AI Gateway async video 413 at 300 KiB: hosted URLs on Sume

Vercel AI Gateway persists the start request for an async video job and caps it at 300 KiB, which includes an 8 KiB allowance for attribution fields. Inline base64 images pass that ceiling quickly and fail with a 413 telling you to use hosted URLs. On Sume, send input media as public HTTPS image URLs; an oversized body returns 413 payload_too_large.

Vercel's limit is from its video generation page and Sume's codes from Errors and credits, read 2026-10-01.

When does the Vercel 413 happen?

The page says it applies when you pass inline base64 in prompt.image, frameImages or inputReferences. Text-to-video is unaffected. The fix it gives is a hosted URL, which "keeps the start request small enough to persist."

What does Sume return for a large request?

413 payload_too_large means the request body exceeds the configured API limit. A body that is not application/json returns 415 unsupported_media_type. The docs I read do not state a byte limit for /v1/videos, so do not assume Vercel's 300 KiB applies; check the 413 response and the docs for the endpoint you call.

How do I send an image by URL on Sume?

Use frame_images with an image_url entry and a frame_type of first_frame or last_frame, as in the video docs. If Sume cannot fetch or mirror the media, the docs say to check that input media is a public HTTPS image URL, then retry or contact support with the request id.

{
  "model": "seedance-2",
  "prompt": "A character walking through a forest",
  "frame_images": [
    {
      "type": "image_url",
      "image_url": { "url": "https://example.com/first-frame.png" },
      "frame_type": "first_frame"
    }
  ]
}

What about 413 from my own proxy?

A 413 can come from a platform in front of the API as well; see Vercel 413 and Sume video. Check the response body for a Sume error code to tell the two apart.