Introduction
This Seedance 2.0 prompt guide explains how to direct AI video generation using a true multimodal workflow. Instead of relying only on text, Seedance 2.0 lets you combine images, video clips, audio, and natural language to control motion, pacing, emotion, and camera behavior.
Rather than thinking like a prompt engineer, Seedance 2.0 encourages you to think like a director—clearly defining references, intent, and the audience experience.
Supported Inputs and Limits
Seedance 2.0 supports four input types. All inputs are optional and can be combined.
Image Input
- Formats: jpeg, png, webp, bmp, tiff, gif
- Up to 9 images
- File size per image: under 30 MB
Images are best used to define:
- Character appearance and clothing
- Composition and lighting
- Environments and product details
Video Input
- Formats: mp4, mov
- Up to 3 videos
- Total duration: 2 to 15 seconds
- File size per video: under 50 MB
- Supported resolution range:
- Minimum: 640 × 640 (480p)
- Maximum: 834 × 1112 (720p)
Video references are ideal for:
- Camera movement and framing
- Action rhythm and pacing
- Transitions and visual effects
Note: Using reference video costs more than image-only generation.
Audio Input
- Formats: mp3, wav
- Up to 3 audio files
- Total duration: up to 15 seconds
- File size: under 15 MB
Audio is used to control:
- Rhythm and timing
- Scene cuts
- Emotional tone
Text Input
Natural language prompts are used to describe:
- Actions and motion
- Camera behavior
- Timing and sequencing
- Emotion and performance
Output Duration
- Selectable from 4 to 15 seconds
File Count Limit
- Maximum of 12 total files across all input types
- Prioritize materials that most strongly affect visuals or rhythm
Generation Modes and Entry Points
Seedance 2.0 provides two generation modes.
First and Last Frame Mode
Use this mode when:
- You upload only a starting image or ending image
- The motion and story are primarily driven by text
All Reference Mode
Use this mode when:
- You combine images, video, audio, and text
- You want precise control over motion, camera language, or editing
If you plan to mix multiple modalities, always choose All Reference Mode.
How to Reference Files in Prompts
Seedance 2.0 assigns meaning to uploaded files using @filename references.
Basic Syntax
- @image1
- @video1
- @audio1
Example Usage
- @image1 as the first frame
- @video1 for camera movement reference
- @audio1 for background music and rhythm
References can be inserted by:
- Typing @ directly in the prompt
- Clicking the @ icon in the parameter toolbar
Always clearly describe how each reference should be used.
Core Prompt Structure
A strong Seedance 2.0 prompt usually includes:
- Reference assignment
- Scene description
- Action and motion
- Camera behavior
- Timing or sequencing
- Audio or emotional intent
Simple Example
@image1 as the first frame.
Reference @video1 for camera movement and pacing.
A woman gracefully hangs laundry, then reaches into a basket and shakes out another piece of clothing.
The motion is natural and continuous.
Using References Effectively
Image References
Use images to define:
- Character appearance
- Clothing and styling
- Lighting and composition
Example:
Use @image1 for character appearance and clothing details.
Use @image2 for the background environment.
Video References
Use videos to define:
- Camera movement
- Action rhythm
- Transitions and visual effects
Example:
Fully reference @video1 for camera movement and transitions.
Match the action rhythm and pacing.
Audio References
Use audio to define:
- Background music
- Rhythm and timing
- Emotional tone
Example:
Use @audio1 as background music.
Synchronize scene cuts with the audio rhythm.
Advanced Prompt Techniques
Extending an Existing Video
Extend @video1 by 5 seconds.
The generated duration should be 5 seconds.
Continue the motion smoothly without abrupt cuts.
Combining Multiple Videos
Insert a new scene between @video1 and @video2.
The new scene shows the character walking into a foggy forest.
Maintain consistent lighting and pacing.
Continuous Actions
The character transitions smoothly from jumping to rolling on the ground.
Maintain motion continuity with no visual breaks.
One Take Shot
One continuous shot with no cuts.
Follow the character from the street, up the stairs, through the hallway, and onto the rooftop.
Editing Existing Videos
Seedance 2.0 supports targeted video edits without recreating everything.
You can:
- Replace characters
- Modify actions
- Extend scenes
- Change emotional tone
Example:
Override the original story of @video1.
The man’s expression turns cold and calculated.
He pushes the woman into the water decisively.
Do not change the environment or camera angle.
Emotion and Performance Control
Emotional intensity can be directed with precise descriptions.
Example:
The character looks at her reflection, pauses in silence, then suddenly breaks down screaming.
Match the emotional intensity and facial expressions from @video1.
Music Sync and Beat Matching
Seedance 2.0 can align visuals with musical rhythm.
Example:
Reference @video1 for rhythm and beat timing.
Cut scenes on musical accents.
Increase motion energy on strong beats.
Important Restrictions
- Uploading realistic, identifiable human faces is not supported
- Content containing real faces will be blocked
- All human examples are illustrative only
Best Practices
- Clearly label every reference with its purpose
- Keep prompts structured and chronological
- Describe intent instead of technical jargon
- Use reference videos for complex camera work
- When extending video, match output duration to extension length
Summary
Seedance 2.0 is not just a video generator. It is a multimodal directing system that understands composition, motion, rhythm, and storytelling when given clear references and intent.
Think less like a prompt writer and more like a director.
Describe what to reference, how to use it, and what the audience should feel.
