Silent clip for your own music: sume/auto rejects audio off

If you send generate_audio: false with model: "sume/auto", Sume answers 400 unsupported_capability and submits nothing. Auto defaults to Gemini Omni Flash 1.1, where native audio is always on, and the docs say Sume never silently routes such a request elsewhere. To get a silent clip for your own music bed, name a catalog model whose generate_audio field allows it.

What auto promises

sume/auto is Sume-only. Its resolution is a pure function of the normalized request plus the catalog version. There is no load balancing and no A/B test, so an idempotent replay gets the same route and the same price. The default target for text, first and end frame, and reference requests is gemini-omni-flash-1.1.

The validation envelope is that model's. Duration is 3 to 10 seconds with a default of 8. Resolution is 360p, 720p, 1080p or 4K with a default of 720p. Aspect is 16:9 or 9:16.

Requests that fail closed

Unsupported constraints fail with 400 unsupported_capability and no provider submission. The docs list generate_audio: false, 2 or 11 seconds, and 480p or 768p as examples. The error names sume/auto, not the resolved model, and supported still lists the values the API accepts.

The fix

Read GET /v1/videos/models and look at each row's generate_audio value. Choose a model that supports a silent output, name it in model, and keep your music from the Music Router or your own file as the Timeline audio spine. The Music Router bills a fixed Music price per generation.

Do not fall back by catching the 400 and retrying on auto. The retry would fail again with the same message.

curl "https://api.sume.com/v1/videos/models" \
  -H "Authorization: Bearer $SUME_API_KEY"

Keep one policy in code

Decide per campaign whether the clip carries its own sound. If it does, use auto or Omni and leave generate_audio out. If the bed is yours, pin the model and render the final cut on Timeline with your track.