Seedance Prompt Structure
Turn an idea into one executable shot
A prompt needs to establish what appears, how the event unfolds, and how the camera observes it. Start with one primary action, then add only the scene, camera, and constraints that affect the result.
Source review: . Examples are original editorial drafts, not measured generation results.
- 01Define the subject and one primary action
- 02Describe movement in chronological order
- 03Keep only constraints that protect the result
Check the version and the tool's scope
This builder provides natural-language structure, not a model API. The version selector supplies capability guidance; versions share the same editorial rules. Set duration, aspect ratio, references, and sound in the video platform. Numbers in a prompt do not configure those controls.
- Seedance 1.5 Pro supports native joint audio-video generation. Describe who speaks, the exact line, and the accompanying action.
- Seedance 2.0 documents text, image, audio, and video references and 15-second audiovisual output. Assign a clear purpose to each reference.
- Seedance 2.5 documents generation up to 30 seconds and time-specific editing. Organize longer scenes into ordered shots, then check which controls your platform exposes.
Give each of the six parts one job
The subject identifies what appears. The action explains what happens during the clip. The scene adds place, time, and light. Camera language defines shot size, position, and movement. Visual style aligns texture and color. Constraints protect identity, structure, direction, and continuity.
Keep the categories separate so conflicts remain visible. Avoid hiding camera instructions inside the character description or repeating clothing details inside every sentence.
A basic and detailed version of the same idea
Use the basic version to test the direction. Add sequencing, timing, and continuity only after the idea works.
At an old train station at sunset, a young woman in a dark long coat stands near the platform edge. She hears a distant train and slowly turns her head toward the end of the tracks. Medium shot, gentle push-in, realistic cinematic texture, warm sunset against a cool gray platform.
An empty old train platform at sunset. A woman around 28 wears a charcoal long coat and stands in profile behind the yellow safety line. She begins still, looking forward. After hearing a train, she lifts her chin slightly, turns her head slowly to the right from the shoulders, and stops while looking down the tracks. Start in a medium shot and push in very slowly to a medium close-up with a stable camera. Warm backlight catches her hair while the platform remains cool gray. Preserve her face, hair, clothing, body proportions, and position. Do not add other people or turn the whole body.
Write a process instead of a final state
“A woman holds a cup” describes a state. A more executable version gives a start, movement, and endpoint: “She looks down, reaches with her right hand, grips the handle, lifts the cup steadily, and holds it in front of her chest.”
When an action needs more than three connected steps, consider splitting the idea into multiple shots. A complex action combined with an elaborate camera move creates many simultaneous requirements.
- State the starting posture
- Name the main movement, hand, or direction
- Define where the action stops
The structure survives model updates
ByteDance states that Seedance 2.0 accepts text, image, audio, and video inputs. Seedance 2.5 adds stronger reference control, long-form continuity, and timestamp-based control. Product features will keep changing, while subject, action, camera, and constraints remain useful ways to audit an idea.
ZHENSHENG does not claim that one sentence pattern produces the same result in every model version. Test against the current model and source material.
References
Model features can change. Check the current official pages before relying on a version-specific capability.
- ByteDance Seed. Seedance 2.0. Accessed 5 Sept. 2026.
- ByteDance Seed. Introducing Seedance 2.5. Accessed 5 Sept. 2026.
- ByteDance Seed. Sound and Vision, All in One Take—The Official Release of Seedance 1.5 Pro. Accessed 5 Sept. 2026.