Vision crowd read

Drop a video to read the crowd

Every person in frame gets a box and a persistent ID. Where a face is visible, the model also estimates age and gender. Nothing leaves this browser — the file is never uploaded to the server.

MP4, WebM or MOV. Works best on footage under two minutes.

Loading detection models
00:00.0 / 0 in frame
00:00.0 00:00.0

Age and gender are estimated from visible faces by a small on-device model. Treat them as a rough distribution across a crowd, not as a fact about any one person. Accuracy drops sharply with distance, motion blur, masks and faces turned away from the camera.