The FaceRater measurement method.
See how your score works.
Measured. Explained. In perspective.
MediaPipe estimates 478 facial landmark positions. FaceRater uses those positions to measure paired alignment and visible width ratios, after rotating the face into an upright coordinate frame.
+ Proportions × 20%
+ Eye spacing × 15%
+ Jaw balance × 15%
+ Lip balance × 10%
+ Nose balance × 5%
These are FaceRater product weights and broad reference bands. Ratio scores vary continuously around each product band’s midpoint. These choices are not a scientifically universal attractiveness formula. Individual readings and the exact scoring version stay attached to your report.
Before a report is created
New scans screen the prepared photo for clipped facial features, extreme exposure, obvious blur, enlarged pixels and eye regions with very little visible detail. A failed check asks for another photo before a score is shown or saved. These limited checks can miss problems; use a clear, unobstructed front photo and follow any retake note.
What the model does not tell us
A front photo cannot reliably measure nose projection, a profile jaw angle, the true hairline, clinical skin conditions or attractiveness in real-life interactions. FaceRater reports these limits instead of inventing a measurement.
Why photos differ
Lighting, camera distance, head turn and expression change a 2D image. Compare photos in the same conditions. A small score change does not prove that a face changed.
Read the MediaPipe Face Landmarker documentation ↗
About the example portrait
The demonstration uses an AI-created adult portrait with a clear, front-facing view and hair tied back. Its readings are produced by the same versioned landmark geometry used for an uploaded photo. The six hairstyle examples are actual Sunburst High edits of that same source. It is an illustration of the product, not a customer or a testimonial.