How Face Shape Detection Works
This implementation measures a visible upper-forehead landmark, cheekbones, jaw corners and chin. It compares mesh height-to-width, forehead-to-cheek width, cheek-to-jaw width and jaw-corner angle with six numerical profiles. The closest profile becomes the headline label.
A match is not a probability
A profile receives a similarity score. Several shapes can be close at once. A strong second-place result is a reason to compare styling suggestions from both groups, rather than to force a permanent label.
Make the input comparable
Use one clear, unfiltered photo with the camera at eye level. Keep the whole face visible and use a relaxed expression. The local scan uses MediaPipe landmarks, so framing and image quality affect the anchors from which ratios are calculated.
Read a result in context
Review the values and the explanation beside them. A match score and a measured ratio answer different questions. Neither predicts how another person will judge you. If the photo fails the pose check, choose a better frame before sharing a result.
Try the relevant tool
Review a photo-based result and the method behind it. Optional AI processing begins only after your permission.
Open the tool More guides