Seedance 2.5 prompt formula
There is no official rigid syntax. Use a short, auditable direction that makes the subject, motion and references clear.
The current official documentation does not prescribe a magic formula. Its examples show concrete scene direction, camera language, timing, audio and reference relationships. BytePlus also recommends keeping current Seedance 2.0-series prompts under 1,000 words; exact behavior and limits may vary for Seedance 2.5 by platform/API.
Subject
Who or what is visible? Include identity or product traits that must stay stable.
Action
Describe the observable movement in a sequence the model can follow.
Environment
Set the location, objects, weather and physical details that affect motion.
Camera
Name one primary camera move, framing and speed instead of a pile of lenses.
Lighting
Give the light direction, quality and color when it changes the scene.
Timing
Use a short sequence or timed beats when the order of actions matters.
Audio
Describe ambience, dialogue, effects or music only when sound is part of the intent.
References
Say what each image, video or audio reference controls and what must not change.
[subject] + [observable action]
in [environment], filmed with [camera + framing + speed]
under [lighting / style], from [start] to [end]
with [audio direction]
[reference instructions, if used]Text-to-video prompts
These are starting directions, not guaranteed outputs. Change the subject and action before adding more adjectives.
Cinematic / rain and neon
CopyableA controlled one-take scene with a clear subject, action and camera path.
A woman in a red raincoat walks through a neon-lit city street after dark and turns toward the camera. Start with a medium tracking shot, then make a slow push in as she passes the glowing signs. Wet asphalt reflects cool blue and warm amber light; rain continues throughout. Cinematic realism, restrained pacing, one continuous shot. Ambient rainfall, footsteps and distant traffic, no dialogue, no music.Product / tabletop commercial
CopyableA product prompt that defines the reveal, material and camera move.
Premium tabletop commercial for a matte black travel mug on a pale stone surface. A narrow beam of morning light moves across the mug as condensation gathers and a small stream of coffee pours into the open lid. Begin with a macro close-up, then slowly pull back to reveal the full product and a clean negative-space background. Crisp commercial lighting, realistic metal and ceramic textures. Soft pour sound and a subtle room tone, no voiceover, no on-screen text.Character / readable performance
CopyableKeep the action simple enough for the character's performance to stay legible.
A young astronomer stands on a quiet rooftop at blue hour and unfolds a paper star chart. She looks from the chart to a bright meteor crossing the sky, then smiles with quiet surprise. The camera begins in a locked-off medium shot and gently arcs to her side without cutting. Natural facial performance, soft wind moving her coat, cool twilight with a warm rooftop light. Rooftop ambience and a distant city hum, no dialogue.Landscape / scale and movement
CopyableDescribe a large environment with one clear camera movement and physical motion.
A wide coastal landscape at sunrise, with tall grass bending in the sea wind and waves breaking below a dark cliff. The camera glides forward from behind the grass, rises slightly over the cliff edge, and settles on the horizon. Natural documentary realism, warm low sunlight, fine mist in the distance, gradual pacing. Wind through grass, rolling surf and distant seabirds, no music.Commercial / vertical social clip
CopyableA short vertical ad direction with a sequence that is easy to follow.
Vertical social ad for a bright citrus drink on a summer picnic table. A hand opens the cold can, ice cubes slide into a clear glass, and the drink pours in with visible bubbles. Use a quick macro push in for the opening, a smooth side track during the pour, and finish on a clean close-up of the can beside the glass. Bright daylight, fresh yellow and green palette, upbeat commercial energy. Crisp fizz and ice sounds, short cheerful music, no extra text.Image-to-video prompts
When an image is already doing the visual description, use the prompt to direct what changes after the reference frame.
Preserve the parts of the image that matter, then describe subject motion, camera movement, environment motion and timing. Re-describing the whole image can create unnecessary conflict with the reference. Accepted roles and input limits depend on the platform/API.
Reference image / animate the subject
CopyableUse [Image 1] as the source appearance and composition. Keep the subject's identity, clothing, colors and background stable. After the first frame, the subject slowly turns toward the window while a curtain moves in a soft breeze. The camera makes a restrained push in and keeps the face readable. Warm morning light remains consistent. Quiet room tone and soft fabric movement, no dialogue.Product reference / controlled reveal
CopyableUse [Image 1] as the exact product reference. Do not redesign the shape, label or material. Animate only the motion: a slow rotation on the tabletop, a narrow highlight travelling across the surface, and a few drops of condensation sliding downward. The camera tracks slightly from left to right and ends in the same centered product framing. Clean studio light, subtle product-commercial sound, no new text or logos.Character reference / motion only
CopyableUse [Image 1] to preserve the character, hairstyle, clothing and setting. Describe only what changes after the first frame: the character takes two careful steps toward the camera, pauses, and looks past the lens as leaves move behind them. Keep the original framing and color palette; use a very gentle handheld drift. Natural outdoor ambience, no dialogue, no costume changes.Camera movement prompts
A camera term works best when it is tied to a subject, direction and speed.
Push in
DirectionBring attention closer to a subject or detail.
The camera makes a slow push in from a medium shot to a close-up as she notices the sound.Pull back
DirectionReveal the larger environment around the subject.
The camera slowly pulls back from the product to reveal the full table and the sunlit room.Tracking
DirectionMove with a subject travelling through the scene.
A side-tracking camera keeps pace with the runner while the background slides past.Orbit
DirectionShow a subject from changing angles while keeping it central.
The camera performs a slow half-orbit around the dancer as the fabric catches the light.Handheld
DirectionAdd documentary energy without changing the subject's action.
Use subtle documentary-style handheld drift while the reporter crosses the market.Locked-off
DirectionKeep attention on performance, timing or a visual transformation.
Use a locked-off static frame as the paper sculpture unfolds by itself.Multi-shot and timed prompt structure
Seedance 2.5 is positioned for 30-second storytelling, but timeline controls and exact timestamp behavior depend on the platform/API.
Use timed beats when the order matters. This is a prompt pattern, not a claim that AstraiLab currently exposes a timeline editor or that every integration will honor every timestamp literally.
Timed beat template
CopyableKeep each beat observable and give the transitions a clear cause.
[0–5 seconds] Wide view of a quiet train platform at dawn. A woman in a blue coat waits beside one suitcase; the camera slowly tracks toward her.
[5–10 seconds] The train enters the background. She turns toward the sound and takes one step forward; cut to a medium profile shot.
[10–15 seconds] The doors open. She picks up the suitcase and walks into the carriage as the camera follows from behind.
[15–20 seconds] Close-up on her hand releasing the suitcase handle. The train begins moving; warm window light crosses her face.
Natural station ambience, rail vibration and soft footsteps. Maintain the blue coat and the same dawn color palette throughout.Reference prompts
Give each reference a job. State what to borrow and what to preserve.
- Subject identity, product shape or wardrobe.
- Composition, camera language or motion rhythm.
- Environment, sound character or visual palette.
- A first frame, final frame or transition target.
- Which reference controls which element.
- What must remain unchanged.
- What changes after the first frame.
- Which action happens first, next and last.
Reference mapping template
CopyableUse [Image 1] for the character's identity and clothing.
Use [Image 2] for the location and background composition.
Use [Video 1] for the slow camera movement only.
Keep the face, outfit and color palette from [Image 1] unchanged.
The character walks from the doorway to the window, pauses, and looks outside.
Use the framing from [Image 2] and the camera rhythm from [Video 1].
Natural room tone and soft footsteps, no dialogue.The current BytePlus documentation describes reference roles and combined image/video/audio inputs for the Seedance 2.0 series. Treat exact reference counts, file requirements and role syntax as platform/API-specific.
Common prompt mistakes
Fix the instruction before adding more style words.
Action conflicts with the reference
If the reference is the first frame, say what changes after it instead of replacing the subject, pose and setting.
Too many camera instructions
Give one primary move per beat. A push in, orbit, crane and handheld change at once is hard to audit.
Ambiguous subject
Name the subject before the action and use the same wording when the action continues.
Conflicting timing
Use a simple order or time window. Do not ask an action to finish before it starts.
Adjectives without motion
Replace “epic, beautiful, dynamic” with something visible: a camera path, physical action, light change or ending beat.
Seedance 2.5 prompt FAQ
Practical answers for the first prompt iteration.
How do I write a Seedance 2.5 prompt?
Start with the subject and observable action, then add the environment, camera movement, lighting, timing and audio direction that matter. Keep the prompt internally consistent and describe what should happen rather than stacking adjectives.
Does Seedance 2.5 support image-to-video?
Seedance is presented as a multimodal video model with reference control, and the current Seedance-family documentation includes image-to-video workflows. Exact image roles, limits and access depend on the platform/API you use.
How do I describe camera movement?
Name one primary camera move, say what the camera follows, and add its speed or timing. For example: “slow side tracking with the cyclist, then settle into a close-up.” Avoid asking for several unrelated camera paths at the same moment.
Can Seedance 2.5 use reference images?
Reference control is a core capability described on the official Seedance 2.5 page. In a prompt, identify what each reference controls and what must remain unchanged. The accepted file types, number of references and role syntax vary by platform/API.
How long should a Seedance prompt be?
The current BytePlus ModelArk documentation recommends keeping Seedance 2.0-series prompts under 1,000 words because lengthy text can scatter attention. Use that as a practical ceiling, not as a universal 2.5 limit; a shorter prompt with clear observable actions is usually easier to audit.
Check the source
Prompt behavior and model limits can change as platforms update their integrations.
- ByteDance SeedSeedance 2.5 official model page
- ByteDance SeedOfficial prompt, camera and reference examples
- BytePlus ModelArkCurrent video generation API and prompt guidance
- AstraiLabSeedance 2.5 Generator