pull · Sep 18, 2026 · Sep 18, 2026
diff --git a/comfy_api_nodes/apis/openrouter.py b/comfy_api_nodes/apis/openrouter.py
@@ -104,6 +104,7 @@ class OpenRouterChatResponse(BaseModel):
 
 class OpenRouterImageData(BaseModel):
     b64_json: str | None = Field(None)
+    url: str | None = Field(None)
     media_type: str | None = Field(None)
 
 

diff --git a/comfy_api_nodes/nodes_bytedance.py b/comfy_api_nodes/nodes_bytedance.py
@@ -3362,6 +3362,7 @@ async def execute(
             cls,
             ApiEndpoint(path="/proxy/byteplus/api/v3/tts/create", method="POST"),
             response_model=SeedAudioResponse,
+            asset_urls=True,
             data=SeedAudioRequest(
                 model=model,
                 text_prompt=text_prompt,
@@ -3374,11 +3375,15 @@ async def execute(
                 ),
             ),
         )
-        if not response.audio:
+        if response.audio:
+            audio_bytes = base64.b64decode(response.audio)
+        elif response.url:
+            audio_bytes = (await download_url_as_bytesio(response.url, cls=cls)).getvalue()
+        else:
             raise Exception(
                 f"Seed Audio returned no audio (code={response.code}): {response.message}"
             )
-        return IO.NodeOutput(audio_bytes_to_audio_input(base64.b64decode(response.audio)))
+        return IO.NodeOutput(audio_bytes_to_audio_input(audio_bytes))
 
 
 _VCUBE_ENHANCE_VIDEO_ENDPOINT = ApiEndpoint(path="/proxy/byteplusmediakit/api/v1/tools/enhance-video", method="POST")

diff --git a/comfy_api_nodes/nodes_elevenlabs.py b/comfy_api_nodes/nodes_elevenlabs.py
@@ -404,6 +404,7 @@ async def execute(
             ),
             data=request,
             as_binary=True,
+            asset_urls=True,
         )
         return IO.NodeOutput(audio_bytes_to_audio_input(response))
 
@@ -449,6 +450,7 @@ async def execute(
             files={"audio": ("audio.mp4", audio_bytes_io, "audio/mp4")},
             content_type="multipart/form-data",
             as_binary=True,
+            asset_urls=True,
         )
         return IO.NodeOutput(audio_bytes_to_audio_input(response))
 
@@ -545,6 +547,7 @@ async def execute(
                 loop=model.get("loop", None),
             ),
             as_binary=True,
+            asset_urls=True,
         )
         return IO.NodeOutput(audio_bytes_to_audio_input(response))
 
@@ -762,6 +765,7 @@ async def execute(
             files={"audio": ("audio.mp4", audio_bytes_io.getvalue(), "audio/mp4")},
             content_type="multipart/form-data",
             as_binary=True,
+            asset_urls=True,
         )
         return IO.NodeOutput(audio_bytes_to_audio_input(response))
 
@@ -901,6 +905,7 @@ async def execute(
             ),
             data=request,
             as_binary=True,
+            asset_urls=True,
         )
         return IO.NodeOutput(audio_bytes_to_audio_input(response))
 

diff --git a/comfy_api_nodes/nodes_fishaudio.py b/comfy_api_nodes/nodes_fishaudio.py
@@ -293,6 +293,7 @@ async def execute(
             ),
             data=request,
             as_binary=True,
+            asset_urls=True,
         )
         return IO.NodeOutput(audio_bytes_to_audio_input(response))
 

diff --git a/comfy_api_nodes/nodes_gemini.py b/comfy_api_nodes/nodes_gemini.py
@@ -8,6 +8,7 @@
 from fnmatch import fnmatch
 from io import BytesIO
 from typing import Any, Literal
+from urllib.parse import urlparse
 
 import torch
 from typing_extensions import override
@@ -277,8 +278,12 @@ async def get_video_from_interaction(
     )
 
 
+GEMINI_FILES_HOST = "generativelanguage.googleapis.com"
+
+
 async def download_interaction_video(uri: str, cls: type[IO.ComfyNode] | None = None) -> InputImpl.VideoFromFile:
-    if "/files/" not in uri:
+    parsed = urlparse(uri)
+    if parsed.netloc != GEMINI_FILES_HOST or "/files/" not in parsed.path:
         return await download_url_to_video_output(uri, cls=cls)
     name = uri.split("?", 1)[0].rsplit("/files/", 1)[-1].split(":", 1)[0]
     await poll_op(
@@ -1921,6 +1926,7 @@ async def execute(cls, model: dict, seed: int) -> IO.NodeOutput:
                 ),
             ),
             response_model=GeminiInteraction,
+            asset_urls=True,
         )
         if interaction.status != "completed":
             model_message = get_text_from_interaction(interaction).strip()
@@ -2181,6 +2187,7 @@ async def execute(cls, model: dict) -> IO.NodeOutput:
                 response_format=response_format,
             ),
             response_model=GeminiInteraction,
+            asset_urls=True,
         )
         if interaction.status != "completed":
             model_message = get_text_from_interaction(interaction).strip()

diff --git a/comfy_api_nodes/nodes_heygen.py b/comfy_api_nodes/nodes_heygen.py
@@ -1,5 +1,3 @@
-import uuid
-
 import torch
 from typing_extensions import override
 
@@ -70,7 +68,7 @@ async def _create_and_poll_video(cls: type[IO.ComfyNode], payload: dict) -> dict
     """POST a /v3/videos payload, poll until terminal, and return the final video data."""
     created = await sync_op_raw(
         cls,
-        ApiEndpoint(path=_VIDEOS_PATH, method="POST", headers={"Idempotency-Key": uuid.uuid4().hex}),
+        ApiEndpoint(path=_VIDEOS_PATH, method="POST"),
         data=payload,
     )
     video_id = (created.get("data") or {}).get("video_id")

diff --git a/comfy_api_nodes/nodes_ideogram.py b/comfy_api_nodes/nodes_ideogram.py
@@ -356,7 +356,6 @@ async def execute(
                 data=edit_request,
                 files=files,
                 content_type="multipart/form-data",
-                max_retries=1,
             )
 
         elif image is not None or mask is not None:
@@ -400,7 +399,6 @@ async def execute(
                 data=gen_request,
                 files=files if files else None,
                 content_type="multipart/form-data",
-                max_retries=1,
             )
 
         if not response.data or len(response.data) == 0:
@@ -514,7 +512,6 @@ async def execute(
                 resolution=resolution.split(" ")[0] if resolution != "Auto" else None,
                 rendering_speed=rendering_speed,
             ),
-            max_retries=1,
         )
 
         if not response.data or len(response.data) == 0:
@@ -649,7 +646,6 @@ async def execute(
             ApiEndpoint(path="/proxy/ideogram/text-to-image/p-image-ideogram", method="POST"),
             response_model=IdeogramGenerateResponse,
             data=request,
-            max_retries=1,
         )
         if not response.data:
             raise Exception("No images were generated in the response")

diff --git a/comfy_api_nodes/nodes_krea.py b/comfy_api_nodes/nodes_krea.py
@@ -35,7 +35,6 @@ async def _upload_image_to_krea_assets(cls: type[IO.ComfyNode], image: Input.Ima
         response_model=KreaAssetResponse,
         files=[("file", (img_io.name, img_io, "image/png"))],
         content_type="multipart/form-data",
-        max_retries=1,
         wait_label="Uploading reference",
     )
     return response.image_url

diff --git a/comfy_api_nodes/nodes_ltxv.py b/comfy_api_nodes/nodes_ltxv.py
@@ -59,7 +59,6 @@ async def _v25_submit_and_poll(cls: type[IO.ComfyNode], route: str, data: BaseMo
         ApiEndpoint(f"/proxy/ltx/v2/{route}", "POST"),
         response_model=Ltx25SubmitResponse,
         data=data,
-        max_retries=1,
     )
     job = await poll_op(
         cls,

diff --git a/comfy_api_nodes/nodes_meta.py b/comfy_api_nodes/nodes_meta.py
@@ -16,6 +16,8 @@
 from comfy_api_nodes.util import (
     ApiEndpoint,
     bytesio_to_image_tensor,
+    download_url_to_image_tensor,
+    pad_images_to_common_channels,
     sync_op,
     upload_images_to_comfyapi,
     validate_string,
@@ -57,15 +59,16 @@ def _size(aspect_ratio: str) -> str | None:
     return None if aspect_ratio == "auto" else aspect_ratio.replace(":", "x")
 
 
-def _decode_images(response: MuseImageResponse) -> torch.Tensor:
-    images = [
-        bytesio_to_image_tensor(BytesIO(base64.b64decode(item.b64_json)))
-        for item in response.data
-        if item.b64_json
-    ]
+async def _decode_images(cls: type[IO.ComfyNode], response: MuseImageResponse) -> torch.Tensor:
+    images = []
+    for item in response.data:
+        if item.b64_json:
+            images.append(bytesio_to_image_tensor(BytesIO(base64.b64decode(item.b64_json))))
+        elif item.url:
+            images.append(await download_url_to_image_tensor(item.url, cls=cls))
     if not images:
         raise Exception("The response contains no images.")
-    return torch.cat(images)
+    return torch.cat(pad_images_to_common_channels(images))
 
 
 def _reasoning_strength_input() -> IO.Combo.Input:
@@ -220,6 +223,7 @@ async def execute(cls, model: dict):
             cls,
             ApiEndpoint(path=GENERATIONS_PATH, method="POST"),
             response_model=MuseImageResponse,
+            asset_urls=True,
             data=MuseImageRequest(
                 model=model["model"],
                 prompt=model["prompt"],
@@ -228,7 +232,7 @@ async def execute(cls, model: dict):
                 tool_enablement=_tool_enablement(model),
             ),
         )
-        return IO.NodeOutput(_decode_images(response))
+        return IO.NodeOutput(await _decode_images(cls, response))
 
 
 class MetaMuseImageEditApi(IO.ComfyNode):
@@ -280,6 +284,7 @@ async def execute(cls, model: dict):
             cls,
             ApiEndpoint(path=EDITS_PATH, method="POST"),
             response_model=MuseImageResponse,
+            asset_urls=True,
             data=MuseImageEditRequest(
                 model=model["model"],
                 prompt=prompt,
@@ -289,7 +294,7 @@ async def execute(cls, model: dict):
                 images=[MuseImageInput(image_url=url) for url in urls],
             ),
         )
-        return IO.NodeOutput(_decode_images(response))
+        return IO.NodeOutput(await _decode_images(cls, response))
 
 
 class MetaApiExtension(ComfyExtension):

diff --git a/comfy_api_nodes/nodes_openai.py b/comfy_api_nodes/nodes_openai.py
@@ -84,12 +84,15 @@ class SupportedOpenAIModel(str, Enum):
 }
 
 
-async def validate_and_cast_response(response, timeout: int = None) -> torch.Tensor:
+async def validate_and_cast_response(
+    response, timeout: int = None, cls: type[IO.ComfyNode] = None
+) -> torch.Tensor:
     """Validates and casts a response to a torch.Tensor.
 
     Args:
         response: The response to validate and cast.
         timeout: Request timeout in seconds. Defaults to None (no timeout).
+        cls: The calling node class; required for relative `/proxy/` URLs so they can be expanded and authenticated.
 
     Returns:
         A torch.Tensor of shape (N, H, W, C) with all returned images; images whose
@@ -112,7 +115,7 @@ async def validate_and_cast_response(response, timeout: int = None) -> torch.Ten
             img_io = BytesIO(base64.b64decode(img_data.b64_json))
         elif img_data.url:
             img_io = BytesIO()
-            await download_url_to_bytesio(img_data.url, img_io, timeout=timeout)
+            await download_url_to_bytesio(img_data.url, img_io, timeout=timeout, cls=cls)
         else:
             raise ValueError("Invalid image payload – neither URL nor base64 data present.")
 
@@ -367,6 +370,7 @@ async def execute(
                 cls,
                 ApiEndpoint(path="/proxy/openai/images/edits", method="POST"),
                 response_model=OpenAIImageGenerationResponse,
+                asset_urls=True,
                 data=OpenAIImageEditRequest(
                     model=model,
                     prompt=prompt,
@@ -385,6 +389,7 @@ async def execute(
                 cls,
                 ApiEndpoint(path="/proxy/openai/images/generations", method="POST"),
                 response_model=OpenAIImageGenerationResponse,
+                asset_urls=True,
                 data=OpenAIImageGenerationRequest(
                     model=model,
                     prompt=prompt,
@@ -396,7 +401,7 @@ async def execute(
                     moderation="low",
                 ),
             )
-        return IO.NodeOutput(await validate_and_cast_response(response))
+        return IO.NodeOutput(await validate_and_cast_response(response, cls=cls))
 
 
 GPT_IMAGE_QUALITIES = ("low", "medium", "high")
@@ -732,6 +737,7 @@ async def execute(
                 cls,
                 ApiEndpoint(path="/proxy/openai/images/edits", method="POST"),
                 response_model=OpenAIImageGenerationResponse,
+                asset_urls=True,
                 data=OpenAIImageEditRequest(
                     model=model_id,
                     prompt=prompt,
@@ -749,6 +755,7 @@ async def execute(
                 cls,
                 ApiEndpoint(path="/proxy/openai/images/generations", method="POST"),
                 response_model=OpenAIImageGenerationResponse,
+                asset_urls=True,
                 data=OpenAIImageGenerationRequest(
                     model=model_id,
                     prompt=prompt,
@@ -759,7 +766,7 @@ async def execute(
                     moderation="low",
                 ),
             )
-        return IO.NodeOutput(await validate_and_cast_response(response))
+        return IO.NodeOutput(await validate_and_cast_response(response, cls=cls))
 
 
 class OpenAIChatNode(IO.ComfyNode):

diff --git a/comfy_api_nodes/nodes_openrouter.py b/comfy_api_nodes/nodes_openrouter.py
@@ -29,6 +29,7 @@
 from comfy_api_nodes.util import (
     ApiEndpoint,
     bytesio_to_image_tensor,
+    download_url_to_image_tensor,
     get_number_of_images,
     pad_images_to_common_channels,
     sync_op,
@@ -279,16 +280,20 @@ def _extract_text(response: OpenRouterChatResponse) -> str:
     return message.content or ""
 
 
-def _image_data_to_tensor(item: OpenRouterImageData) -> torch.Tensor:
-    try:
-        return bytesio_to_image_tensor(BytesIO(base64.b64decode(item.b64_json)))
-    except Exception as e:
-        raise ValueError(f"OpenRouter returned an image that could not be decoded: {e}") from e
+async def _image_data_to_tensor(cls: type[IO.ComfyNode], item: OpenRouterImageData) -> torch.Tensor:
+    if item.b64_json:
+        try:
+            return bytesio_to_image_tensor(BytesIO(base64.b64decode(item.b64_json)))
+        except Exception as e:
+            raise ValueError(f"OpenRouter returned an image that could not be decoded: {e}") from e
+    if item.url:
+        return await download_url_to_image_tensor(item.url, cls=cls)
+    raise ValueError("OpenRouter returned an image with neither inline data nor a URL.")
 
 
-def _extract_images(response: OpenRouterImageResponse) -> torch.Tensor:
+async def _extract_images(cls: type[IO.ComfyNode], response: OpenRouterImageResponse) -> torch.Tensor:
     _raise_on_error(response.error)
-    tensors = [_image_data_to_tensor(item) for item in response.data or [] if item.b64_json]
+    tensors = [await _image_data_to_tensor(cls, item) for item in response.data or [] if item.b64_json or item.url]
     if not tensors:
         raise ValueError("OpenRouter returned no image.")
     return torch.cat(pad_images_to_common_channels(tensors))
@@ -617,6 +622,7 @@ async def execute(cls, model: dict) -> IO.NodeOutput:
             cls,
             ApiEndpoint(path=OPENROUTER_IMAGES_ENDPOINT, method="POST"),
             response_model=OpenRouterImageResponse,
+            asset_urls=True,
             data=OpenRouterImageRequest(
                 model=slug,
                 prompt=prompt,
@@ -625,7 +631,7 @@ async def execute(cls, model: dict) -> IO.NodeOutput:
                 input_references=input_references,
             ),
         )
-        return IO.NodeOutput(_extract_images(response))
+        return IO.NodeOutput(await _extract_images(cls, response))
 
 
 class OpenRouterExtension(ComfyExtension):