There are three families of motion capture, and they fail in different places. Knowing which one you are reaching for saves days. The short version: video capture is the cheapest to start and the hungriest for cleanup; suits cost money and setup but hand you dependable body data; generated motion produces a usable clip out of thin air for generic moves and nothing at all for specific ones.
The three ways people capture motion for Blender
- Video-based (markerless) - one or two cameras and an AI solver that estimates the skeleton per frame. Cheapest and fastest to start; every frame is an estimate, so every frame is candidates for cleanup.
- Inertial suits and gloves - sensors on the body, no occlusion problems for the torso and limbs, and gloves that actually deliver fingers. Costs money and calibration time, and the data still needs a polish pass.
- Generated motion - text-conditioned models that produce a walking, dancing or lifting clip without anyone performing it. Weekend-fast for generic moves; useless the moment the shot has to match a specific performance.
Capture footage that solves well
A solver can only be as good as the video it is fed. The checklist is short, and almost everyone ignores it:
- Full body in frame for the entire take - feet leave the frame first, and the solver invents the rest.
- Plain background; no other people drifting through the shot.
- Steady camera, no zoom, no panning. Put the phone on something.
- Loose clothing is the enemy - fabric breaks the silhouette the solver reads. Fitted clothes, no props.
- One move per take. Perform it, stop, reset, repeat.
From clip to keyframes in six steps
01 - Generate the character first. A rigged humanoid from a prompt gives the solve a target and gives you something to preview against immediately.
02 - Record the clip. Follow the checklist above, then record three takes of the same move - solvers disagree with themselves between takes, and the best one is not always the first.
03 - Solve. Run the footage through your capture tool and export the skeleton as FBX or BVH. Check the solve in the tool first: a bad take re-solved is still a bad take.
04 - Retarget, do not re-rig. In Blender, map the source skeleton onto your character with the retarget tools already built into the rigging workflow. Match the rest poses before you judge the result.
05 - Clean up in passes. Foot contacts first, then hip drift, then smoothing, then hands. Fixing in that order stops you from smoothing problems you were about to delete.
06 - Loop or trim, then commit. If the shot is a locomotion loop, close the cycle deliberately - the first and last frames are the ones nobody fixes and everybody notices.
What markerless capture gets right - and wrong
| Move | Result |
|---|---|
| Walking, running, idle | Reliable. The cleanup is small and mostly foot contacts. |
| Fast direction changes and dance | Workable, expect jitter and popped shoulders. Budget the cleanup. |
| Hands and fingers | Weak from single video. A glove or dedicated hand solve is still the honest answer. |
| Faces | A separate pipeline that drives shape keys, not bones. Plan it as its own pass. |
| Interaction with props and furniture | No data. Block the contact points by hand; the solver does not know what it cannot see. |
The cleanup is the project
Most people arrive expecting the solve to be the work and the cleanup to be the formality. It is the other way round. The cleanup is a short, repeatable pass once you have done it twice: hunt sliding feet, pin the contact frames, damp the jitter the solver added to shoulders and wrists, and close loops. The faster your cleanup pass, the more capture starts to feel like a pipeline instead of a gamble.
Where it stops
Captured motion is raw material, not a performance. The solve does not know your character, your camera or your intent - it knows silhouettes. Treat the result as a first pass from an enthusiastic junior animator: direction, timing and contact decisions are yours. What capture does remove is the blank-page problem of blocking a full body move from scratch, and that is usually the difference between an animated shot and an imagined one.