What does a face rating score mean?

A score summarizes a method, not a whole person.

In FaceRater, a face rating is a 0–10 harmony score: how close the proportions in one photograph sit to the common range of adult faces, and how evenly the two sides mirror each other. A typical photo reads 7.5, and the percentile beside the score says how many reference faces the photo is ahead of. That is the useful definition to keep beside your result: it tells you what this tool measured, rather than claiming to know how attractive everyone finds you.

There is no shared scale across all face-rating websites. An 8 here is not necessarily equivalent to an 8 elsewhere. Another service may use different landmarks, an “ideal” such as the golden ratio, or an AI model’s subjective response. Compare the method before comparing the number.

Five parts feed the score.

The score combines 17 proportions in five parts: facial thirds (25%), facial fifths (25%), the eye area (20%), jaw and chin (20%) and symmetry (10%). The local model locates 478 facial landmarks and corrects head roll, turn and nod before measuring them. Symmetry carries the least weight because it changes most with camera angle and expression.

Each proportion is compared with 2,560 front-facing adult photos from FairFace, a public research set balanced across seven perceived race groups and two genders. Values near the common range of the closest group score highest; more open eyes and smaller left–right differences also score higher. The combined result is ranked against the same photos, and the 10-point score places that rank on a curve: 7.5 is typical, the top tenth reaches 8.4 and the top hundredth 8.9.

Read the sample part by part.

Our synthetic adult example scores 8.7/10, ahead of about 95% of reference faces. Its symmetry (9.0) and facial thirds (8.4) are its strongest parts. Its jaw is narrow relative to cheek width, which the score does not mark down, so its jaw and chin (8.3) read well. Its facial fifths (7.0), the spacing of eyes, nose and mouth across the face, sit furthest from the common range. That is a description, not a flaw: glasses width, brows and hair at the temples change how that spacing reads.

The percentile is a rank against the reference photos, not against everyone, and 8.7/10 does not mean 87% attractive. Being typical is not the same as being attractive either: striking faces often differ from the average in distinctive ways. The ten extra measurements, such as a 99.5% paired-landmark alignment, are separate readings and do not change the score.

Use the result in this order.

Check the photo and its overlay first, then the part with the most room, and only then compare a second photo. If the lines miss an eye corner or the outline, the resulting number deserves less confidence. If the result is marked approximate, retake the photo straight on with a relaxed expression. In the part with the most room, read its proportions and common ranges. A second photo is only worth comparing when it was taken under similar conditions.

• Check that the photo is frontal and the eyes, mouth and outline are visible.
• Open each of the five parts to see which proportions drive it.
• Retake an approximate or unclear photo before interpreting a small score difference.
• Use styling suggestions to explore a look you like, rather than chase a decimal.

The AI stylist adds interpretation, not a better score.

The optional AI stylist adds more detail and practical suggestions; using it does not raise or recalculate the original score. Your free harmony score, its five parts and ten more measurements remain available without payment. The full style report also includes visual styling concepts and a four-week plan.

If your question is about one feature, you can first use the free canthal tilt, symmetry or proportion tools. They inspect the same local scan. If you want to understand the exact calculation, the methodology page explains the reference set, the weights and the known limits.