1.52
 comfyui-embedded-docs==0.2.4
 comfyui_manager
 torch