agentscope-ai/agentscope · error · ValueError

Multimodal embedding API only supports URL input for video d

Error message

Multimodal embedding API only supports URL input for video data, got {type(source).__name__}.

What it means

The DashScope multimodal embedding API can only ingest video as a public URL, so _format_data_block rejects video DataBlocks whose source is not URLSource (e.g. Base64Source) before ever calling the API.

Source

Thrown at src/agentscope/embedding/_dashscope/_model.py:512

            block (`DataBlock`):
                A data block with a ``Base64Source`` or ``URLSource``.

        Returns:
            `dict[str, str]`: E.g.
            ``{"image": "data:image/png;base64,..."}`` or
            ``{"video": "https://..."}``.

        Raises:
            `ValueError`: If the media type is unsupported or a video
                block uses a non-URL source.
        """

        source = block.source
        media_type = source.media_type

        if media_type.startswith("video/"):
            if not isinstance(source, URLSource):
                raise ValueError(
                    "Multimodal embedding API only supports URL input "
                    f"for video data, got {type(source).__name__}.",
                )
            return {"video": str(source.url)}

        if media_type.startswith("image/"):
            if isinstance(source, Base64Source):
                return {
                    "image": f"data:{media_type};" f"base64,{source.data}",
                }
            if isinstance(source, URLSource):
                return {"image": str(source.url)}

        raise ValueError(
            f"Unsupported media type {media_type!r} in DataBlock. "
            f"Expected image/* or video/*.",
        )

View on GitHub (pinned to e90f1c7592)

Solutions

1. Host the video at a URL reachable by DashScope and pass DataBlock(source=URLSource(url=...))
2. If you only have local video, upload it to object storage (OSS/S3) first and embed the public/pre-signed URL
3. Extract frames and embed them as images if video URLs are impossible

Example fix

# before
DataBlock(source=Base64Source(media_type="video/mp4", data=b64))
# after
DataBlock(source=URLSource(url="https://cdn.example.com/clip.mp4"))

from agentscope.message import URLSource
if block.source.media_type.startswith(\"video/\") and not isinstance(block.source, URLSource):\n    block.source = URLSource(url=upload(block.source.data))

def video_has_url_source(block) -> bool:\n    from agentscope.message import URLSource\n    return not block.source.media_type.startswith(\"video/\") or isinstance(block.source, URLSource)

try:\n    await model([block])\nexcept ValueError as e:\n    if \"URL input for video\" in str(e): block.source = URLSource(url=host_video(block)); await model([block])\n    else: raise

When it happens

Trigger: Passing a DataBlock with media_type video/* whose source is Base64Source (raw bytes) instead of URLSource.

Common situations: Reusing an image pipeline that base64-encodes local files and pointing it at video files; uploading user-uploaded videos stored locally with no public URL.

Related errors

• Invalid input: {item!r}. Expected str or DataBlock.
• DashScope multimodal embedding API error: {res}
• Unsupported media type {media_type!r} in DataBlock. Expected
• Text embedding model {self.model!r} only accepts str inputs,
• DashScope text embedding API error: {response}

AI-assisted analysis of agentscope-ai/agentscope@e90f1c7592 (2026-08-28). Data as JSON: /api/errors/32e2863a9409b703. Report an issue: GitHub.