What is a good Seedance prompt formula?
Use subject and setting, visible action, camera movement, timing, light, sound or dialogue, and explicit constraints.
Prompt guide
Facts checked August 12, 2026
Seedance prompts: direct a changing shot, not a static image
Across Seedance versions, the useful prompt is a shot specification: what is in frame, what changes, how the camera moves, when events happen, what is heard, and what must not appear. Version pages then own the model-specific recipes.
Core formula
7 parts
Subject through constraints
First test
5 seconds
One dominant question
2.0 recipes
6
Original with test status
Iteration rule
1 change
Per rerun
The general Seedance prompt formula
Subject and setting → visible action → camera movement → timing → light and visual treatment → sound or dialogue → constraints.
- Subject: only details the shot must preserve.
- Action: verbs, physical interaction and an end state.
- Camera: lens or shot size plus one main movement.
- Timing: order events when sequence matters.
- Light: concrete sources and direction instead of “cinematic”.
- Sound: exact quoted dialogue or named ambience and effects.
- Constraints: no captions, logo, watermark, extra subject or cut when required.
Camera language that controls a short take
Push in / pull out
Changes emotional distance without changing the subject action.
Lateral tracking
Follows movement while keeping scale more stable.
Orbit
Reveals volume and background, but needs physical room around the subject.
Rack focus
Moves attention between two clear depth planes; use it once in a five-second test.
Crane reveal
Start close, then rise to reveal the environment; keep the action simple.
Static camera
Often the best diagnostic choice when subject motion is already difficult.
Text-to-video and image-to-video need different prompts
| Mode | Spend prompt detail on | Avoid |
|---|---|---|
| Text to video | Subject, setting, action, composition and camera | Leaving the scene underdefined |
| Image to video | Subject motion, environment motion, camera and end state | Redescribing details already visible in the image |
Route model-specific intent to the right page
Seedance 2.0 Prompts
Native-audio recipe structure and first-party Fast/Standard evidence.
Open 2.0 recipesSeedance 2.5
2.5 text, photo and 30-second prompting stays on the 2.5 hub until search data supports a child page.
See 2.5 prompt moduleImage to Video
Source-image preparation and a motion-first prompt formula.
Open motion guideDebug the prompt without changing everything
- If the action is wrong, keep camera and style fixed and rewrite only the verbs and end state.
- If the camera is wrong, keep subject action fixed and reduce to one camera instruction.
- If dialogue is wrong, shorten the spoken line and simplify the shot.
- If identity drifts, reduce subject motion or use a clearer image source.
- If a long shot drifts, return to five seconds and validate each beat.
Frequently asked questions
How long should a Seedance prompt be?
Long enough to control visible and audible decisions, but no longer. Contradictory or decorative details do not automatically improve output.
How do I prompt Seedance image to video?
Describe what changes: subject motion, environmental motion, one camera move, timing and the end state. Do not repeat the source image as prose.
Where are the Seedance 2.0 prompt examples?
The dedicated Seedance 2.0 Prompts page contains six original recipes and states the evidence status of each one.
Prompt examples are separated by test status. First Take Lab does not present provider gallery media or unrun recipes as first-party output evidence.