Sume Auto video: which model ran? Not disclosed, so log this

When you send model: "sume/auto" to /v1/videos, Sume does not tell you which model family served the request. The poll response reports sume/auto, and the docs say you should not infer the family from any trait of the output. If you must know which model made a clip, pin a catalog id such as seedance-2.5; if you accept Auto, log the request and the job instead.

This matters when a vendor launches a new model and a client asks whether the new one made their ad. With Auto you cannot answer that from Sume's side, so decide before you submit whether you want convenience or provenance. Facts below come from Sume's video generation docs (read 2026-10-02).

What does Auto promise?

Auto picks a family for you, and resolution is a pure function of the normalized request and the catalog version. That makes an idempotent replay price and route identically. The defaults are 720p and 8 seconds, with 3 to 10 second clips at 16:9 or 9:16.

What it does not promise is a named model. The response echoes sume/auto and the served family is never disclosed.

What should I log for every Auto clip?

Log what you control, because that is what you can reproduce and defend later.

• The exact request body, including model, prompt, duration, resolution and aspect_ratio.
• The job id from the submit response and your Idempotency-Key.
• The date and the response's usage.cost, which is the Sume billable amount.
• Your own metadata if you use the legacy route; it is stored on the job and not sent to the provider.

When should I pin instead?

Pin when a contract, a brand rule, a comparison test or a launch claim depends on the model name. Pin also when you need a length or ratio outside Auto's envelope, such as 30 seconds, because Auto's create range is 3 to 10 seconds.

Keep Auto for drafts and placeholder clips where the choice of family does not matter. A hybrid works well: Auto while you iterate on the idea, a pinned id for the clip that ships.