AI Generation

Motion Transfer: Driving a Character with Reference Video

Also called pose transfer, motion reference, character animation, driving video

Motion transfer copies the movement of a subject in a driving video onto a different character, so a still image or a new identity performs the same action. The pipeline extracts a pose track from the driver, retargets it to the target's proportions, then generates frames conditioned on that track plus your reference identity, which is why it is also called pose transfer.

The pipeline in three stages

Understanding where the failure happened means knowing which of three stages produced it.

  1. Extraction. A pose estimator reads the driving video and produces a per-frame track: 2D keypoints, a skeleton, a dense body map, or facial landmarks. Nothing about the driver's appearance is carried forward, only geometry.
  2. Retargeting. The track is remapped onto your target's proportions. A tall driver's shoulder positions mean something different on a short character, and this is where most visible damage originates.
  3. Generation. Frames are produced conditioned on the retargeted pose plus your identity reference, so the character keeps its design while adopting the movement.

Choosing a driving video

The driver matters more than the model. In practice:

  • One subject, fully framed, for the whole clip. A limb that leaves frame produces a limb that vanishes or flails.
  • Static camera. Camera motion in the driver becomes subject motion in the output, since the pipeline cannot tell the two apart.
  • No cuts. Every cut is a discontinuity in the pose track, which reads as a snap.
  • Moderate speed with clear silhouettes. Fast motion means blur, and blur means an unreliable pose estimate.
  • Minimal self-occlusion. Arms crossing the torso, hands behind the back, and turning fully away are all points where the estimator guesses.
  • Framing matched to the target. Full-body driver for a full-body character, upper body for a portrait.

Why proportions dominate everything

Noodle limbs, sliding feet, and joints that bend backward are almost always retargeting artifacts, not generation artifacts. The pose track is a set of coordinates, and if those coordinates were produced by a body shaped differently from your character, no amount of prompting fixes the mismatch.

Two concrete habits help a lot. Match the reference image's pose to the driver's first frame, so the sequence starts from agreement rather than a jump. And pick a driver whose build resembles your character, which matters more than whether the motion itself is impressive.

What transfers well and what does not

Body motion, timing, and weight shift transfer well, and that is the whole point. Hands are the standard casualty, since fingers occupy few pixels and the pose track for them is sparse. Fine facial performance rarely survives, so pair motion transfer with a separate mouth pass when the character speaks. Cloth and hair are generated rather than simulated, so they follow the body loosely and can lag or clip.

It also helps to keep the three related tasks distinct. Motion transfer drives a subject's performance from a driving video. Lip sync drives only the mouth, from audio. Camera motion control specifies how the camera moves, not what the subject does. Reaching for the wrong one of the three is a common way to spend a lot of credits on the wrong problem.

Models that support this

Pulled from the live ZOOOP model catalog, so this list stays current as new models ship.

See all 18 models

The prompt for this

A starting point that reliably produces the effect. Adjust the subject and setting; keep the technical clauses.

Apply the motion of the driving video to the reference character, keep the character design and outfit unchanged, static camera, feet planted on the ground

Try Motion Transfer yourself

Open the generator with a starting point already filled in.

Frequently asked questions

What is the difference between motion transfer and pose transfer?
They are used interchangeably. Pose transfer more often describes a single image target with a per-frame pose driving it, while motion transfer describes the video-level result. Both mean the same pipeline: extract skeleton, retarget, generate.
What makes a good driving video?
One subject, whole body in frame for the entire clip, plain background, a static camera, no cuts, moderate speed, and no limbs disappearing behind the torso. The framing should also match your target image, since a full-body dance clip cannot drive a half-body portrait.
Why do the limbs stretch or bend the wrong way?
Retargeting error. The pose track is in the driver's proportions, and if your character is a different height or build the joints land in the wrong places. Match the driver's body proportions and starting pose to the reference image as closely as you can, and pick a driver with less extreme motion.
Can I transfer motion onto a non-human character?
Only when the skeleton maps. A stylized human, a humanoid robot, or a character in a costume usually works. Quadrupeds, characters with no legs, or wildly non-human proportions do not, because there is nothing sensible to retarget the human joints onto.
Does the driving video's background come across?
No, and it should not. Only the pose track is used, so the background and the driver's appearance stay behind. Camera movement in the driver is the exception: it gets read as subject movement, which is why a static camera driver is safer.

Related terms