Seventeen points are enough
In 1973, Gunnar Johansson filmed people with a few points of light on their joints, and viewers recognized a person walking in a fraction of a second. Try it: your camera drives a figure of points, and its twin follows you, one step behind.
Before you turn on your camera
- Everything is computed in this browser. No image or point leaves your device.
- The video is neither shown nor recorded: only the 17 points are drawn.
- About 20 MB downloads once, from this site, when you ask for it.
- The Stop button turns the camera off right away.
Camera off. You're watching a recording.
- Points tracked
- –
- Twin delay
- –
Real recording: CMU subject 02, trial 01, 17 markers.
What this has to do with your clone
A clone doesn't need to copy everything. It needs the right points: your decisions, your sources, your exceptions. This page only explains the idea: your clone learns nothing from it, not your movements, not your image.
See how a clone captures your decisionsWhat happens technically
A pose model (MediaPipe, Apache 2.0 licence) estimates 33 points on each frame in your browser. We keep 17, the same ones as the CMU recording. The twin shows those points 220 milliseconds later.