AI Technology

How AI Facial Landmark Detection Works (In Plain English)

· 9 min read · Face Shape Detector Team

When you upload a photo, a neural network called a face mesh model examines the image and returns 478 coordinates — one for each anatomical landmark it has been trained to find: the corners of your eyes, the arc of your jaw, the peak of your hairline, the tip of your chin.

The model runs entirely in your browser via WebAssembly — the photo is processed on your own device and never touches a server.

Landmarks alone do not name a face shape. The next stage is measurement: the app computes distances between specific landmark pairs and normalizes them into ratios, so photo resolution and camera distance stop mattering.

Classification is a nearest-profile match. Each face shape has a signature ratio profile: oblong means length far exceeds width; triangle means the jaw out-measures the cheekbones. Your ratios are compared against all seven profiles and converted into probabilities.

The confidence score reflects how decisively your measurements match the winning profile. A 97% oval is unambiguous; an 86% result usually means you are a blend of two shapes.

Not sure of your face shape?

Find out in seconds — free, private, analyzed entirely on your device.

Detect My Face Shape

Related articles