Your webcam is now a four-way gaze controller

What it does

gaze-detection is a thin JavaScript wrapper around TensorFlow.js’s face-landmarks-detection model. It pulls frames from a webcam, infers facial geometry, and collapses that data into four discrete directions: RIGHT, LEFT, TOP, and STRAIGHT. The result is a coarse-grained eye-tracking API that runs entirely in the browser without native binaries or specialized hardware.

The interesting bit

The library bundles the whole ML pipeline—model loading, camera setup, and frame inference—into a few method calls that return human-readable directions instead of raw landmark arrays. The README also claims it runs on mobile browsers, relying only on a standard webcam rather than specialized eye-tracking hardware.

Key highlights

• Wraps TensorFlow.js face-landmark inference into four cardinal gaze directions
• Consumes any standard webcam feed via a <video> element
• Claims to work on mobile browsers, not just desktop
• Inspired by the Android accessibility app Look to speak
• Returns discrete string labels (RIGHT, LEFT, etc.) rather than raw coordinates

Verdict

Worth a look if you are prototyping hands-free web interfaces, accessibility tools, or creative-coding experiments. Skip it if you need fine-grained pupillometry or sub-degree gaze accuracy; this is directional detection, not precision instrumentation.