ATH-MaaS/Pixelle-Video · error · ValueError

无法解析 base64 图片: {e}

Error message

无法解析 base64 图片: {e}

What it means

_to_dashscope_file_url wraps any exception raised while decoding/writing a base64 image data URL into ValueError('无法解析 base64 图片: {e}'). The decode, mime/header parsing, or temp-file write failed.

Source

Thrown at pixelle_video/services/api_services/vlm_client.py:89

        if path.startswith("data:"):
            if not allow_data_url:
                raise ValueError("DashScope video input does not support data URLs in this adapter.")

            import base64 as b64
            import tempfile

            try:
                header, b64_data = path.split(",", 1)
                mime_type = header.split(";")[0].replace("data:", "")
                image_data = b64.b64decode(b64_data)
                suffix = f".{mime_type.split('/')[-1]}" if "/" in mime_type else ".png"
                with tempfile.NamedTemporaryFile(suffix=suffix, delete=False) as tmp:
                    tmp.write(image_data)
                    temp_path = tmp.name
                return f"file://{os.path.abspath(temp_path)}"
            except Exception as e:
                print(f"Error processing base64 image: {e}")
                raise ValueError(f"无法解析 base64 图片: {e}")

        if path.startswith("http") or path.startswith("file://"):
            return path

        return f"file://{os.path.abspath(path)}"

View on GitHub (pinned to 848b054e4f)

Solutions

1. Validate the data URL format before calling: starts with 'data:image/', contains ';base64,', and the payload is valid base64 (try base64.b64decode(payload, validate=True))
2. Strip whitespace/newlines from the base64 payload and re-encode URL-safe base64 to standard
3. Check the error suffix ({e}) in the message — it names the underlying cause (binascii.Error, OSError, etc.)
4. Bypass the data URL path: save the image to a file and pass the local path instead

Example fix

// before
images=[f"data:image/png;base64,{raw_b64}"]  # raw_b64 may contain newlines
// after
import base64
clean = raw_b64.replace("\n", "").replace("\r", "").replace("-", "+").replace("_", "/")
base64.b64decode(clean, validate=True)  # fail fast with a clear error
images=[f"data:image/png;base64,{clean}"]

import base64, re
def validate_image_data_url(p: str) -> str:
    m = re.fullmatch(r"data:image/(png|jpe?g);base64,([A-Za-z0-9+/=\s]+)", p)
    if not m:
        raise ValueError("not a valid image data URL")
    payload = "".join(m.group(2).split())
    base64.b64decode(payload, validate=True)
    return p

def is_valid_base64(s: str) -> bool:
    import base64
    try:
        base64.b64decode("".join(s.split()), validate=True)
        return True
    except Exception:
        return False

try:
    answer = vlm.query(prompt=p, image_paths=[img], model="qwen-vl-max")
except ValueError as e:
    if "base64" in str(e):
        logging.error(f"Malformed base64 image input: {e}")  # fix encoding upstream or write to a file
    else:
        raise

When it happens

Trigger: Passing an image_paths entry like 'data:image/png;base64,...' whose base64 payload is malformed, header lacks a mime type/extension the parser understands, or whose decoded bytes cannot be written to the temp file (disk full, permissions).

Common situations: Truncated base64 string (cut off during serialization); whitespace/newlines inside base64; data URL missing the 'data:...;base64,' prefix; URL-safe base64 (-, _) from another service without re-encoding; no space in /tmp.

Related errors

• DashScope video input does not support data URLs in this ada
• DashScope reference-to-video models require at least one ref
• DashScope video edit models require video input and may use
• DashScope legacy video models require image_path.
• Invalid DashScope media combination: {'+'.join(sorted(media_

AI-assisted analysis of ATH-MaaS/Pixelle-Video@848b054e4f (2026-08-30). Data as JSON: /api/errors/8ba6d388237a48c6. Report an issue: GitHub.