F/S FACE STUDYLOCAL ANALYSIS · NO SIGN-UP

YOUR PROPORTIONS, EXPLAINED

Face shape detector.

One face. Four measurements. A result you can understand.

A clear view of you

ON YOUR DEVICE

Choose one front-facing portrait. Your photo stays in this browser.

How to take a useful photo
  • Face straight ahead. Turning your head compresses one side and changes width ratios.
  • Use even light. Shadows can obscure your cheek and jaw outline.
  • Pull your hair back. We need to see your forehead, hairline and jaw.
  • Avoid wide-angle selfies. Step back and use a normal lens to reduce perspective distortion.
  • Keep a neutral expression. Smiling changes the cheek and jaw outline. Use one person only.
YOUR RESULTWaiting for your input

More than a shape label.

Your four measurements and the reasoning behind your closest match will appear here.

NO PHOTO STORAGE
NO ACCURACY CLAIMS

OPEN METHOD

No hidden verdict.

A transparent prototype for styling, not a validated scientific classification.

See every ratio, threshold & limitation

F = forehead width; C = cheekbone width; J = jaw width; L = hairline-to-chin length. Ratios use the values displayed in your result.

For each interval [a,b], violation = max(a − ratio, 0, ratio − b). A shape's distance is the mean of its three violations. The smallest distance wins. Equal distances use the listed order. If the next distance is within 0.04, we flag a close call and show alternatives; if the best distance exceeds 0.10, we flag a poor fit. These margins are our explicit prototype choices, not validated thresholds.

Adapted from PFPMaker's published ratio rules and the proportion approach on Face Shape Tool. We added interval bounds and forehead constraints to make comparisons deterministic. No claim that these competitors or Google validated our classifier. Round vs square also depends on jaw curvature, which four lengths cannot establish.

Photo landmarks: Google MediaPipe Face Landmarker, exactly 478 points. We use pairs 103–332 (forehead proxy), 234–454 (cheekbone proxy), 172–397 (jaw proxy), and 10–152 (upper forehead to chin proxy). These endpoint choices are our implementation. Click your actual hairline and check the endpoints before accepting a photo result. Pixels are not centimeters: no real-world scale can be inferred from an uncalibrated portrait.

The model detects landmarks; our rules produce the shape label. Neither this rule classifier nor its photo measurement choices have been independently validated. Perspective, hair, expression and endpoint placement can change the result.

Local limits: one photo, at most 10 MB and 24 megapixels; analysis image resized to at most 1600 pixels on its longest side. These are our memory limits, not a model accuracy specification. No daily limit, account, analytics or image storage. Only static app files are read from this machine; the server rejects uploads.