Seedance 2.5 and Wan 3.0 claim 30 s: what Sume's ids accept

ByteDance lists up to 30 seconds per generation for Seedance 2.5, and Alibaba lists native 30-second generation for Wan 3.0. On Sume, the ids seedance-2.5 and wan-3.0 both reach 30 seconds, but they start at different minimums: 4 seconds for Seedance and 2 seconds for Wan.

That makes them the two catalog ids that beat Luma's 20-second Ray 3.2 clip length, which Sume does not run, in one job.

What each vendor says

The Seedance 2.5 post from ByteDance Seed (read 2026-10-05) describes audio-video joint generation, up to 30 seconds per generation with multiple rounds of extension, and per pass up to 30 images, 10 video clips and 10 audio clips. It also lists timestamp-level editing, green-screen background replacement and camera perspective change.

The Wan 3.0 repository (read 2026-10-05) lists native 30-second generation, up to 20 reference assets, instruction-based video editing, auto scene splitting and text rendering in 12 languages.

Those are vendor feature lists. Sume's docs describe what its ids accept, and the two lists do not match one to one.

What Sume's docs say

The Video generation docs state that seedance-2.5 accepts 4 to 30 seconds at 480p, 720p and 1080p, and wan-3.0 accepts 2 to 30 seconds. The Seedance 2.x models, Wan 3.0, MiniMax H3 and MiniMax H3 Max accept audio and video references. Always read supported_durations and supported_resolutions from GET /v1/videos/models before you send a request, because each model has its own envelope.

How to pick between them

If you need a very short clip, Wan's 2-second floor is lower. If you need a catalog id that documents 1080p, Seedance 2.5's list names it. For a 30-second single take with audio and video references, either id fits, and the cheaper one on your target resolution wins. Prices are per model and resolution; read them from GET /v1/videos/models (pricing_skus) rather than from a blog post.

Sume bills the provider list times 1.25 for every model, and reserves that amount at submit. The poll response's usage.cost is the billable amount.

A request that tests the ceiling

A 30-second job is the longest single clip you can ask for on these ids. Use async mode or a callback, because a long clip can take several minutes.

curl -X POST https://api.sume.com/v1/videos \
  -H "Authorization: Bearer $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: ceiling-test-001" \
  -d '{"model":"wan-3.0","prompt":"One continuous take through a night market",
       "duration":30,"aspect_ratio":"16:9"}'

Past 30 seconds

Sume has no extension endpoint for these ids. For more than 30 seconds, extract the last still with video frames, send it as the next first_frame, and join the clips in Timeline 1.0, which renders up to 1800 seconds.

A 60-second example

Two 30-second jobs make 60 seconds. Run one on each id, or two on the same id for a matched look. Join them in Timeline: ceil(60/60) = 1 minute, so $0.10. The generation cost is the two per-second list prices times the clip length times 1.25, which you can compute from pricing_skus on each model.

Because a 4-second minimum applies to seedance-2.5, a 3-second cut needs a different id or a trim after generation.

Treat vendor lists as a wish list

Features such as 20 reference assets, 12 sequential images or green-screen replacement come from the vendors' pages. Unless Sume's docs list an input for them, do not assume the Sume id takes it. The reliable inputs on Sume are the ones in the request table of the video docs: frame_images, input_references, generate_audio and the model's own durations and resolutions.