Which video models does Sume list in October 2026? Twelve ids

Sume's video catalog has twelve ids, from seedance-2.5 at 30 seconds to gemini-omni-flash-1.1 at 4K. Length and resolution for each; Veo 3.1 is not listed.

5 min readSume
All posts

Sume lists twelve video model ids behind one POST /v1/videos route: higgsfield-genjutsu, seedance-2.5, seedance-2-mini, seedance-2, seedance-2-fast, kling-3, wan-3.0, grok-imagine-video-1.5, minimax-h3, minimax-h3-max, gemini-omni-flash-1.1 and h3-max-recast. Veo 3.1 is not one of them. You pick the id in the model field, and the request shape stays the same.

Why this list matters now

After the Sora API shutdown on September 24, 2026 (OpenAI's deprecations page lists no replacement), most teams need to know which of these ids covers the length and resolution they used. The table gives the documented limits, and the first column is the exact string to send.

Counts matter for planning: twelve ids, but one of them appears only when its provider is configured, and one is a specialist for replacing people in an existing video. For general text or image to video work you are choosing among about ten.

The catalog

Sume video catalog limits (read 2026-10-07)
Model idLengthResolutionNotes
seedance-2.54 to 30 s480p, 720p, 1080pLongest Seedance
seedance-2, -fast, -mini4 to 15 sSee the video docsThree tiers of the same family
kling-34 to 15 s720p, 1080p16:9, 9:16, 1:1, no references
wan-3.02 to 30 s480p, 720p, 1080pReferences: 10 images, 5 videos, 5 audios
grok-imagine-video-1.54 to 15 s480p, 720pimage_url required, no audio
minimax-h35 to 15 s480p, 768p nativeNative stereo audio, 2K and 4K are upscales of 768p
minimax-h3-max5 to 15 s480p, 768p, 1080p1080p is a latent refinement of 768p
gemini-omni-flash-1.13 to 10 s360p, 720p, 1080p, 4K16:9 or 9:16, native audio always on
h3-max-recastSource videoSee the video docsReplaces people in a source video
higgsfield-genjutsuSee the video docsSee the video docsListed only when its provider is configured

Picking by constraint

Pick by constraint, not by name. A 25 second clip leaves two ids: seedance-2.5 and wan-3.0. A 4K clip leaves one, gemini-omni-flash-1.1, with a ceiling of 10 seconds. A 2 second bumper is wan-3.0 only.

  • Longer than 15 seconds: seedance-2.5 or wan-3.0.
  • 4K output: gemini-omni-flash-1.1.
  • Start from a still and spend little: grok-imagine-video-1.5.
  • Swap the people in an existing video: h3-max-recast.
  • Edit an existing video with a prompt: gemini-omni-flash-1.1 with video_url.

The auto id

sume/auto routes to gemini-omni-flash-1.1, so a request with no model preference lands on the one id that does 4K and native audio. Ask for it explicitly if you want the run to stay stable when the default changes.

If you route programmatically, keep a small table in your code from use case to model id, and read the live list from GET /v1/videos/models at startup and compare it with your table, so an id that is no longer offered fails in your check instead of in a customer's request.

What the route refuses

The catalog also tells you what the route refuses: size, seed and non-empty provider.options return 400 unsupported_parameter. Read the full field rules in the video docs, and use this id-by-id comparison for references.

Sources

Related posts

More in Models

All Models posts

Written by Sume