Motion Transfer
Recast any motion with your characters, locations and products or swap specific elements and keep the rest.
Motion Transfer: Put Your Character Into Any Video’s Moves
Film the move once, or find a clip that already has it. Then hand it to a character who was never there.
Upload a video and 1 to 8 photos of your character. You get a new video where the character does exactly what the person in the clip does: the same movement, the same camera, the same timing. Powered by Genjutsu, Higgsfield’s motion model.
What you can do with it
- Make a mascot dance. Take a TikTok dance and give it to your brand’s mascot, your 3D character or your dog.
- Recast a performance. A friend delivering a rap verse, a colleague doing the choreography from a music video.
- Reuse one shoot for many characters. Film a presenter once. Generate the same walk-and-talk with 5 different people.
- Bring a stylized character to life. An illustrated character, a game character or a concept art figure, moving like real footage.
- Rebuild a scene around the motion. The moves and the camera stay, everything else is up to your references and your prompt.
How to use it
- Upload the motion video. 4 to 30 seconds, one person doing the thing.
- Upload your character. 1 photo works. Up to 8 photos, from different angles and showing the full outfit, hold up better when the character turns.
- Optional: describe how the character should look in the video. Clothes, a hairstyle, a prop.
- Hit generate. The video lands in your library.
Pick 480p in the settings for a quick test, then run the keeper at 720p.
What you get
- The exact motion. Every step, gesture and turn from the source, on the same beats.
- The exact camera. The framing, the pans and the cuts stay as filmed.
- Your character, consistently. Face, hair and outfit follow your photos through the whole clip.
- A clip you can use. Results are cleared for commercial work.
Tips for a better result
- Clean source footage. One person, fully in frame, clearly lit. Crowds and heavy cuts confuse the transfer.
- Match the body. A character with roughly the same build as the person in the clip lands the moves more convincingly.
- Show the whole outfit. A full-body photo tells the model what the legs and shoes look like. A headshot leaves it guessing.
- Photos on plain backgrounds. The model reads the character faster when there’s nothing else in the shot.
- Keep the prompt short. “Red leather jacket, silver chain” does more than a paragraph about their personality.
- Run it twice. Each generation comes out a little different. Keep the one where the hands look right.
FAQ
How long can the video be? 4 to 30 seconds.
How many photos of the character do I need? 1 is enough to start. More angles and a full-body shot give a more stable character across the clip.
Does the character have to be a real person? No. Illustrated characters, mascots, 3D renders and game characters all work, as long as your photos show them clearly.
Does it keep the audio? The result follows the source video’s motion and timing. If your clip has music or a voice, put the original track back on the finished video with Mitte’s video tools, or generate a new one.
Can I change the background too? The motion and camera are the fixed part. Your character, their outfit and details are steered by your photos and your prompt.
What resolution do I get? 720p by default. 480p is available in the settings for cheaper test runs.
Can I use the videos commercially? Yes.