30-second ad in one request: which Sume video models accept 30 s

Two catalog models on Sume accept a 30-second ad in one request: seedance-2.5 at 4 to 30 seconds and wan-3.0 at 2 to 30 seconds. minimax-h3 stops at 15 seconds, minimax-h3-max at 15, gemini-omni-flash-1.1 at 10, and the docs say that every other catalog model has a limit of 15 seconds. For a 30-second ad you have the choice of a 30-second model, or two or more shorter clips that you join in Timeline. Vendors quote longer figures than a hosted route may allow, so write your code against GET /v1/video-router/models, not against a launch post.

The ranges

Sume figures are from the Video Router page. Vendor figures are as read on 2026-10-05.

Where the numbers disagree

Three gaps matter. First, MiniMax: the third-party summary gives a 4 to 15 second range, and Sume's page gives 5 to 15 for minimax-h3, so a 4-second request is a Sume validation error and not a vendor limit. Second, Gemini Omni: Google describes scene extension that analyzes up to 10 seconds of earlier context and reaches 40 seconds in 10 second steps. The Sume page lists 3 to 10 seconds per request and does not list an extension capability, so do not plan a 40-second Omni ad on Sume. Third, Seedance: ByteDance says that its API is coming soon through BytePlus ModelArk, while Sume lists seedance-2.5 in its own catalog, so Sume is the route to read here.

The request

Pin the model and ask for the duration explicitly, with a vertical ratio for a short-form ad. Use an Idempotency-Key so a retry does not start a second paid job. Sume bills list price times 1.25, so check the price of your resolution first.

curl -X POST https://api.sume.com/v1/video-router/generate \
  -H "Authorization: Bearer $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: ad-30s-seedance-001" \
  -d '{
    "model": "seedance-2.5",
    "prompt": "A 30 second vertical ad for a travel kettle: morning, pour, pack, end card",
    "resolution": "720p",
    "duration": 30,
    "aspect_ratio": "9:16",
    "mode": "async"
  }'

One 30-second clip or two 15-second clips

A single 30-second request is simple, and its weakness is that one bad second spoils the clip and you pay again for all of it. Two 15-second requests cost the same time in total, can run side by side, and let you redo one half. Join them in Timeline 1.0, which owns the clip sequence, transitions and audio. For a first ad, try the single request at 480p to read the story and the pacing, then decide.

For anything that must show the same product in both halves, give both requests the same reference images, so the item looks alike across the cut.

Think about the audio too. A 30-second clip carries one continuous soundtrack, and two clips joined later carry two. If you want one music bed, make the clips without music where the model allows it, or lay a single bed over the joined result in Timeline, and keep any spoken line inside one clip so that a cut does not land in the middle of a word.

Last, the price check. Sume bills the provider list times 1.25 for each catalog model, so the cost of a 30-second request depends on the model and the resolution that you pick. Read the figure for your own choice from the pricing page or the catalog before you queue a batch of them, and write the number into the job record. A 30-second model at 1080p is not the place to learn the price by surprise.