Seedance 2.5 one-take prompt: one continuous shot, 30 seconds

To get a one-take clip from Seedance 2.5, write the prompt as one unbroken camera path, say so in words ("single continuous take, smooth camera movement, no cuts"), and request the full length with duration: 30 on model seedance-2.5. That wording and structure come from ByteDance Seed's own launch examples; Sume's side is only the request: seedance-2.5 accepts 4 to 30 seconds at 480p, 720p or 1080p.

ByteDance's Seedance 2.5 launch post says the model can generate a 30-second clip in a single pass and organise several connected shots inside it, so a story has a setup, development, turning points and a resolution (read 2026-10-02). Sume's limits come from Video generation and Video Router.

What does Seed's one-take prompt look like?

Seed publishes two. The text-to-video one opens with "One-take handheld gimbal tracking shot", then walks the camera through a backstage curtain, a dressing room, a corridor and onto a stage, ending on a wide shot of the arena. Each sentence moves the camera to a new place, so the clip reads as one path rather than a list of scenes.

The reference-to-video example opens with "16:9 widescreen, cinematic texture, single continuous take, smooth camera movement, no cuts." It then gives time ranges (0-5s, 6-10s, 11-20s) and tells the camera what to do in each one. Those ranges describe beats inside one take, not edits.

How do I send it to Sume?

Submit once, with an Idempotency-Key, and poll the job; a 30-second video usually outlasts a short HTTP wait, so use the default async mode. generate_audio is optional. The docs say it defaults to the model's audio capability.

If you want the prompt to hold at a narrower frame, set aspect_ratio as well. 9:16, 16:9, 1:1, 4:3, 3:4 and 21:9 are in the Seedance 2.5 envelope on the Video Router catalog.

curl -X POST https://api.sume.com/v1/videos \
  -H "Authorization: Bearer $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: one-take-001" \
  -d '{
    "model": "seedance-2.5",
    "prompt": "Single continuous take, smooth camera movement, no cuts. A handheld gimbal shot follows a courier from a loading dock through a corridor and out onto a rooftop at dusk. The camera pulls back to a wide shot of the city.",
    "duration": 30,
    "resolution": "720p",
    "aspect_ratio": "16:9"
  }'

Does "no cuts" guarantee no cuts?

No. It is a prompt instruction, and neither Seed's page nor Sume's docs promise it as a hard setting. Seed describes the model as keeping subjects stable across cuts and transitions, so a take can still contain a shot change if the prompt implies one. Keep the camera path unbroken, avoid words like "cut to", and review the result.

If you need more than 30 seconds, Seed describes multi-round extension: continue from an existing clip with a new prompt. Extending past 30 seconds covers what each round costs on Sume.

What should I check before paying for 30 seconds?

A 30-second job reserves the full length, so a few cheap checks come first.

• Run a shorter draft first: the same prompt at duration: 8 and resolution: "480p" shows whether the camera path reads.
• Confirm the model's limits with GET /v1/video-router/models rather than assuming one envelope; the Video Router docs say limits are per model.
• Reuse the same Idempotency-Key when retrying a submit, so a retry returns the original job instead of a second one.