Seedance 2.5, Wan 3.0, H3 Max, Kling 3, Omni 1.1: one limits matrix
Five video ids on Sume side by side: length, resolution, aspect ratios, references and audio. Read it before you choose, then confirm on the live catalog.
Of the five ids, only Seedance 2.5 and Wan 3.0 reach 30 seconds in one job. Kling 3 and H3 Max stop at 15, and Omni 1.1 stops at 10. Kling 3 takes no reference inputs, Omni takes no audio reference, and the other three take image, video and audio references. These numbers come from Sume's Video Router catalog, which is the place to confirm them.
The newest model names in the news this season, such as Seedance 2.5, Wan 3.0, MiniMax H3 and Kling 3.0, are ids you can call on Sume. Models that are not in the catalog are not available here, however loudly they are announced.
The matrix
One row per id. Read the notes column for the rule that most often causes a rejected request.
| Id | Length | Resolutions | Aspect ratios | References | Note |
|---|---|---|---|---|---|
| seedance-2.5 | 4 to 30 s | 480p, 720p, 1080p | 8 values incl. 21:9, 4:3 | image, video, audio | No bitrate_mode |
| wan-3.0 | 2 to 30 s | 480p, 720p, 1080p | 7 values incl. 4:3, 3:4 | image, video, audio | Limits 10 / 5 / 5 |
| minimax-h3-max | 5 to 15 s | 480p, 768p, 1080p | 7 values incl. 21:9 | image, video, audio | Native stereo audio, no toggle |
| kling-3 | 4 to 15 s | 720p, 1080p | 16:9, 9:16, 1:1 | none | Sound on costs more |
| gemini-omni-flash-1.1 | 3 to 10 s | 360p to 4K | 16:9, 9:16 | image, video | Native synced audio, no toggle |
How to choose
Start from the one constraint you cannot bend. Need a single shot longer than 15 seconds: Seedance 2.5 or Wan 3.0. Need a square or ultrawide frame: Kling takes 1:1 but not 21:9, while Seedance and H3 Max take 21:9. Need audio references: not Kling, not Omni. Need 4K: Omni is the only row that lists it.
After the hard limits, compare price. Rates differ by a factor of several across these ids, and Seedance 2.5 bills on video tokens rather than flat seconds, so use the catalog quote instead of a mental per-second figure.
Honest tradeoffs
A longer single shot avoids visible seams, but one bad moment in second 22 costs you the whole job. Stitching two 15-second clips lets you redo half. Neither is free, and both carry a continuity risk: faces and props can drift between jobs. Test your actual subject on your two best candidates before you commit to a pipeline.
When you do test, keep a small fixed set of prompts and references, and score each id on the same set. Otherwise you are comparing prompts, not models.
Sources
Related posts
More in Comparisons
- Seedance 2 on Sume vs Veo 3.1 Standard at Google: 720p and 1080p
Seedance 2 on Sume is $0.378 a second at 720p and about $0.85 at 1080p. Google lists Veo 3.1 Standard at $0.40 for both, so 1080p favors Veo on price.
- Sora API ended: which Sume route replaces a talking presenter clip?
OpenAI's Sora API ended 2026-09-24. For a presenter who says exact words, the replacement is a still-plus-audio or avatar route, not a video model.
- Sora API shutdown alternatives: OpenAI lists no replacement
OpenAI's deprecations page removes the Videos API and sora-2 on 2026-09-24 and names no replacement. Pick a Sume video model by clip length and resolution.
- Seedance 2, Omni or Wan 3.0 for 9:16 ads on Sume: how to choose
Three Sume video models side by side for vertical ads after Sora: clip length, resolutions, 9:16 in the docs, references, and the price of a 5 s clip.
Written by Sume