Seedance 2.0 Prompt Guide: Cinematic Templates That Work

Seedance 2.0 can use text, images, video, and audio as references. That makes prompting more powerful, but it also makes vague instructions more expensive: every reference and sentence should have a clear job.

Using Seedance 2.5? The same shot-brief structure remains useful. BytePlus now documents 4-to-30-second output and as many as 50 reference assets, with task-specific constraints. Check the Seedance 2.5 release, specs, and access status before assuming every mode or resolution is enabled in your account.

The best prompt is not the longest one. It is the shortest prompt that unambiguously defines the subject, action, camera, environment, sound, and continuity constraints.

A Reliable Seedance 2.0 Prompt Structure

Use this order:

Subject: who or what is on screen, including identity anchors.
Action: one primary physical or emotional beat.
Camera: shot size, movement, angle, and lens behavior.
Environment: location, depth, weather, and time of day.
Lighting and style: one coherent visual direction.
Audio: dialogue, ambience, effects, or silence.
Continuity constraints: what must not change.

Template:

[Subject] performs [single action]. [Shot size and camera movement]. [Environment and depth]. [Lighting and visual style]. [Audio direction]. Preserve [identity, wardrobe, geometry, or reference constraints]. [Duration and aspect ratio].

Text-to-Video Prompt Examples

Character close-up

A tired courier in a weathered red jacket pauses under a station awning and looks toward an arriving train. Medium close-up, slow push-in, stable eye line, shallow depth of field. Blue-hour rain with warm platform lights. Quiet rain, distant rail noise, one soft breath. Keep facial features and jacket details stable. Six seconds, 16:9.

Vertical motion comic scene

A young swordswoman steps from shadow into a lantern-lit alley and raises her blade once. Start on a waist-up panel-like composition, then make a controlled lateral track as cloth and hair react to the movement. Inked comic texture with restrained color. One metal draw sound and low city ambience. No costume or face changes. Six seconds, 9:16.

Product shot

A matte-black wireless speaker rotates slowly on a stone pedestal while a narrow band of light travels across its surface. Locked macro camera with a subtle 10-degree orbit. Dark studio, soft haze, accurate material reflections. Minimal electronic sound bed. Keep logo shape and product proportions identical to the reference. Five seconds, 1:1.

Image-to-Video Prompt Template

When an image already defines appearance, do not redescribe every visible detail. Tell the model what may move and what must remain locked.

Use the reference image as the first-frame identity and composition. Add [one subject motion] and [one environmental motion]. [Camera instruction]. Preserve face, clothing, object geometry, color palette, and background layout. No new people or objects. [Audio direction]. [Duration].

Example:

Use the reference as the first frame. The character turns toward the window once while curtain fabric moves in a light breeze. Slow 5% push-in with no roll or shake. Preserve facial identity, hairstyle, uniform, room geometry, and color palette. Add soft room tone and distant rain. Five seconds.

Using Video and Audio References

Assign one role to each reference:

image reference: identity, costume, prop, scene, or composition;
video reference: motion rhythm, camera path, performance, or transition;
audio reference: timing, dialogue cadence, ambience, or sound design.

If two references conflict, state the priority in the prompt. For example: “Use image A for character identity; use video B only for camera motion.” This is clearer than asking the model to imitate both in full.

Camera Language That Produces Clearer Results

Prefer one deliberate move per shot:

locked shot for dialogue and product geometry;
slow push-in for realization or tension;
lateral track for walking and environmental reveal;
controlled orbit for products or spatial emphasis;
pull-back reveal for scale and scene context.

Avoid stacking “handheld, orbit, zoom, dolly, whip pan” in one short clip. Conflicting camera verbs often produce apparent subject deformation or an incoherent background.

Common Failures and Fixes

Identity drift

Use a clear identity reference, reduce unrelated style adjectives, and repeat only the attributes that must remain fixed. Break a long performance into shorter shots that reuse the same references.

Too much motion

Keep one primary action and one secondary environmental motion. Remove simultaneous gestures that compete for attention.

Camera ignores the instruction

Move the camera instruction earlier, name one move, and specify a stable start composition.

Dialogue feels disconnected

Keep the face visible, reduce camera motion, provide the exact line or audio reference, and define the emotional delivery without adding multiple competing actions.

Multi-shot scene resets

Generate a planned shot list instead of forcing an entire sequence into one prompt. Reuse character, scene, and style references for each shot, then assemble the approved clips in an editor.

Prompt Workflow for a Story Episode

Parse the script into shots and assign one story beat to each shot.
Lock character and scene references before video generation.
Draft storyboards to validate composition and continuity.
Generate short motion tests at a cost-efficient setting.
Promote only approved shots to final resolution.
Save prompt versions and generation settings with every output.

This is the workflow AniKuku is designed around: the model generates shots, while the workspace preserves the story context that individual API calls do not know.

FAQ

What is the best Seedance 2.0 prompt format?

Use subject, one action, one camera move, environment, lighting, audio, and explicit continuity constraints in that order.

Can Seedance 2.0 use several references?

The official product page describes multimodal reference support. Assign each reference a specific role and verify current limits in the ModelArk tutorial.

How long should a Seedance prompt be?

Long enough to resolve ambiguity, but not padded with synonyms. A compact shot brief usually gives the model clearer priorities.