By · Updated

Head Tracking Demo Browser

Face-Tracked 3D Window Illusion, Layers Slide at Different Speeds

No ratings yet

OneTask is free, with no sign-up, and built by one person.
If this tool saved you time, a coffee helps keep new ones coming.

How the Head-Tracked Parallax works

This head tracking demo browser page moves a 3D scene as you move. Press Start camera, sit back a little, and move your head from side to side and up and down. The layered scene shifts as you move — near shapes fast, far shapes slowly — so the screen behaves like a window you are looking through rather than a picture you are looking at. Turn the strength up for a more exaggerated effect and untick the camera preview to lose the thumbnail. Snap saves the frame.

Head tracking demo browser: what to know first

A flat screen normally shows the same picture no matter where you sit. Head-tracked parallax breaks that rule: the picture is redrawn for the position of your head, so the screen behaves like a window into a space behind the glass. In 2007 Johnny Lee demonstrated the idea with a Wii remote clipped to a pair of glasses, and the video became one of the most watched technology demos of its year. This page does the same thing with nothing but a webcam and a face-tracking model running in your browser.

Why head-tracked parallax looks 3D

When you move your head, nearby objects slide across your view faster than distant ones. That is motion parallax, and it is one of the strongest depth cues the brain has. Stereo vision, the difference between your two eyes, fades as a depth cue beyond a few metres because the eyes are only about 6.3 cm apart; motion parallax keeps working at any distance as long as you move. It is how a one-eyed person judges depth, and how a pigeon does it by bobbing its head.

The page fakes it with four layers of discs. Each layer is shifted opposite to your head movement by an amount proportional to how near it is meant to be: the nearest layer moves the most, the farthest barely at all. Because the shift follows your own motion, your brain reads the layers as sitting at different depths, even though every disc is flat and exactly the same distance from your eyes.

Where the technique is used

How to read the result

The only readout is Head, your position in the camera frame, as a percentage across and down from the centre. Zero across and zero down means you are centred; +50 % across means you are halfway from the centre to the edge of the frame. It shows Not in view when the tracker loses your face, which usually means poor light, turning too far or leaving the frame. The green square on the camera preview marks the point being tracked: the midpoint between the outer corners of your eyes.

The layers follow that point with some smoothing, so tracking jitter does not make the scene tremble. The cost is that a sudden move settles over a few frames rather than snapping into place.

Getting the best illusion

Mirror and why direction matters

The Mirror toggle is on by default, so moving your head left reveals more of the right side of the scene, exactly as looking through a window would. Untick it and the coupling flips: the scene moves the wrong way, and the illusion collapses immediately. It is a quick demonstration that the depth comes from your brain predicting how the world should move, not from the picture itself.

Limits

The effect works for one viewer. Anyone looking over your shoulder sees a scene sliding for no reason, which is the main reason head-coupled perspective never became a standard display feature. The page tracks only left-right and up-down; a full head-coupled display also tracks distance so the view widens as you lean in, but a webcam estimate of distance from face size is noisy enough to make the scene visibly breathe.

The face model runs in WebAssembly in the page and is downloaded once, then cached. The camera stream is never uploaded. The only data the page uses is a pair of head coordinates, replaced every frame. Snap saves the current frame as an image. For more face-tracking effects, try the face mesh filter.

FAQ

Why does the effect look right only for me?

Because it is drawn for one viewpoint — yours. Anyone else looking at the screen sees a scene shifting for no reason, which is the inherent limit of head-coupled perspective on a flat display and why it never became a mainstream feature.

Can I use my own image or a 3D model?

Not on this page — the scene is fixed. The technique is simple enough that the layered approach here transfers directly to any set of images at different depths, and full 3D scenes do the same thing by moving the camera in the renderer.

Why does it lag or judder?

The tracker runs at camera frame rate, and the smoothing adds a few frames of settling. Judder usually means the face was lost and re-found; better light and facing the camera help. Slow, deliberate head movement gives the cleanest illusion.

Does it work with glasses?

Yes. The model tracks the whole face and is trained with glasses in the data; only sunglasses dark enough to hide the eyes entirely give it trouble, and even then it usually holds.

Does anything get uploaded?

No. Tracking runs locally in the page, and the head position is a pair of numbers that never leaves your browser and is discarded every frame.

See also: all Video & Screen tools, Images, Audio, PDF, SVG, Simulations & Screensavers, Generative Art, Chart Makers, 3D Model Viewers, Drawing Pads, Mockups & 3D Making, Diagram Makers, 3D Science, Tracker Music.

Related tools