Merge branch 'comfyanonymous:master' into master

comfy_api_nodes/apis/gemini_api.py

@@ -58,8 +58,14 @@ class GeminiInlineData(BaseModel):
     mimeType: GeminiMimeType | None = Field(None)
 
 
+class GeminiFileData(BaseModel):
+    fileUri: str | None = Field(None)
+    mimeType: GeminiMimeType | None = Field(None)
+
+
 class GeminiPart(BaseModel):
     inlineData: GeminiInlineData | None = Field(None)
+    fileData: GeminiFileData | None = Field(None)
     text: str | None = Field(None)
 
 

comfy_api_nodes/apis/veo_api.py

@@ -1,34 +1,21 @@
-from typing import Optional, Union
+from typing import Optional
-from enum import Enum
 
 from pydantic import BaseModel, Field
 
 
-class Image2(BaseModel):
+class VeoRequestInstanceImage(BaseModel):
-    bytesBase64Encoded: str
+    bytesBase64Encoded: str | None = Field(None)
-    gcsUri: Optional[str] = None
+    gcsUri: str | None = Field(None)
-    mimeType: Optional[str] = None
+    mimeType: str | None = Field(None)
 
 
-class Image3(BaseModel):
+class VeoRequestInstance(BaseModel):
-    bytesBase64Encoded: Optional[str] = None
+    image: VeoRequestInstanceImage | None = Field(None)
-    gcsUri: str
+    lastFrame: VeoRequestInstanceImage | None = Field(None)
-    mimeType: Optional[str] = None
-
-
-class Instance1(BaseModel):
-    image: Optional[Union[Image2, Image3]] = Field(
-        None, description='Optional image to guide video generation'
-    )
     prompt: str = Field(..., description='Text description of the video')
 
 
-class PersonGeneration1(str, Enum):
+class VeoRequestParameters(BaseModel):
-    ALLOW = 'ALLOW'
-    BLOCK = 'BLOCK'
-
-
-class Parameters1(BaseModel):
     aspectRatio: Optional[str] = Field(None, examples=['16:9'])
     durationSeconds: Optional[int] = None
     enhancePrompt: Optional[bool] = None
@@ -37,17 +24,18 @@ class Parameters1(BaseModel):
         description='Generate audio for the video. Only supported by veo 3 models.',
     )
     negativePrompt: Optional[str] = None
-    personGeneration: Optional[PersonGeneration1] = None
+    personGeneration: str | None = Field(None, description="ALLOW or BLOCK")
     sampleCount: Optional[int] = None
     seed: Optional[int] = None
     storageUri: Optional[str] = Field(
         None, description='Optional Cloud Storage URI to upload the video'
     )
+    resolution: str | None = Field(None)
 
 
 class VeoGenVidRequest(BaseModel):
-    instances: Optional[list[Instance1]] = None
+    instances: list[VeoRequestInstance] | None = Field(None)
-    parameters: Optional[Parameters1] = None
+    parameters: VeoRequestParameters | None = Field(None)
 
 
 class VeoGenVidResponse(BaseModel):

comfy_api_nodes/nodes_gemini.py

@@ -20,6 +20,7 @@ from comfy_api.latest import IO, ComfyExtension, Input
 from comfy_api.util import VideoCodec, VideoContainer
 from comfy_api_nodes.apis.gemini_api import (
     GeminiContent,
+    GeminiFileData,
     GeminiGenerateContentRequest,
     GeminiGenerateContentResponse,
     GeminiImageConfig,
@@ -38,6 +39,7 @@ from comfy_api_nodes.util import (
     get_number_of_images,
     sync_op,
     tensor_to_base64_string,
+    upload_images_to_comfyapi,
     validate_string,
     video_to_base64_string,
 )
@@ -68,24 +70,43 @@ class GeminiImageModel(str, Enum):
     gemini_2_5_flash_image = "gemini-2.5-flash-image"
 
 
-def create_image_parts(image_input: torch.Tensor) -> list[GeminiPart]:
+async def create_image_parts(
-    """
+    cls: type[IO.ComfyNode],
-    Convert image tensor input to Gemini API compatible parts.
+    images: torch.Tensor,
-
+    image_limit: int = 0,
-    Args:
+) -> list[GeminiPart]:
-        image_input: Batch of image tensors from ComfyUI.
-
-    Returns:
-        List of GeminiPart objects containing the encoded images.
-    """
     image_parts: list[GeminiPart] = []
-    for image_index in range(image_input.shape[0]):
+    if image_limit < 0:
-        image_as_b64 = tensor_to_base64_string(image_input[image_index].unsqueeze(0))
+        raise ValueError("image_limit must be greater than or equal to 0 when creating Gemini image parts.")
+    total_images = get_number_of_images(images)
+    if total_images <= 0:
+        raise ValueError("No images provided to create_image_parts; at least one image is required.")
+
+    # If image_limit == 0 --> use all images; otherwise clamp to image_limit.
+    effective_max = total_images if image_limit == 0 else min(total_images, image_limit)
+
+    # Number of images we'll send as URLs (fileData)
+    num_url_images = min(effective_max, 10)  # Vertex API max number of image links
+    reference_images_urls = await upload_images_to_comfyapi(
+        cls,
+        images,
+        max_images=num_url_images,
+    )
+    for reference_image_url in reference_images_urls:
+        image_parts.append(
+            GeminiPart(
+                fileData=GeminiFileData(
+                    mimeType=GeminiMimeType.image_png,
+                    fileUri=reference_image_url,
+                )
+            )
+        )
+    for idx in range(num_url_images, effective_max):
         image_parts.append(
             GeminiPart(
                 inlineData=GeminiInlineData(
                     mimeType=GeminiMimeType.image_png,
-                    data=image_as_b64,
+                    data=tensor_to_base64_string(images[idx]),
                 )
             )
         )
@@ -338,8 +359,7 @@ class GeminiNode(IO.ComfyNode):
 
         # Add other modal parts
         if images is not None:
-            image_parts = create_image_parts(images)
+            parts.extend(await create_image_parts(cls, images))
-            parts.extend(image_parts)
         if audio is not None:
             parts.extend(cls.create_audio_parts(audio))
         if video is not None:
@@ -562,8 +582,7 @@ class GeminiImage(IO.ComfyNode):
         image_config = GeminiImageConfig(aspectRatio=aspect_ratio)
 
         if images is not None:
-            image_parts = create_image_parts(images)
+            parts.extend(await create_image_parts(cls, images))
-            parts.extend(image_parts)
         if files is not None:
             parts.extend(files)
 
@@ -702,7 +721,7 @@ class GeminiImage2(IO.ComfyNode):
         if images is not None:
             if get_number_of_images(images) > 14:
                 raise ValueError("The current maximum number of supported images is 14.")
-            parts.extend(create_image_parts(images))
+            parts.extend(await create_image_parts(cls, images))
         if files is not None:
             parts.extend(files)
 

comfy_api_nodes/nodes_veo2.py

@@ -1,6 +1,7 @@
 import base64
 from io import BytesIO
 
+import torch
 from typing_extensions import override
 
 from comfy_api.input_impl.video_types import VideoFromFile
@@ -10,6 +11,9 @@ from comfy_api_nodes.apis.veo_api import (
     VeoGenVidPollResponse,
     VeoGenVidRequest,
     VeoGenVidResponse,
+    VeoRequestInstance,
+    VeoRequestInstanceImage,
+    VeoRequestParameters,
 )
 from comfy_api_nodes.util import (
     ApiEndpoint,
@@ -346,12 +350,163 @@ class Veo3VideoGenerationNode(VeoVideoGenerationNode):
         )
 
 
+class Veo3FirstLastFrameNode(IO.ComfyNode):
+
+    @classmethod
+    def define_schema(cls):
+        return IO.Schema(
+            node_id="Veo3FirstLastFrameNode",
+            display_name="Google Veo 3 First-Last-Frame to Video",
+            category="api node/video/Veo",
+            description="Generate video using prompt and first and last frames.",
+            inputs=[
+                IO.String.Input(
+                    "prompt",
+                    multiline=True,
+                    default="",
+                    tooltip="Text description of the video",
+                ),
+                IO.String.Input(
+                    "negative_prompt",
+                    multiline=True,
+                    default="",
+                    tooltip="Negative text prompt to guide what to avoid in the video",
+                ),
+                IO.Combo.Input("resolution", options=["720p", "1080p"]),
+                IO.Combo.Input(
+                    "aspect_ratio",
+                    options=["16:9", "9:16"],
+                    default="16:9",
+                    tooltip="Aspect ratio of the output video",
+                ),
+                IO.Int.Input(
+                    "duration",
+                    default=8,
+                    min=4,
+                    max=8,
+                    step=2,
+                    display_mode=IO.NumberDisplay.slider,
+                    tooltip="Duration of the output video in seconds",
+                ),
+                IO.Int.Input(
+                    "seed",
+                    default=0,
+                    min=0,
+                    max=0xFFFFFFFF,
+                    step=1,
+                    display_mode=IO.NumberDisplay.number,
+                    control_after_generate=True,
+                    tooltip="Seed for video generation",
+                ),
+                IO.Image.Input("first_frame", tooltip="Start frame"),
+                IO.Image.Input("last_frame", tooltip="End frame"),
+                IO.Combo.Input(
+                    "model",
+                    options=["veo-3.1-generate", "veo-3.1-fast-generate"],
+                    default="veo-3.1-fast-generate",
+                ),
+                IO.Boolean.Input(
+                    "generate_audio",
+                    default=True,
+                    tooltip="Generate audio for the video.",
+                ),
+            ],
+            outputs=[
+                IO.Video.Output(),
+            ],
+            hidden=[
+                IO.Hidden.auth_token_comfy_org,
+                IO.Hidden.api_key_comfy_org,
+                IO.Hidden.unique_id,
+            ],
+            is_api_node=True,
+        )
+
+    @classmethod
+    async def execute(
+        cls,
+        prompt: str,
+        negative_prompt: str,
+        resolution: str,
+        aspect_ratio: str,
+        duration: int,
+        seed: int,
+        first_frame: torch.Tensor,
+        last_frame: torch.Tensor,
+        model: str,
+        generate_audio: bool,
+    ):
+        model = MODELS_MAP[model]
+        initial_response = await sync_op(
+            cls,
+            ApiEndpoint(path=f"/proxy/veo/{model}/generate", method="POST"),
+            response_model=VeoGenVidResponse,
+            data=VeoGenVidRequest(
+                instances=[
+                    VeoRequestInstance(
+                        prompt=prompt,
+                        image=VeoRequestInstanceImage(
+                            bytesBase64Encoded=tensor_to_base64_string(first_frame), mimeType="image/png"
+                        ),
+                        lastFrame=VeoRequestInstanceImage(
+                            bytesBase64Encoded=tensor_to_base64_string(last_frame), mimeType="image/png"
+                        ),
+                    ),
+                ],
+                parameters=VeoRequestParameters(
+                    aspectRatio=aspect_ratio,
+                    personGeneration="ALLOW",
+                    durationSeconds=duration,
+                    enhancePrompt=True,  # cannot be False for Veo3
+                    seed=seed,
+                    generateAudio=generate_audio,
+                    negativePrompt=negative_prompt,
+                    resolution=resolution,
+                ),
+            ),
+        )
+        poll_response = await poll_op(
+            cls,
+            ApiEndpoint(path=f"/proxy/veo/{model}/poll", method="POST"),
+            response_model=VeoGenVidPollResponse,
+            status_extractor=lambda r: "completed" if r.done else "pending",
+            data=VeoGenVidPollRequest(
+                operationName=initial_response.name,
+            ),
+            poll_interval=5.0,
+            estimated_duration=AVERAGE_DURATION_VIDEO_GEN,
+        )
+
+        if poll_response.error:
+            raise Exception(f"Veo API error: {poll_response.error.message} (code: {poll_response.error.code})")
+
+        response = poll_response.response
+        filtered_count = response.raiMediaFilteredCount
+        if filtered_count:
+            reasons = response.raiMediaFilteredReasons or []
+            reason_part = f": {reasons[0]}" if reasons else ""
+            raise Exception(
+                f"Content blocked by Google's Responsible AI filters{reason_part} "
+                f"({filtered_count} video{'s' if filtered_count != 1 else ''} filtered)."
+            )
+
+        if response.videos:
+            video = response.videos[0]
+            if video.bytesBase64Encoded:
+                return IO.NodeOutput(VideoFromFile(BytesIO(base64.b64decode(video.bytesBase64Encoded))))
+            if video.gcsUri:
+                return IO.NodeOutput(await download_url_to_video_output(video.gcsUri))
+            raise Exception("Video returned but no data or URL was provided")
+        raise Exception("Video generation completed but no video was returned")
+
+
 class VeoExtension(ComfyExtension):
     @override
     async def get_node_list(self) -> list[type[IO.ComfyNode]]:
         return [
             VeoVideoGenerationNode,
             Veo3VideoGenerationNode,
+            Veo3FirstLastFrameNode,
         ]
 
 

comfy_api_nodes/util/upload_helpers.py

@@ -4,7 +4,7 @@ import logging
 import time
 import uuid
 from io import BytesIO
-from typing import Optional, Union
+from typing import Optional
 from urllib.parse import urlparse
 
 import aiohttp
@@ -48,8 +48,9 @@ async def upload_images_to_comfyapi(
     image: torch.Tensor,
     *,
     max_images: int = 8,
-    mime_type: Optional[str] = None,
+    mime_type: str | None = None,
-    wait_label: Optional[str] = "Uploading",
+    wait_label: str | None = "Uploading",
+    show_batch_index: bool = True,
 ) -> list[str]:
     """
     Uploads images to ComfyUI API and returns download URLs.
@@ -59,11 +60,18 @@ async def upload_images_to_comfyapi(
     download_urls: list[str] = []
     is_batch = len(image.shape) > 3
     batch_len = image.shape[0] if is_batch else 1
+    num_to_upload = min(batch_len, max_images)
+    batch_start_ts = time.monotonic()
 
-    for idx in range(min(batch_len, max_images)):
+    for idx in range(num_to_upload):
         tensor = image[idx] if is_batch else image
         img_io = tensor_to_bytesio(tensor, mime_type=mime_type)
-        url = await upload_file_to_comfyapi(cls, img_io, img_io.name, mime_type, wait_label)
+
+        effective_label = wait_label
+        if wait_label and show_batch_index and num_to_upload > 1:
+            effective_label = f"{wait_label} ({idx + 1}/{num_to_upload})"
+
+        url = await upload_file_to_comfyapi(cls, img_io, img_io.name, mime_type, effective_label, batch_start_ts)
         download_urls.append(url)
     return download_urls
 
@@ -126,8 +134,9 @@ async def upload_file_to_comfyapi(
     cls: type[IO.ComfyNode],
     file_bytes_io: BytesIO,
     filename: str,
-    upload_mime_type: Optional[str],
+    upload_mime_type: str | None,
-    wait_label: Optional[str] = "Uploading",
+    wait_label: str | None = "Uploading",
+    progress_origin_ts: float | None = None,
 ) -> str:
     """Uploads a single file to ComfyUI API and returns its download URL."""
     if upload_mime_type is None:
@@ -148,6 +157,7 @@ async def upload_file_to_comfyapi(
         file_bytes_io,
         content_type=upload_mime_type,
         wait_label=wait_label,
+        progress_origin_ts=progress_origin_ts,
     )
     return create_resp.download_url
 
@@ -155,27 +165,18 @@ async def upload_file_to_comfyapi(
 async def upload_file(
     cls: type[IO.ComfyNode],
     upload_url: str,
-    file: Union[BytesIO, str],
+    file: BytesIO | str,
     *,
-    content_type: Optional[str] = None,
+    content_type: str | None = None,
     max_retries: int = 3,
     retry_delay: float = 1.0,
     retry_backoff: float = 2.0,
-    wait_label: Optional[str] = None,
+    wait_label: str | None = None,
+    progress_origin_ts: float | None = None,
 ) -> None:
     """
     Upload a file to a signed URL (e.g., S3 pre-signed PUT) with retries, Comfy progress display, and interruption.
 
-    Args:
-        cls: Node class (provides auth context + UI progress hooks).
-        upload_url: Pre-signed PUT URL.
-        file: BytesIO or path string.
-        content_type: Explicit MIME type. If None, we *suppress* Content-Type.
-        max_retries: Maximum retry attempts.
-        retry_delay: Initial delay in seconds.
-        retry_backoff: Exponential backoff factor.
-        wait_label: Progress label shown in Comfy UI.
-
     Raises:
         ProcessingInterrupted, LocalNetworkError, ApiServerError, Exception
     """
@@ -198,7 +199,7 @@ async def upload_file(
 
     attempt = 0
     delay = retry_delay
-    start_ts = time.monotonic()
+    start_ts = progress_origin_ts if progress_origin_ts is not None else time.monotonic()
     op_uuid = uuid.uuid4().hex[:8]
     while True:
         attempt += 1