Seedance 2.0 max length: 15 seconds, and how to go longer

Seedance 2.0 clips are 4 to 15 seconds long. Sume's seedance-2, seedance-2-fast and seedance-2-mini take whole-second durations in that range, and fal's Seedance 2.0 fast page lists the same 4 to 15 seconds. A request for 20 seconds returns 400 unsupported_capability. For a single clip of 16 to 30 seconds, Sume's docs list seedance-2.5 at 4 to 30 seconds.

Sume facts come from the Video generation docs and the API's request checks; the fal fact is from its page, read 2026-09-29.

What duration does each Seedance model accept?

The reserve column is what Sume holds at submit for a 720p 16:9 clip of the longest length, computed from the provider list price times 1.25, before the 5.5% agent fee.

What error do I get for 20 seconds?

The duration is checked against the model's list. The message names the model and the value, for example "seedance-2 does not support duration 20." Read supported_durations from GET /v1/videos/models and clamp the value in your code.

What does length do to the reserve?

Sume counts video tokens by frame size and duration, so the reserve grows in step with the seconds. On seedance-2 at 720p and 16:9, a 5-second clip reserves $1.89 and a 15-second clip reserves $5.67. Draft a scene at 5 seconds, settle the prompt, then ask for the long version once. The final amount is on the job as usage.cost.

How do I get more than 15 seconds?

There are two routes. The first is one clip on seedance-2.5, which the docs list at 4 to 30 seconds. The second is to chain clips: make a 15-second clip, take its last frame as a still, and start the next clip with that still as first_frame. The request parameters in the docs include no extend option, so the still is your bridge.

curl -X POST https://api.sume.com/v1/videos \
  -H "Authorization: Bearer $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "seedance-2.5",
    "prompt": "A slow dolly along a rain-soaked street at night",
    "duration": 30,
    "resolution": "720p",
    "aspect_ratio": "16:9"
  }'

Does a chained clip keep the same look?

The docs do not promise continuity between two jobs. Repeat the subject, wardrobe and lighting words in both prompts, keep the aspect ratio and resolution identical, and watch the join before you publish. A cut on a change of angle is easier to make than a join in the middle of one continuous move. See extending an AI video past 30 seconds for the chaining steps.