sume/auto for a former Sora feature: when to pin a model

Use sume/auto when users send arbitrary prompts and you care that a clip arrives. Pin a catalog model id when a brand, a series or a test needs one consistent look. The video generation docs say auto responses echo sume/auto and the family that ran is never disclosed, so you cannot build on traits of the output to infer it. Sora gave you one model with one look; auto does not, which matters if a feature was sold on that look.

What auto does and does not promise

The docs describe auto as a Sume-only addition. Resolution is a pure function of the normalized request and the catalog version, so an idempotent replay prices and routes identically. The Video Router docs say auto create controls default to 720p and 8 seconds, with 3 to 10 second clips at 16:9 or 9:16.

Pick by use case

A throwaway preview in an editor, a social post from a free-text box, or a test of whether video is worth adding all suit auto. A recurring series, an advert reviewed by a client, or a feature with tuned prompts suit a pin, because you can test that prompt against that model and keep the result.

• User-generated prompts: start with auto, log the cost.
• Branded or reviewed output: pin and test.
• Needs a longer clip: pin; auto stops at 10 seconds.

Moving from auto to a pin later

Because the poll response does not reveal the family, you cannot read your way from auto to the model that served it. Instead run your golden prompts against two or three pinned ids from GET /v1/videos/models, compare, and set the winner in config. Check supported_durations and supported_aspect_ratios first, since limits differ per model, and treat a prompt tuned on one model as a starting point on another.

OpenAI's deprecations page names no replacement for the Sora models, so this choice is yours to make and to revisit.