Navigate this site with your hand.
Turn on the camera in the corner. Pinch and drag to scroll this page. Open your palm and flick sideways to move to the next one. Nothing leaves your machine.
Two gestures, four pages.
Every page here is deliberately long, because scrolling is the half of this that is easy to get wrong. A swipe is a single discrete event and forgiving of noise. A scroll is continuous: it has to track your hand frame by frame, feel proportional, and stop the instant you release the pinch. If it lags or overshoots, you notice immediately.
Pinch drag
Touch thumb to index, move your hand up or down. The page follows like a sheet of paper under your fingers.
Open palm swipe
Show five fingers, flick left or right. Right goes forward, left goes back.
Keyboard
← → change page, ↑ ↓ scroll, exactly as they always did.
Why pinch and not point.
Pointing is the obvious choice and the wrong one. A pointed finger has no off state: the moment your hand enters the frame it starts issuing commands, and you end up holding your arm in an unnatural pose to avoid scrolling by accident. A pinch has a clear open and closed position, which gives you a clutch — the same reason a mouse has a button rather than scrolling whenever you move it.
The pinch also survives distance. Because the thumb-to-index gap is divided by the width of your palm, the threshold holds whether your hand is filling the frame or sitting a metre back.
What you should feel.
Scrolling should feel like the page is attached to your hand: move down four centimetres, the page moves down a fixed amount, every time. Nothing accelerates, nothing coasts. Inertia is lovely on a trackpad, where the surface tells your fingers exactly where they are, and disorienting in mid-air, where nothing does.
Swiping should feel like a button press, not a drag. Speed is what triggers it, not distance, so a short sharp flick works and a slow wide sweep does not. That distinction is what keeps you from changing page while you are reaching for your coffee.
Limits worth knowing.
One hand at a time. Keep it roughly upright, since finger extension is measured vertically and a sideways hand reads as a fist. Light in front of you rather than behind, or the camera hands the model a silhouette. And there is a lower bound on frame rate below which no amount of smoothing helps — if the tracker is running at fifteen frames a second, every gesture will feel like it is happening underwater.
Scroll down to the bottom of this page and then flick right. The next page explains what the model is actually giving us.