Seedance 2.5 prompt guide: direct the shot, second by second
A useful video prompt describes change over time. Build it from subject, action, setting, camera, look and audio—then add time ranges only when the shot has distinct beats.
Clarity beats adjective density. Start with one subject, one primary action and one camera move; add constraints only to solve a visible problem.
Site-made visual reference · not model output
Core formula
6 prompt layers
Subject, action, setting, camera, look, audio.
Long clips
Use time ranges
Assign one visible beat to each interval.
First test
5–10 seconds
Diagnose motion before extending the shot.
Avoid
Conflicting directions
Do not request several camera moves at once.
The foundation
A six-layer Seedance 2.5 prompt formula
- 01
Subject
Name the person, object or environment that must remain visually stable.
- 02
Action
Use a concrete verb and describe what changes from the beginning to the end.
- 03
Setting
Define the location, time of day and environmental motion that matters.
- 04
Camera
Choose shot size, lens feeling and one primary move with a start and stop.
- 05
Look
Specify lighting, palette, texture and realism without piling on incompatible styles.
- 06
Audio
Describe ambience, effects or dialogue intent only when audio generation is enabled.
Single-shot formula
A bicycle courier in a cobalt rain jacket accelerates through a wet night market. Low tracking shot beside the rear wheel, 35mm documentary feel, warm stall light reflected on asphalt, realistic spray and fabric movement, distant vendors and tire hiss, no music.
Temporal direction
Use timecodes for visible beats
Time ranges help when a shot contains sequential actions. Each interval should describe something the viewer can see or hear. Do not use timecodes to repeat styling words.
15-second narrative beat
[0–4s] Wide shot: a lone botanist enters a foggy glasshouse carrying a dim lantern; camera slowly cranes down. [4–10s] She brushes a giant leaf; blue bioluminescent veins wake and travel across the plant, reflected in her eyes. [10–15s] Camera arcs behind her as the entire glasshouse pulses awake; humid ambience, leaf movement, one low resonant tone.
10-second product reveal
[0–3s] Locked macro of condensation on an unbranded amber bottle, hard rim light on black. [3–7s] Slow half-orbit as the bottle rotates once; liquid catches a warm highlight. [7–10s] Movement settles into a clean hero frame, subtle room tone, no text, no extra objects.
Motion vocabulary
Write camera direction that can be executed
| Intent | Useful language | Common mistake |
|---|---|---|
| Reveal context | Start close, slowly dolly back to a wide shot | Asking for zoom, orbit and crane simultaneously |
| Follow action | Low lateral tracking shot, matched to the runner's pace | Saying only cinematic camera |
| Build attention | Slow push-in that stops at a medium close-up | No endpoint or subject named |
| Show an object | Controlled 120-degree clockwise orbit, stable product geometry | Requesting a full spin in too little time |
| Documentary energy | Handheld shoulder-height follow, small natural corrections | Combining handheld with perfectly locked framing |
Fix the prompt
Diagnose common failures before adding more words
| Problem | Likely prompt issue | Next test |
|---|---|---|
| Subject changes identity | Too many subjects or weak reference priority | Reduce the cast; put identity first; use a clean reference |
| Camera move is ignored | Move conflicts with action or has no endpoint | Use one move and state where it stops |
| Shot feels static | Prompt describes appearance but no change | Add one physical action with a clear beginning and end |
| Sequence order drifts | Several beats have no time ranges | Assign non-overlapping intervals |
| Visual artifacts multiply | Scene complexity is too high for the first test | Shorten duration and remove secondary motion |
Change one variable at a time
When a test fails, adjust the prompt, reference, duration or model—not all four. Controlled iteration teaches you which change helped.
Reusable starting points
Three prompt templates to adapt
Image-to-video
Preserve the subject, clothing, composition and background from the first frame. [subject] performs [single action]. [one camera move with endpoint]. [environmental motion]. [lighting and texture]. Avoid new objects, text and identity changes.
Character consistency
The same [character description] remains consistent throughout. [0–Xs] [first action and shot]. [X–Ys] [second action and camera transition]. Preserve face, hairstyle, wardrobe colors and body proportions; natural motion, no duplicate limbs.
Product commercial
An unbranded [product] on [surface]. [camera move] reveals [material detail]. Controlled studio lighting, accurate geometry, clean reflections, no text or extra accessories. End on a stable centered hero frame.
Common questions
How long should a Seedance 2.5 prompt be?+
Long enough to define the shot, not to fill a target word count. A focused single shot may need one paragraph; a 15–30 second sequence benefits from timecoded beats.
Does Seedance 2.5 understand timecodes?+
Time ranges are a useful way to communicate sequential intent in supported workflows, but they are direction rather than a frame-accurate editing guarantee.
Should I include negative prompts?+
Use a short avoid list only for recurring failures: no text, no duplicate limbs, no identity change. A long negative list can compete with the main instruction.
What resolution should I use while testing prompts?+
Use 480p and 5–10 seconds for early motion tests. Move to 720p and longer duration after the structure works.
Sources reviewed
- Kie.ai — Seedance 2.5 prompt and timecode input field
- fal — Seedance 2.5 workflow guide
- fal — Seedance 2.5 text-to-video schema
Examples are original editorial guidance, not guaranteed outputs. Model results remain probabilistic; use short tests before committing credits to long, high-resolution clips.
Direct one clear shot
Paste a template, replace the brackets, and test it at a short duration first.