AI workout video generator: Seedance 2.5, 4 to 30 seconds
For a workout demo, one Seedance 2.5 request runs 4 to 30 seconds on Sume, with video and audio references; most other catalog models stop at 15 seconds.

For a workout demo, pick the clip length first. On Sume, seedance-2.5 accepts 4 to 30 seconds in one request and honors video and audio references, while most other catalog models top out at 15 seconds (wan-3.0 also reaches 30). A set of one exercise fits comfortably inside 30 seconds; a full routine is several clips.
Limits are from the video generation docs; vendor capabilities from the Seedance 2.5 post, both read 2026-10-01.
How long can one workout clip be?
The catalog endpoint reports supported durations per model, and the docs say limits are not uniform.
| Model | Duration window | Fits |
|---|---|---|
seedance-2.5 | 4 to 30 seconds | One exercise with a few reps, or a short cue-and-demo |
wan-3.0 | 2 to 30 seconds | Same, shorter minimum |
| Most other catalog models | Up to 15 seconds | A single rep set |
How do I keep the same movement in the clip?
Send a reference. Audio and video references are honored by the Seedance 2.x models, Wan 3.0 and the MiniMax H3 family, so a short clip of the move you want can guide the result. Seed's post says Seedance 2.5 takes up to 30 images, 10 video clips and 10 audio clips as references in one pass; Sume's own accepted counts are in the docs and the catalog entry, so check those before assuming the vendor's numbers.
What about a routine longer than 30 seconds?
Generate one clip per exercise and join them, or extend from a last frame; see Seedance 2.5 multi-round extension. Treat each exercise as its own request so a bad form render only costs one clip.
Is generated exercise form safe to publish?
A generated body can show a movement that is wrong or unsafe, and no model guarantees correct form. Review each clip against a qualified source before it ships, and say in the video that it is generated if your audience would expect real footage.
Sources
Related posts
More in Use cases
- Seedance real-face reference images rejected: what to do
BytePlus says Seedance 2.5 and 2.0 do not accept reference images or videos with real human faces. The exceptions, and a safe reference plan on Sume.
- Sepia or blue tint filter for video by API
Tint a clip sepia or blue with colorchannelmixer, colorbalance or monochrome in Sume's video filter. Presets and .cube LUT files are not available.
- Sharpen video after an upscale: unsharp or cas filter API
Sharpen an upscaled clip with the unsharp or cas filter in Sume's video filter: one MP4 in, one new MP4 out, a flat per-job rate, and a free check first.
- Shopify AI image editor scenes: one at a time vs n
Shopify's editor generates one scene at a time and deletes unsaved ones on close. On Sume, ask for several scenes in one call with n and keep the job results.
Written by Sume