Mission 05 of 05
Explorer
AI Face Lab
Turn on your camera and watch an AI model find your face in real time. Draw on the canvas, then watch your drawing wrap around your face as you move.
The webcam lab
Video never leaves this devicecamera · local only
Let's see what the AI sees
Press start, allow the camera, then draw on your screen. Your drawing sticks to your face as you move.
Nothing is uploaded, saved, or shared. Close the tab and the camera stops.
What happens to a frame
- Camerareads frames locally
- Light pixelsa grid of numbers
- Trained modellooks for patterns
- Landmarksa guess, not a measurement
A model is a pattern finder
This model was trained on thousands of face photos. It learned which arrangements of light and dark usually mean “face”, and it looks for those arrangements in your video.
Landmarks are guesses
The dots mark the eyes, nose, ears, and mouth. They aren't measured. They are the model's best guess, which is why they can wobble.
Nothing leaves this device
The video is read straight from your camera into this browser tab. There is no upload, no recording, and no server involved.
Why does AI need so much data?
A model doesn't understand faces the way you do. It learns from examples: show it millions of photos labelled “face” and “not a face” and it adjusts until it gets them right. The catch is that it also learns the biases in that data, which is why face recognition works better for some people than others, and why researchers spend so long checking training sets. This one is small enough to run on a phone, and it still gets things wrong.