Skip to content

Mocap

Kitly captures body and face motion from ordinary video or photos, on your machine. No suit, no markers, no upload: the capture runs on-device with MediaPipe, so a phone clip of your own performance becomes an animation clip you can retarget.

Open the dialog

Three doors lead to the "Generate Mocap from Image/Video" dialog:

  • The burger menu's Generate Mocap...
  • Right-click a video or image asset and choose Create Mocap...
  • The mocap entry in the Sandbox operations tree (it opens this same dialog)

Set it up

  1. Choose what to Detect:
    • Body: a "Skeleton-only motion clip (BVH)". This is the one you retarget onto characters.
    • Face: a "Textured, animated face mesh (GLB)". A moving face mesh carrying the actor's captured texture.
    • Both: run the two captures together.
  2. Pick your source from the thumbnail list (the asset you right-clicked is preselected).
  3. Give the capture a Name.
  4. Face captures offer "Keep head steady (no root motion)", which locks the head in place so you get pure facial performance without the bobbing. Useful when the face will drive a character whose head moves for other reasons.
  5. Video sources offer Refine passes with hints showing the effective fps. More passes trade time for smoother, more stable tracking.
  6. Click Queue Generation. The confirmation points you at the Jobs panel.

While it runs

Mocap is a job with visible phases: Loading, Detecting (with a frame count like 123/456), Building animation, Baking mesh, Exporting GLB, Capturing texture, and Registering. Long clips take a while at the Detecting phase; the counter tells you how far along it is. Cancel from the Jobs panel any time.

Where results land

Everything registers into geo/source:

  • Body: Body <Name>.bvh, an animation asset.
  • Face: Face <Name>.glb plus a .face.json sidecar and a texture image sequence.

Reviewing the capture

Select the result in the viewer:

  • A body BVH plays on a paired or default rig, side by side with the source footage and a pose overlay, so you can judge the capture against the performance directly.
  • A face GLB plays its captured texture sequence synced with the source footage.

The animation transport gives you loop modes, speed from 0.1x to 4x, smoothing, and skeleton or wireframe display.

Getting it onto your character

A BVH clip carries its own capture skeleton, so the usual next stop is Retarget: map the capture skeleton to your rig's bones, preview against the synced footage, and export the clip onto your character.

What to expect from video mocap

Markerless capture from a single camera is a previz and gameplay-animation tool, not an optical stage. Framing matters: keep the full body in frame for body capture, face the camera reasonably for face capture, and light the subject evenly. Fast motion blur and heavy occlusion degrade the track. Refine passes help; a cleaner source helps more.