Two parts to the analysis
Google’s MediaPipe Face Landmarker estimates 478 three-dimensional landmarks from a photograph. Our geometric engine then measures selected points and compares ratios with six shape prototypes. MediaPipe does not supply the face shape categories, and the style recommendations are written in a local database, not generated by an AI service.
Quality before classification
The tool checks that there is exactly one detected face, that it is sufficiently large and inside the frame, and that the head is approximately frontal. Initial pose limits are ten degrees for yaw, pitch and roll. Face-region exposure and sharpness checks reject clearly unsuitable images. These engineering thresholds are approximate and not validated across every camera or population.
Landmarks can still be estimated behind hair, glasses or other coverings. Our contour checks cannot reliably detect every occlusion, so the tool asks you to confirm that your whole face is unobstructed. If the overlay does not follow your features, retake the photo instead of relying on the result.
Transparent, approximate matching
Normalized face length, estimated upper-face width, jaw width and a lower-face contour angle are compared with documented prototypes. Weighted distances become independent match scores from 0 to 100. The scores do not add up to 100 and are not probabilities. If the leading scores are close or all weak, the result is explicitly labeled mixed.
Suggestions, not rules
Hair and glasses recommendations describe possible visual effects: adding height, introducing softer curves or drawing attention to a particular region. They do not prescribe what anyone should wear. Style filters are selected by you; gender and other sensitive characteristics are never inferred.
What this release does not claim
There is no claimed clinical accuracy, demographic fairness validation, beauty rating, identity recognition, hairline detection, or virtual try-on. A photograph is a limited view of a three-dimensional person. Treat the result as a useful conversation starter and compare it with your own observations.
Software and model credits
MediaPipe Tasks Vision 0.10.32 and the official Face Landmarker float16 version 1 bundle are distributed locally with their Apache 2.0 license. See the official model documentation. The interface illustrations are original schematic drawings, not examples of actual analysis.
Try the free tool ↗