Write a 30-second Seedance 2.5 prompt as six five-second beats

To get a coherent 30-second clip from Seedance 2.5, write the prompt as an ordered list of about six five-second beats, name the subject once, and keep the camera language consistent. Then send it as one seedance-2.5 job with duration: 30. ByteDance describes the model as producing 4 to 30 second clips natively, so there is no stitching step to plan around.

This post is a prompt-writing practice, not a vendor rule: the BytePlus prompt guide could not be read as plain text on 2026-10-04, so nothing here claims official syntax. Treat it as a structure to test.

Why beats instead of one long paragraph

A 30-second request gives the model far more time to wander than a 5-second one. A paragraph that describes a mood leaves the middle of the clip undefined. A beat list pins what must be on screen at roughly each five-second mark, and it gives you something to check against when you review the result.

BytePlus markets the model for "30-second cinematic narrative" work and lists film, advertising, e-commerce and previsualization among its use cases (BytePlus: Dreamina Seedance 2.5). A narrative needs a start, a turn and an end, which is what a beat sheet encodes.

A six-beat template

Keep each beat to one action and one camera instruction. Put the subject description first and do not rename it later.

• Beat 1 (0-5 s): establishing shot, subject enters, steady camera.
• Beat 2 (5-10 s): subject starts the main action, camera pushes in slowly.
• Beat 3 (10-15 s): detail close-up of the product or face, same lighting.
• Beat 4 (15-20 s): wider shot showing the result of the action.
• Beat 5 (20-25 s): second action or reaction, same location.
• Beat 6 (25-30 s): resolve on a clean final frame you could caption.

The request on Sume

On the Video Router, seedance-2.5 accepts 4 to 30 seconds at 480p, 720p and 1080p. duration is a whole number of seconds. The example below drafts the idea at 480p first; raise resolution only after the beats read well.

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: beat-sheet-001" \
  -d '{
    "model": "seedance-2.5",
    "prompt": "A baker in a small kitchen. 0-5s: wide shot, she enters and ties her apron. 5-10s: she kneads dough, slow push-in. 10-15s: close-up of flour on her hands. 15-20s: tray of loaves into the oven. 20-25s: she checks the oven window. 25-30s: loaves on a rack, steam rising.",
    "resolution": "480p",
    "duration": 30,
    "aspect_ratio": "16:9",
    "mode": "async"
  }'

What a draft costs before you commit

The price table below is computed from Sume's token-based pricing: width x height x seconds x 24 / 1024 tokens, at the Seedance 2.5 list rate, times Sume's 1.25 margin (Video Router states the 1.25 factor). Read the live number from GET /v1/video-router/models before you budget.

Review the result against the beats

Pull stills at 2.5, 7.5, 12.5, 17.5, 22.5 and 27.5 seconds with video frames (the clip must already be stored on media.sume.com, so import a downloaded MP4 first) and check one question per still: is the action of that beat visible? If beats 3 and 4 collapse into one, shorten the prompt rather than adding words.

Sume does not promise any beat will land; the model is probabilistic and the docs list no seed parameter to repeat a take. Budget a second attempt, and see the single continuous take prompt if you want no cuts at all.