Upload a video as a reference, and Seedance 2.0 can replicate its camera movements, transfer its motion to new characters, extract its expressions, or restyle it entirely. This video-to-video capability is unique in the AI video space—no other model accepts video as a reference input. If you're a filmmaker, animator, or content creator who works with existing footage, this changes your entire workflow.
What Video-to-Video Can Do
Seedance 2.0's video reference system isn't a single feature—it's a versatile tool that serves multiple creative purposes depending on how you prompt it.
| Use Case | What Happens | Example |
|---|---|---|
| Motion Transfer | Character movements from the reference video are applied to new characters | Dance choreography → animated character performs the same dance |
| Camera Replication | Camera path and movements from the reference are reproduced | Severance-style orbiting shot → applied to a different scene |
| Expression Transfer | Facial expressions and emotions are mapped to a new character | Actor's performance → animated avatar delivers same emotions |
| Style Transfer | The scene content is restyled while preserving the motion | Live-action footage → anime version with same movements |
| 3D Previz to Final | Rough 3D animation becomes polished final output | Blender blocking → cinema-quality rendered scene |
How to Use Video References
Step by Step
- Open Dreamina → Video Generation → Seedance 2.0
- Select "All-Round Reference" mode
- Upload your reference video (up to 15 seconds, up to 3 videos total)
- Optionally upload character images as additional references
- Write your prompt using @ tags to specify how each reference should be used
- Set duration to match your reference video length
- Generate
Video Reference Specifications
| Spec | Limit |
|---|---|
| Max video references | 3 per generation |
| Max duration per video | 15 seconds |
| Combined with images | Up to 9 images + 3 videos + 3 audio = 12 total references |
| Tag format | @Video1, @Video2, @Video3 |
Motion Transfer
The most powerful application. Upload a video showing specific movements, then apply those movements to entirely different characters or scenarios.
Example: Fight Choreography
Upload a rough 3D fight animation as @Video1. Upload character reference images as @Image1 and @Image2. Prompt: "A fight between @Image1 and @Image2. They are fighting in a forest. Use the motion of @Video1. Use the camera movement in the 3D video as well."
This produces remarkably accurate results—the characters follow the choreography from the 3D previz while being rendered with full detail, lighting, and physics. Using 3D previz as a motion source virtually eliminates the "extra sword" or "phantom limb" errors that can occur in pure prompt-based fight scenes.
Example: Dance Replication
Upload a dance video as @Video1 and a character image as @Image1. Prompt: "@Image1 performs the exact dance moves from @Video1. Concert lighting, wide shot." The model replicates the dance choreography with your character while generating appropriate music and audience reactions.
Camera Replication
Extract complex camera movements from existing footage and apply them to entirely new scenes.
Example: Severance-Style Shot
Upload a clip with distinctive camera work (orbiting, zooming, tracking). Upload a character image and a background image. Prompt: "@Image1 is running through @Image2 background. The character and camera movement should follow @Video1 exactly."
The model reproduces the zoom patterns, orbit speeds, and tracking movements from the reference video, applied to your new scene. Results aren't frame-perfect replicas, but they closely capture the feel and rhythm of the original camera work.
Expression Transfer
Map an actor's facial performance onto a different character while maintaining that character's visual identity.
How It Works
Upload a video of expressions as @Video1 and a character image as @Image1. The model extracts the emotional performance (micro-expressions, eye movements, mouth shapes, head tilts) and applies them to your character.
This is particularly powerful for:
- Animated characters: Give cartoon or anime characters realistic emotional performances
- Avatar content: Transfer your own expressions to a digital presenter
- Consistent branding: Same character, different emotional performances for different videos
3D Previz to Final Render
This is a game-changer for 3D animators and filmmakers. Create rough blocking in any 3D software (Blender, Maya, Cinema 4D), export as video, and let Seedance turn it into polished footage.
Why This Matters
- Absolute motion control: You decide exactly how characters move and how the camera travels
- Drastically reduced errors: The model follows your previz rather than improvising, eliminating most physics and continuity errors
- Rapid iteration: Adjust the 3D previz and regenerate in minutes instead of hours of manual rendering
- No rendering farm needed: Seedance handles the final visual quality
Best Practices for 3D Previz Input
- Keep the 3D scene simple—clear silhouettes and movements matter more than detailed textures
- Match the video duration to your previz length
- Use distinct character colors/shapes so the model can differentiate between subjects
- Include camera movements in the previz export for full control
Combining Video with Other References
Video references become even more powerful when combined with image and audio references:
- Video + Character Images: Motion from the video, appearance from the images
- Video + Background Image: Camera work from the video, environment from the image
- Video + Audio: Visual movement from the video, soundtrack from the audio reference
- Video + Multiple Characters: Choreography from the video applied to specific character designs
The @ tag system makes these combinations explicit. For example: "Use the motion of @Video1. @Image1 is Character A. @Image2 is Character B. @Image3 is the background. @Audio1 provides the soundtrack."
Known Limitations
- Not frame-perfect: Camera replication captures the feel and rhythm but not exact frame-by-frame matching
- Complex multi-person scenes: Motion transfer works best with 1-2 characters; large groups may lose accuracy
- Reference video quality: Low-resolution or highly compressed reference videos produce lower-quality output
- Processing time: Video reference generations are the most computationally expensive—expect longer wait times
- Unique to Seedance: No other major model supports video references, so there's no benchmark for comparison
Video-to-Video vs Other Modes
| If You Need... | Use This Mode |
|---|---|
| Maximum creative freedom from text | Text-to-Video |
| Character-specific animation from photos | Image-to-Video |
| Motion/camera replication from existing footage | Video-to-Video (this page) |
| Talking head with lip-sync | AI Avatar |
| Music-synced dancing | Video + Audio references combined |
Frequently Asked Questions
Q: Can any AI video model use video as a reference input?
A: As of February 2026, Seedance 2.0 is the only major AI video generator that accepts video references. Sora 2 only accepts a single image or text. This is one of Seedance's most significant competitive advantages.
Q: Does the output include audio?
A: Yes. Seedance 2.0 generates native audio for all output regardless of the input mode—dialogue, sound effects, ambient sounds, and music are all generated simultaneously.
Q: Can I use copyrighted footage as a reference?
A: Using copyrighted video as a motion or camera reference to generate original content falls into a legal gray area. The output is a new generation, not a copy—but consult legal advice for commercial use with copyrighted reference material.
Q: How close is the motion replication?
A: Very close for character movement and camera path. Not frame-perfect—think of it as a skilled animator interpreting your reference rather than a rotoscope trace. 3D previz inputs produce the most accurate results.
Q: What video formats does Seedance accept?
A: Standard video formats up to 15 seconds in duration. For best results, use clear, well-lit footage with distinct character movements.
Q: Can I combine video reference with image references?
A: Yes—this is the recommended approach. Upload the video for motion/camera reference and images for character appearance. You can combine up to 3 videos + 9 images + 3 audio files = 12 total references.
Master the prompt techniques for video references in the Prompt Guide. Compare Seedance's unique video reference capability against competitors in Seedance vs Sora 2. For access and costs, see the Pricing Guide.