Seedance prompts: direct one clear, testable shot
A useful video prompt behaves like a compact shot brief. It separates what the viewer sees from what moves, how the camera behaves, what the scene sounds like, and which details must stay stable.
Keep the prompt ordered, change one variable per iteration, and save the exact settings with every successful result.
Use a six-part prompt order
Write the subject and setting first, then framing, subject action, camera movement, atmosphere or sound, and constraints. This order makes a failed result easier to diagnose because each part of the brief has one job.
Example: A ceramic tea cup on a rain-dark windowsill at dawn, medium close shot. Steam curls slowly while droplets move down the glass. The camera makes a gentle push in. Quiet room tone and distant rain. Preserve the cup shape; no text, no cut, no extra objects.
- Subject and setting
- Framing and viewpoint
- One subject action
- One camera movement
- Light, atmosphere, and sound direction
- Preservation and exclusion constraints
Adapt the brief to each creation mode
For text to video, describe the complete visible shot. For image to video, the source frame already defines appearance, so spend more words on motion and preservation. For reference remix, name uploaded slots exactly as the composer shows them, such as @Image1, @Image2, @Video1, or @Audio1.
References are instructions, not automatic permission. Upload only material you are allowed to use, explain what each reference contributes, and avoid asking the model to copy protected characters, brands, or a person's likeness without the necessary rights and consent.
- Text mode: define appearance and motion.
- Image mode: preserve appearance and define motion.
- Remix mode: identify each reference and define what to keep or change.
Iterate with one controlled change
If the action is wrong, revise the action phrase. If the frame drifts, simplify or lock the camera. If the atmosphere dominates the subject, reduce secondary movement. Rewriting the entire prompt after every result makes it difficult to learn what improved the shot.
Keep a small prompt log with the model label, version, mode, ratio, resolution, duration, sound option, references, and outcome. A prompt is not portable by itself; the surrounding settings are part of the recipe.
- Change one instruction between comparable tasks.
- Review the same failure category after each change.
- Keep the shortest prompt that still preserves the important decisions.
Seedance prompt examples to adapt
Use these as structures, not copy-and-paste promises. Replace the subject, setting, motion, camera, atmosphere, and constraints with decisions that fit your own material. For image-to-video, remove visual details already fixed by the start frame and strengthen the preservation instruction.
A prompt generator can help organize a brief, but it cannot know whether the chosen model supports a setting or whether you hold the necessary rights. Review every generated prompt, compare it with the live controls, and keep only instructions that can be observed in the final shot.
- Product reveal: A matte black bottle on wet stone; slow turntable motion; measured push in; narrow rim light; preserve label geometry; no extra text.
- Portrait: A chef pauses beside a warm kitchen window; subtle breathing and cloth movement; locked medium shot; quiet room tone; preserve face and hands.
- Landscape: Wind moves through tall grass above a coastal cliff; slow lateral track; late-afternoon haze; no people, no cut, stable horizon.
- Architecture: Morning light crosses a concrete gallery; one visitor walks through frame; slow symmetrical dolly; preserve straight walls and openings.
- Image animation: Keep @Image1 product shape and label fixed; camera arcs ten degrees right; soft steam moves behind it; no new objects.
- Reference remix: Use @Image1 for palette and @Video1 for camera pace; create a new licensed subject and setting; do not copy logos, people, or protected characters.
Frequently asked questions
How long should a Seedance prompt be?
Long enough to specify one shot, one subject action, one camera instruction, atmosphere, and essential constraints. Remove adjectives or subplots that do not change an observable decision.
Should I include camera terms?
Yes, when the camera movement matters. Use one clear instruction such as locked frame, slow dolly in, lateral track, or measured orbit instead of stacking incompatible moves.
How do I mention uploaded references?
Use the slot names displayed by the composer, such as @Image1 or @Video1, and state what to preserve and what may change.
Can the same prompt work across every Seedance version?
Do not assume so. Save the exact model selection and task settings with the prompt, then re-test after a model or provider change.