Which AI video models make a 20-second clip in one job on Sume?

On Sume, two text-to-video models cover a 20-second clip in one generation: seedance-2.5 accepts 4 to 30 seconds and wan-3.0 accepts 2 to 30 seconds. Gemini Omni Flash 1.1 accepts 3 to 10 seconds, MiniMax H3 and H3 Max accept 5 to 15, and each other catalog model has a maximum of 15. Ask GET /v1/videos/models before you submit, because limits differ by model.

The duration table

The table lists what the Sume docs state for supported_durations. Two rows are special: Motion Transfer and Recast take a source video, so their length follows that source and they are not text-to-video.

Why one job beats two clips

A 20-second shot in one generation keeps one camera, one light, and one set of faces. Two 10-second clips joined by a cut can drift: a sleeve changes, a logo moves. If the shot has no cut in it, ask for 20 seconds and keep the continuity. If it is two beats, make two clips on purpose.

If you need more than 30 seconds, you need more than one job. Timeline joins them: its limit is 200 slots and 1800 seconds, at $0.10 per output minute.

Check the request fits

Duration is an integer number of seconds, validated against the model. duration is preferred over the alias duration_seconds, and if you set both they must agree. Ask for a length the model does not list and the request is rejected before it starts, which is cheaper than a failed job.

curl -sS "https://api.sume.com/v1/videos/models" \
  -H "Authorization: Bearer $SUME_API_KEY"
# read data[].supported_durations and supported_resolutions

Choosing between them

Pick by what the shot needs. Seedance 2.5 takes 4 to 30 seconds with 480p, 720p, and 1080p. Wan 3.0 starts lower, at 2 seconds, which suits short loops, and also takes audio and video references. If you need a source clip as input, the Motion Transfer and Recast rows above are the choice, and their output length follows the source.

Whichever you pick, read the model's supported_durations from the list endpoint at submit time instead of copying this table into code. The registry can change, and the endpoint is the source of truth.

Pricing is per second

Sume reserves the provider list price times 1.25 at submit, and each model has its own pricing_skus. Seconds drive the bill, so a 30-second job costs three times a 10-second job on the same model and resolution. Check pricing_skus for the model you pick, since rates vary by model and resolution.