Seedance 2.5, Seedance 2 or fast: pick the model for episode length

Choose by the length of the shot and the resolution you need. On Sume, Seedance 2.5 accepts 4 to 30 seconds at 480p, 720p or 1080p, Seedance 2 accepts 4 to 15 seconds at the same resolutions, and Seedance 2 Fast and Mini accept 4 to 15 seconds at 480p or 720p. A 25-second shot is Seedance 2.5 or Wan 3.0; a 10-second draft is a candidate for fast.

The catalog rows

Mapping to episode length

A 15-second shot is inside every row, so there you choose on resolution and price. Above 15 seconds only Seedance 2.5 (and Wan 3.0, which goes to 30 seconds) can render a shot in one job. Anything longer than the model's limit has to be made as two jobs and joined; the Timeline endpoint does that for $0.10 per output minute.

There is also a quality question that a table cannot answer. A 30-second shot has more chances to drift than a 10-second one, so a long single take is worth it only when the action is continuous. A cut every few seconds is how most micro-drama is edited, and it fits the 15-second rows without effort.

Whichever row you choose, check the catalog before the season, because limits are validated per model and a value the model does not advertise returns a 400 rather than being dropped.

Read the live rates

We do not copy Seedance prices here. They are token based, and the exact values change with the catalog. Call the catalog and read the pricing_skus field of each row; it holds Sume's billable rates as decimal-USD strings. Multiply by your own token or duration estimate to compare two models for the same shot.

curl -sS "https://api.sume.com/v1/videos/models" \
  -H "Authorization: Bearer $SUME_API_KEY"

A simple policy

Write the choice down once per season so the whole team renders episodes the same way.

• Draft every shot on the cheaper row you are allowed to use, at 480p or 720p.
• Finalize on the row whose duration and resolution match the shot.
• Pin the model by id in your season config instead of relying on sume/auto, which is deterministic but defaults to Omni Flash 1.1 and may choose differently from what you tested.
• Remember that explicit routing is needed for the fast and mini rows.