Motion transfer AI explained: depth, pose and reference video

Motion transfer is the idea behind most "do the same dance" AI tools: take the movement from one video and apply it to a different subject. The three methods behind it, depth, pose and plain reference video, behave differently, and knowing which one a tool uses explains why some results look like the original and others do not. This guide explains each in plain words and shows how PrismPoster uses them in Remake.

What is motion transfer AI?

Motion transfer AI separates how something moves from who or what is moving, then applies the movement to a new subject. The input is a reference video that shows the motion. The output is a new video in which a different performer, or a drawn or described character, makes the same moves.

It is not video editing. Nothing from the reference is pasted into the result. The engine renders a new clip that follows the motion.

How do depth, pose and reference video differ?

Method	What it reads from the clip	Strength	Weakness
Depth map	Distance of every pixel from the camera, as a grey image	Keeps body shape, camera moves and space	Can carry the outline of the original person
Pose tracking	A stick-figure skeleton of joints over time	Clean, face-free, small	Loses hands, clothes and props unless added
Plain reference video	The footage itself	Easiest to use	The engine sees the original person and may refuse it

Depth maps are good at the shape of a scene. Pose tracking is good at the movement of a body. Plain reference video is the simplest, but it hands the engine everything, including the face of the person in it.

Which method does Remake use?

Remake derives a silent, face-free motion sheet from your clip, using depth and pose together, and sends only that. Depth keeps the body's volume and the camera path; pose keeps the timing of the moves; neither contains a face, a voice or a song. The sheet is then used as the motion reference for your performer.

That is why Remake can accept a clip with a person in it without passing the person on. The viral video remake page lists what it accepts and what it declines.

Why not send the clip straight to the engine?

Two reasons. First, the engine would receive the original person's face and voice, which are not yours to use. Our piece on why AI video tools refuse real faces covers the refusals that follow. Second, a derived sheet is smaller and cleaner, so the engine spends its effort on the movement. A raw clip carries lighting, clothing and backgrounds that compete with your scene.

How do you get the best result from motion transfer?

• Pick a clip with one clear performer and a steady camera. Crowds and fast cuts blur the pose.
• Keep the whole body in frame if the moves use it. Cropped limbs are guessed.
• Describe the scene around the moves, not the moves themselves; the sheet already carries them. AI video prompts shows how to phrase a scene.
• Use a performer with a clear reference. A consistent character holds up better; see character consistency in AI images.

Reference to video AI covers the wider family of references an engine can take, and how to remake a viral video with AI walks through the full flow.

Looking for clips that transfer cleanly? Viral video formats you can remake sorts the common ones.

Frequently Asked Questions

Is motion transfer the same as a face swap?

No. A face swap replaces a face in existing footage. Motion transfer renders a new video that follows the movement and keeps nothing else from the clip.

Does motion transfer copy the music?

Not in Remake. The motion sheet has no sound, and the AI engine never receives audio.

How long can the reference clip be?

3 to 30 seconds in Remake. Trim a longer clip to the section you want.

Can motion transfer work on animals or characters?

Yes, when the performer you choose is a described character. Results are best when the character has a body plan close to the clip's performer.