sgl-project/sglang · error · HTTPException

Failed to process image source: {str(e)}

Error message

Failed to process image source: {str(e)}

What it means

Raised by the mesh generation endpoint when saving the uploaded image to the server's uploads directory fails. The endpoint wraps save_image_to_path in a try/except and converts any exception (I/O error, invalid file object, unwritable directory) into an HTTP 400 with the underlying message.

Source

Thrown at python/sglang/multimodal_gen/runtime/entrypoints/openai/mesh_api.py:170

        urls = url or url_array
        image_list = merge_image_input_list(images, urls)

        if not image_list:
            raise HTTPException(
                status_code=422,
                detail="Field 'image' or 'url' is required for mesh generation",
            )

        uploads_dir = os.path.join("outputs", "uploads")
        os.makedirs(uploads_dir, exist_ok=True)
        img = image_list[0]
        filename = img.filename if hasattr(img, "filename") else "input_image"
        try:
            input_path = await save_image_to_path(
                img, os.path.join(uploads_dir, f"{request_id}_{filename}")
            )
        except Exception as e:
            raise HTTPException(
                status_code=400, detail=f"Failed to process image source: {str(e)}"
            )

        req = MeshGenerationsRequest(
            prompt=prompt or "generate 3d mesh",
            model=model,
            seed=seed,
            generator_device=generator_device,
            num_inference_steps=num_inference_steps,
            negative_prompt=negative_prompt,
            output_format=output_format,
            **(
                {"guidance_scale": guidance_scale} if guidance_scale is not None else {}
            ),
        )
    else:
        try:
            body = await request.json()

View on GitHub (pinned to 0132848349)

Solutions

1. Verify the uploaded file is a valid image (open it client-side before sending)
2. Check the server's uploads directory exists and is writable by the server process
3. Retry with a smaller/standard format (PNG/JPEG) image to rule out decode issues

Example fix

// before
files = {"image": ("mesh.txt", open("mesh.txt","rb"), "text/plain")}
// after
files = {"image": ("input.png", open("input.png","rb"), "image/png")}

from PIL import Image
img = Image.open(path); img.verify()  # raises if not a valid image

try:
    resp = await client.post('/v1/mesh/generations', files=files)
except httpx.HTTPStatusError as e:
    if e.response.status_code == 400 and 'image source' in e.response.text:
        # re-encode image and retry once
        ...

When it happens

Trigger: POST to the mesh generations endpoint with a multipart image that cannot be read/saved: corrupted UploadFile stream, missing uploads directory, or filesystem permission errors on the server.

Common situations: Server container running as non-root without write access to the uploads dir; client sending a non-file field where an image is expected; disk full.

Related errors

• Invalid request body: {e}
• An input image is required for mesh generation
• Mesh not found
• Mesh has been uploaded to cloud storage. Please use the clou
• Generation is still in-progress

AI-assisted analysis of sgl-project/sglang@0132848349 (2026-08-28). Data as JSON: /api/errors/c275169fdb64d4ff. Report an issue: GitHub.