The Formula
Subject + action, then Scene, Visual style, Camera movement or cut, Audio. Fill only the slots you need; two or three sentences is the sweet spot.
- Subject + action
- Required. Who or what moves and what happens, written as motion over time, never a still picture.
- Scene and environment
- Where it happens: place, light, weather, surface details.
- Visual style
- The look: live action, animation, film stock, palette.
- Camera movement or cut
- One specific move with a visible result, or a cut, such as a slow push in that settles on her hands.
- Audio
- Music in ( ), sound effects in < >, dialogue in { } and on-screen subtitles in 【 】.
The Rules
Subject and action come first
It is the one required slot. Describe what the subject does over time; a prompt that reads like a still photo gives the model nothing to animate.
Fill only the slots that matter
Leave out scene, style, camera or audio when they add nothing. Two or three sentences is the length the guide recommends.
Name the camera move and its result
Write "slow push in that settles on her hands", not "cinematic". Translate unusual terms into what the viewer sees, such as how a rack focus shifts from the leaves to the face.
Use the four brackets for sound and text
Music beds go in ( ), sound effects in < >, spoken dialogue in { } and on-screen subtitles in 【 】. Dialogue and subtitles are separate channels, so a line in { } is spoken, not printed.
Name the language before non-Chinese dialogue
Start with the language and the delivery: Dialogue language: American English. The man says calmly: {I thought you weren't coming.}
Write long takes as stages
Past about 10 seconds, break the take into stages, each with one main change and one observable end state. Seedance 2.5 runs up to 30 seconds in one take.
Show emotion with visible markers
Pair any emotion word with what the viewer sees or hears: the trigger, the immediate reaction, the change in eyes, breathing or hands, and the settled expression.
Give every reference image one job
In reference mode, bind each image to one named subject and say what to leave out: [Image1] defines the character's face, hair and wardrobe. Do not use the image background.
A Template You Can Copy
[Subject] [primary action, as motion over time] in [scene]. [Visual style and light]. [Shot size and one camera move with its end state]. Dialogue language: [language]. [Speaker] says [delivery]: {[exact line]} <[sound effect]> ([music bed])
A Real Example
The shot: Night, rain-streaked diner window. A woman in a red coat sits alone in the last booth, turning a ring on the table. Slow push in from the window. Mara says: "You kept it." (6 seconds, 16:9)
A woman in a red coat sits alone in the last booth of a diner at night, slowly turning a ring between her fingers on the table. Rain streaks down the window beside her, blurring the glow of the street outside; warm amber light from the diner interior contrasts with the cold blue night beyond. Slow push in from outside the window, moving through the rain-streaked glass until it settles close on her face and hands. Dialogue language: American English. Mara says quietly: {You kept it.} <soft rain ticking on glass> <faint diner hum> (sparse, melancholic piano underscore, soft dynamics)
It keeps the slot order: her action with the ring, the rain-streaked scene and its light, one push in with a clear end point, then audio. The line sits in { } after a language and delivery cue, the rain and diner hum go in < >, and the piano bed goes in ( ), so each sound lands in the right channel.
Dialogue
Put the exact spoken words in curly braces { }, which Seedance 2.5 lip-syncs. For any language other than Chinese, name the language first, then the speaker and delivery. Keep subtitles separate in 【 】 if you want text on screen. Seedance 2.0 and 2.0 Fast do not speak, so leave dialogue out for them.
Dialogue language: American English. The man says calmly: {I thought you weren't coming.}
Camera Moves
| Move | How to write it for Seedance 2.5 |
|---|---|
| Slow push in | Slow push in that settles on her hands. |
| Whip pan | At 4 seconds, whip-pan right until a bus blocks frame. |
| Aerial glide | Low aerial gliding over a skyline at dusk, one unbroken push toward a single lit tower. |
| Rack focus | Rack focus: shift focus smoothly from the foreground leaves to the person behind; the leaves blur while the face sharpens. |
| Push in, then cut | Begin with a medium shot of the wheel, slowly push in toward the cup's surface, then cut to a frontal view of the shelf. |
| Orbit | The camera orbits the table once and stops facing her. |
Common Mistakes
| Mistake | Fix |
|---|---|
| Describing a still image instead of motion over time. | Say what the subject does from start to finish, such as "lifts the cup from the wheel and places it on the shelf". |
| Writing "cinematic" without a specific move. | Name one move and what it ends on. |
| Cramming a whole sequence into one long take with no structure. | Split it into stages, each with one main change and one end state. |
| Putting dialogue in quotation marks or in 【 】. | Spoken words go in { }; 【 】 is for on-screen subtitles only. |
| Group-labelling reference images ("these are the characters"). | Bind one image to one subject and add an exclusion such as "Do not use the image background". |
Length, Resolution and Price
Seedance 2.5 makes single takes from 4 to 30 seconds, tops out at 720p, speaks dialogue with optional sound, accepts an end frame and up to 50 references, and costs $0.2205 per second at 480p or $0.473 per second at 720p. Seedance 2.0 runs 4 to 15 seconds at $0.3034 per second at 720p or $0.682 at 1080p, and Seedance 2.0 Fast is $0.2419 per second at 720p; neither 2.0 model speaks. All three turn away photoreal human faces (checked 2026-10-22). Every model side by side is on the model spec sheet.