AI motion transfer: keep the people and the moves, change the world around them
Motion transfer takes a performance you already filmed and puts it somewhere else. The people stay — same faces, same bodies, same clothes — and so do their movements, the camera path and the timing; only the location is rebuilt from photos you supply. On VdoBloom it is the Motion Transfer preset of Genjutsu at /dashboard/video-creation/genjutsu/, and it runs on ByteDance Seedance 2.5's editing mode, which treats your clip as the thing to preserve rather than a loose reference.
That distinction matters, because "motion transfer" is used for two different things. The older meaning is retargeting: copy a dance from a video onto a new character generated from scratch, where framing and identity drift. What this page describes is scene replacement with the performance locked. In our own test a football clip was moved from a stadium to a container port: all the players, their poses, the ball, the referee and even the boom microphone were identical, and only the ground, the stands and the skyline changed.
You describe the new place with photos and, optionally, a line of text. Each photo can be labelled with the part of the scene it shows — "the street", "the sky at dusk", "the building behind them" — so several photos compose one location. Clips are 2 to 30 seconds at 480p to 720p and the output keeps the clip's length and aspect ratio at 480p, 720p or 1080p.
What this tool does
- Motion transfer on VdoBloom keeps every person, pose, face, garment, camera move and beat of timing, and replaces only the location.
- It is scene replacement with the performance locked — not retargeting a dance onto a generated character, where faces and framing drift.
- The new location comes from up to 10 labelled photos plus an optional line of text; "@Image1 shows the street; @Image2 shows the sky at dusk" is how the model receives it.
- In VdoBloom's September 2026 test a stadium became a container port with all eleven players, the ball, the referee and the boom microphone unchanged.
- Source clips are 2-30 seconds at 480p-720p; the output matches the clip's length and ratio at 480p, 720p or 1080p.
- Use it to relocate an ad for a new market, move a performance into a set you could not afford, or restage the same take in several eras or styles.
- Billing is per second of clip on the with-reference-video rate: 15 seconds is 212 credits at 480p, 473 at 720p, 852 at 1080p.
How it works
1.Open Genjutsu and choose Motion Transfer
At /dashboard/video-creation/genjutsu/. The card shows the stadium-to-port test so you can see what "location only" means.
2.Upload the performance
The clip whose people and movement you want to keep: 2-30 seconds, 480p-720p. The price appears when the file is chosen.
3.Add photos of the new location and label them
Wide shots work best; several photos can describe one place. Labels say what each shows — "the street", "the skyline", "the floor".
4.Describe the setting in a line, then generate
Optional but useful: "a neon Tokyo street at night", "a 1970s newsroom". Four to seven minutes later the same take plays out in the new world.
Frequently asked questions
What is AI motion transfer?
Here it means keeping a filmed performance — people, movement, camera, timing — and rebuilding the location around it from reference photos. Elsewhere the phrase can mean retargeting motion onto a new character; that version loses faces and framing, this one keeps them.
Can I change the background of a video with AI?
Yes, and more than the background: the whole environment, including what the people stand on and what is behind and beside them, is rebuilt to match your photos while the people themselves are untouched.
Will the people still look like themselves?
Yes. Identity, clothing and movement are preserved; that is the point of the preset. If you also want to change who is in the shot, run Object Swap on the result.
Can I use several photos for one location?
Yes — up to ten, each labelled with the part of the scene it shows. Two or three wide shots of the same place from different angles give the model the most to work with.
Does it work for an ad I want to localise?
That is one of the main uses: the same performance restaged on a street, in a kitchen or in a store that looks like the target market, with the actors and the product handling unchanged.
What does motion transfer cost?
Per second of clip at the output resolution: a 15-second clip is 212 credits at 480p, 473 at 720p or 852 at 1080p, shown before you confirm and refunded if the job fails.
Related tools
Genjutsu AI video: change what is in a clip without reshooting it
Upload a 2-30s clip, label a photo with what it replaces, get the same shot back changed. Faces, motion and camera stay. Seedance 2.5, up to 1080p.
AI object swap: replace a person, outfit or product in a video and keep the shot
Replace a person, outfit or product in a video: upload the clip and a labelled photo of the replacement; the rest stays as filmed. 2-30s, up to 1080p.
Higgsfield Genjutsu alternative: the same engine with labelled targets and credits that don't expire
VdoBloom vs Higgsfield Genjutsu: same Seedance 2.5 engine. Differences: per-photo labels, price before upload, no plan gate, credits that never expire.
Seedance 2.5: the 30-second model, and how to run it
ByteDance Seedance 2.5 renders 4-30 seconds at 480p, 720p or 1080p (10-bit). Pick it in Text to Video, Image to Video or Effects. 30s at 480p is 276 credits.
Ready to try it?
Move a clip somewhere new