AI video resolution on Sume: 360p, 480p, 720p, 768p, 1080p or 4K

Sume's video catalog does not use one resolution ladder. Gemini Omni Flash 1.1 goes from 360p to 4K, Wan 3.0 and Seedance stop at 1080p, and the MiniMax models have a native 768p tier that is not the same as 720p. Pick the tier from the model's own list in supported_resolutions.

Who takes what

• gemini-omni-flash-1.1: 360p, 720p, 1080p and 4K, with 16:9 or 9:16.
• wan-3.0, seedance-2.5, seedance-2: 480p, 720p and 1080p.
• minimax-h3: native 480p and 768p. Sume bills 2K and 4K upscales if your request includes them.
• minimax-h3-max: 480p, 768p and 1080p; 1080p is a latent refinement from native 768p.
• h3-max-recast: 768p and 1080p. higgsfield-genjutsu: 480p and 720p, when its provider is configured.

Price per second at each tier

The four models below are priced per second. Seedance is priced per video token, so it is left out.

Reading the table

The cheapest tier overall is Omni at 360p, and the cheapest per second at 1080p is Omni too. Wan 3.0 doubles in price at each step from 480p to 1080p. H3 and H3 Max sit between 480p and 768p with a small step. The docs call 768p first-class for H3, not 720p, so do not ask minimax-h3 for 720p; use 768p.

Docs also state that higher resolution takes more time and has a higher price, so use it where the extra pixels show. A common pattern is to render a draft at the lowest tier and re-render the keepers. Omni 360p at 3 seconds costs $0.12 against $0.57 at 1080p.

Resolution and generation time

The docs say a higher resolution takes more time to generate and has a higher price. Video usually takes from 30 seconds to several minutes, depending on the model and the parameters. So resolution costs you twice: more dollars and a longer wait. If you poll in a user-facing flow, budget for the 4K case, and prefer callback_url for long jobs.

Omni 4K costs exactly twice 1080p per second in the catalog: $0.375 against $0.1875. A 4K clip is worth it when the output will be shown on a large screen or cropped hard in post. For a feed thumbnail, it is not.

Before you ship anything, read the live pages again: the catalog is public, the pricing page is public, and the docs describe the request fields. A blog post is a snapshot. The catalog, the plan grid and the error table are the things that change, so write your code to read them instead of copying numbers from a page, and re-check when a new model is added.

A good habit is a small log line per submit with the model, resolution, duration, estimated cost, job id and the Idempotency-Key you used. When a job misbehaves, those six fields answer most of the questions support will ask, and they let you compare your estimate with usage.cost and the usage ledger without re-running anything.

If you are new to the API, start with one clip, one model and the lowest resolution, read the full response once, and only then build a loop around it. Most surprises with video jobs come from fields that were defaulted, not from fields that were set.