AI face shape detection: how it works
An AI face shape detector doesn't eyeball it — it maps hundreds of points on your face and measures your actual proportions: length versus width, forehead versus jaw, the angle of your jawline. Here's what happens under the hood.
Step 1: landmark mapping
The system finds anatomical landmarks on your face — BeU maps 478 of them: the contours of your forehead, cheekbones, jawline and chin, plus eyes, nose and brows. This turns your photo into a measurement grid, not a picture.
Step 2: real measurements
From the landmarks it computes the ratios that define face shape: face length versus width, forehead width versus cheekbone versus jaw width, and jaw angle — soft and rounded, or angular with corners. These numbers are what separate oval from oblong, square from round.
Step 3: classification
The measurements are matched against the eight shape profiles — oval, round, square, heart, diamond, oblong, rectangle, triangle — trained on thousands of labeled faces. The output isn't a vibe; it's the profile your proportions fit best.
Why it beats the mirror
Mirror-guessing fails for consistent reasons: you're seeing your face in 3D and in motion, your brain normalizes what it sees every day, and most people measure with hair falling over the landmarks. A photo-based measurement is still, standardized and repeatable.
Honest limits
- Hair must be pulled back. Hair over the forehead or jaw corrupts the landmarks. This is the number one cause of wrong results.
- Face the camera straight on, head level. Tilt changes every ratio.
- Even lighting. Hard side-light carves false shadows into the jawline.
- Weight changes read as shape changes. The bone structure is the same, but fuller cheeks genuinely change the measured widths.
How BeU does it
BeU's face scan maps 478 landmarks with your phone camera, measures your proportions, and returns your face shape plus a contour map drawn on your own face, your eye shape and your brow map. Measuring happens on your phone; your photo is never saved. Learn the 8 face shapes.
Questions women ask
How accurate is AI face shape detection?
For proportions, very — landmark measurement is objective. The shape label is partly judgment even between human stylists, so treat borderline results as 'between two shapes' rather than wrong.
Why did an app get my face shape wrong?
Almost always one of: hair over the landmarks, tilted head, or harsh side lighting. Retake with hair back, head level, facing a window.
Can face shape change over time?
Bone structure doesn't, but weight, aging and hairstyle change the measured read. Re-scan when something's visibly different.
Is the BeU face shape scan free?
No — a one-time face scan gives your face shape, contour map and brow map; membership adds the skin check, colors and the rest.