Seedance 2.5: how to write a prompt that works the first time
The prompt structure that gives consistent multi-shot scenes with Seedance 2.5.
September 12, 2026, 5 min read
Seedance 2.5 understands long, structured prompts. Here is the template that gives the most usable shots on the first try.
The five-line structure
- The subject: who, with two or three fixed physical details (“lighthouse keeper, gray beard, yellow raincoat”).
- The place and time: “spiral staircase, dusk, cold light through the portholes”.
- The action: a single main verb per shot.
- The camera: shot type and movement (“medium shot, slow upward tracking shot”).
- The sound: ambience, music or a line of dialogue in quotes.
Chaining several shots
Separate the shots with “Shot 1:”, “Shot 2:” and so on, and repeat the character’s physical details in every shot. Seedance 2.5 then keeps the same face for up to 30 seconds.
Using references
Add a photo of the character and an image of the setting. These are the two references that stabilize the result the most. Beyond five, the gain becomes small.
Common mistakes
- Describing three actions in the same shot: the model picks one at random.
- Using different words for the same character (“the old man”, then “the keeper”): it may create two people.
- Asking for on-screen text: add it in the edit instead.
You can test these prompts for free on Seedance Studio.
| # | Model | TopVideoAI score | API price | Try | |
|---|---|---|---|---|---|
| 1 | Seedance 2.5ByteDance | 9.4 | $0.29/s | Try |
Bar scale: 5 to 10. API prices from Apiframe’s public price list, checked 08/20/2026.
tryseedance.ai
Seedance 2.5, the top model in our video ranking, in your browser.
Write a scene and add your reference images: Seedance Studio generates the video with sound, dialogue and lip sync.
- Up to 30 s per video, in 4K
- Generated sound, music and dialogue
- Multiple shots with the same character

