frame in
Your camera, untouched
A raw video frame is grabbed from your webcam via getUserMedia. It never leaves the page — there's no server in this loop.
Runs entirely in your browser
A green screen you were never wearing.
Point your webcam at this page. We'll trace you out of your room in real time, no fabric, no lighting rig, no upload — just a segmentation model doing the work your walls used to.
Click “Start camera” to begin. Nothing leaves this tab.
fps —
·
model: idle
What's actually happening
segment
BodyPix finds the person
TensorFlow.js runs the BodyPix model on-device over WebGL, scoring every pixel for "person" vs. "background" — the same job a green screen used to do with color, done here with a trained eye.
frame out
The background is rewritten
Background pixels get blurred, swapped, or dropped, frame after frame, and the composite is painted straight to a canvas — 25–30 times a second on most laptops.
Questions people actually ask
Do I need a green screen for this to work?
No. BodyPix detects a person against any background — a bookshelf, a bedroom, a hallway — because it's trained on human shapes, not a specific color.
Is my video uploaded anywhere?
No. Everything runs in this browser tab. Close the tab and the feed is gone; there's nothing on a server to delete.
Why does the edge sometimes flicker?
Segmentation confidence dips in low light or fast motion. Even lighting on your face and a still-ish frame give the model the cleanest read.
Will this work on my phone?
Most modern phone browsers support WebGL and getUserMedia, so yes — performance just depends on the device's GPU.