Seedance Performance Guide
Use Seedance in Flick to direct acting, dialogue, voice, and timing.
Write performance as visible or audible actions in the order they happen.
- Do not rely on abstract emotion labels alone.
- Do not overload short shots with too many actions.
- Do not assume the model will infer action order, speaker order, or audio hierarchy by itself.
Treat these examples as starting points. Review each result and adjust one instruction at a time.
Core rule
Write performance as things the model can visibly or audibly execute.
Instead of writing:
- she is sad
- he looks suspicious
- the delivery feels emotional
write:
- her eyes stop moving for one beat
- his jaw tightens slightly
- she takes one small inhale before speaking
- the voice is low, breathy, and restrained
- after hearing the line, she does not answer immediately
Concrete actions and timing give the model clearer direction than emotion labels alone.
One short shot = one primary performance beat
Short clips are easier to control when each shot has one main performance task.
Instead of fitting a walk, a sit, and a book-opening into one short clip, choose one beat:
He reaches the table and sits down.
Use another shot for opening the book.
Write actions in the order they happen
Do not stack actions as a list and expect the model to resolve timing correctly.
Use clear sequencing words:
- first
- then
- after
- only after
- before
- throughout
- while maintaining
This is especially important for:
- emotional turns
- prop handling
- dialogue timing
- action continuity
- lip sync
- multi-character scenes
Good structure:
She initially remains still.
After hearing his words, her eyes stop.
Her jaw tightens slightly.
She looks back at him but does not speak.
Separate character action from camera action
Do not merge performance and cinematography into one vague sentence.
Write them as two different layers:
Character:
She slowly turns her head toward the sound.
Camera:
Locked medium shot. The camera does not move.
If the camera moves, give it one clear job only.
Good example:
Character:
Both characters walk forward at a steady pace.
Camera:
The camera remains in front of them and moves backward at their walking speed.
Maintain the same distance, eye-level framing, and shot scale throughout.
Use time windows when timing matters
Timecodes are not mandatory, but they become useful when:
- the shot is longer than a few seconds
- the action has multiple stages
- the emotion evolves over time
- audio and performance need pacing control
Examples:
- 0-3 seconds
- 3-6 seconds
- first... then...
- after a brief pause...
Time windows are approximate instructions, not exact editing keyframes.
Use them to describe pacing, then check the generated timing.
Break voice performance into controllable dimensions
Do not just write sad voice or angry voice.
Break vocal performance into parts:
- Volume: quiet or barely above a whisper.
- Pace: measured, rushed, or slow with pauses.
- Pause: a brief pause after a phrase.
- Breath: one small inhale before speaking.
- Emphasis: slight stress on the final word.
- Delivery: restrained urgency, guarded warmth, or dry disbelief.
Example:
He says, "Who saw him leave?"
Low, direct, and controlled.
Natural conversational pace.
A brief pause before "leave," with slight emphasis on the final word.
He does not raise his voice.
Related guides
- Voice Consistency for creating the voice a character speaks with.
- Generate Video for the settings around the prompt.
- Character Dialogue for turning a still and an audio track into a talking clip.
- in 'twixt vows behind the scenes for how a director applies these rules on a real film.