Seedance 2.5 R2V Guide: Direct Motion with References

R2V, or reference-to-video, uses a video or spatial reference to communicate movement, blocking, framing, or interaction that is difficult to specify with text alone. Dreamina's official Seedance 2.5 page illustrates R2V with green-screen footage and simple white-model references. Treat that as a supported workflow claim, then test the behavior with assets you are allowed to use.

The central discipline is role separation: one reference demonstrates motion, another may establish identity, and the prompt states which properties should not transfer.

Choose the right reference

A useful motion reference has:

• one readable action;
• an unobstructed subject silhouette;
• stable framing or an intentional camera path;
• a clear beginning and ending pose;
• minimal motion blur;
• and no private or unlicensed material.

A rough phone recording can be more useful than a polished montage when it explains the body path cleanly. For object motion or several characters, a simple 3D blockout can make position and timing explicit without adding unwanted appearance cues.

Assign every input a job

@Video 1: transfer the performer's step timing, arm path, and final pose only.
@Image 1: preserve the lead character's face, hair, jacket, and proportions.
@Image 2: preserve the room layout and window position.
Do not copy clothing, face, background, color grade, or camera artifacts from @Video 1.

This does not guarantee perfect separation, but it makes the intended mapping inspectable. If an unwanted property transfers, you know which instruction to refine or which reference to simplify.

Build the first R2V test

1. Trim the reference to one action and remove unused lead-in or trailing frames.
2. Note the reference duration, frame shape, camera behavior, and ending pose.
3. Add only the minimum identity or environment assets needed for the target shot.
4. Describe the target setting and subject without repeating every visible reference detail.
5. State what transfers from each reference and what must remain original.
6. Generate the shortest version that can reproduce the full motion.
7. Compare timing, spatial path, identity, anatomy, camera, and final pose separately.

Example R2V brief

Use @Video 1 only for the dancer's three-step path, clockwise turn, and final hand position.
Use @Image 1 for the dancer's identity, silver jacket, black trousers, and body proportions.
Set the performance on a clean concrete stage with a dark blue background and one overhead softbox.
Camera remains locked in a full-body front view; preserve headroom and floor contact.
Do not transfer the room, clothing, face, camera shake, compression artifacts, or color from @Video 1.
End with both feet planted and the right hand held at shoulder height for one second.
Output: 9:16, clean audio-free plate, no text.

Diagnose R2V failures

• Appearance from the motion clip transfers: use a cleaner silhouette or green-screen reference and strengthen the “motion only” assignment.
• Feet slide or lose contact: shorten the action, preserve the full body in frame, and choose a reference with a visible floor plane.
• Camera movement appears unexpectedly: state that camera motion does not transfer and request a locked frame.
• Identity changes during a turn: reduce rotation, add a stronger identity reference, or create separate front and rear shots.
• Multiple characters collide: block one character at a time or use a simple spatial reference with distinct positions.
• Ending is unusable: define the final pose, composition, and hold duration.

Change one factor per retry. Replacing the motion reference, identity image, camera instruction, and environment at once destroys the evidence needed to learn what worked.

Rights and safety

Use reference footage you created, licensed, or have permission to process. A motion reference can still contain a person's likeness, distinctive choreography, private location, or protected performance. Do not use R2V to impersonate a person or conceal the source of sensitive footage.

For a text-first structure, read how to write Seedance prompts. For current access and feature wording, use the Seedance 2.5 status brief.

Official reference: Dreamina Seedance 2.5.