Camera vision · seven scenes · runs on your device
A camera model finds 21 points on each of your hands, sixty times a second. Your right hand builds the thing; your left brings it to life. Close them and it all runs backwards. Seven scenes to try — and in Classroom, every hand in the room grows its own numbered flower.
Everything happens inside this browser tab. No video is uploaded, recorded or sent anywhere — there is nowhere for it to go.
Works best in good light, with your hands about an arm's length away.
The model returns a coordinate for every knuckle, joint and fingertip — 21 landmarks on each hand, plus which hand it thinks it is. Turn on Show what the model sees to watch them move.
Openness is just distance. Each fingertip's distance from the wrist is divided by the size of your hand, so it works whether you are close to the camera or far away. Five fingers, averaged, smoothed.
Nothing here is a video effect. Two numbers between 0 and 1 decide how tall the stem is, how many leaves have unfurled and how far the petals have opened. Everything else is ordinary canvas drawing.
This is the same shape of problem as reading a gesture, a posture or a production line: a model gives you points, and you decide what they mean. The interesting work is in that second step.
Your camera stays yours. The model file downloads to your browser and runs there. No frame is ever uploaded — this page has no endpoint to send video to, and its security policy blocks it from gaining one. Close the tab and nothing remains.
We teach AI the hands-on way — for schools, teams and whole organisations across South and Southeast Asia.
Book a call →