Read the motion, ignore the dancer
Pose tracking
The reference video is analyzed frame by frame to estimate a skeleton: body joints, hand and finger positions, head orientation and facial landmarks. Everything about how the dancer looks — clothes, skin, the room — is discarded. Only the movement survives.
- Why it matters
- Clean pose data is the ceiling on quality. A missed hand or a left/right limb swap here turns into a glitch in the final video.
- Tip
- Use a locked-off shot with the whole body in frame and nothing crossing in front of the dancer.






