sume/auto vs a pinned video model: three jobs where each wins

Use sume/auto when your clip is 3 to 10 seconds, 16:9 or 9:16, and you do not care which model makes it; pin a model id when you need a longer clip, a specific model or a price you can predict. Auto currently resolves to gemini-omni-flash-1.1, so an 8-second 720p auto job is $1.00.

The video generation docs set the defaults: 8 seconds, 720p, audio always on, and a 400 for generate_audio: false.

Job 1: a quick 8-second clip, auto wins

You send a prompt and nothing else. Auto picks Omni, takes the default 8 seconds at 720p, and bills $1.00. At 360p it is $0.30. This is the case the alias exists for: no catalog reading, and a request body that stays valid if Sume changes the underlying default.

Job 2: a 12-second clip, pin wins

Auto's default row stops at 10 seconds, so a 12-second ask will not fit. Pin wan-3.0 (2 to 30 seconds) and the 720p clip is $1.50, or pin minimax-h3-max (5 to 15 seconds) at 768p for $1.20.

Job 3: a specific look, pin wins

If a client has approved a look from one model, pin it. A pinned id keeps job.model as the id you sent, so your logs and invoices stay consistent when auto changes. Recast and Genjutsu are never auto-routed, so they are always pinned.

Side by side

A rule you can code

If duration is 10 or less and you have no model requirement, send sume/auto. Otherwise pin and check supported_durations in GET /v1/videos/models first. The pin or auto post covers what to do when new ids launch.