Are face rating tests accurate?
Short answer: they measure proportion accurately and cannot measure attractiveness at all. Below is our entire algorithm, so you can decide how much to trust it.
What it measures
MediaPipe Face Landmarker places 478 points on your face in the browser. We level out head tilt using the line between your outer eye corners, then measure ten proportions from those points: symmetry, facial thirds, five-eye width, eye spacing, width-to-height, jaw to cheekbone, canthal tilt, nose to mouth width, lip ratio and face length to width. Each is compared against a common aesthetic reference value, scoring higher the closer it sits, then weighted into a total.
Weights: symmetry 25%, thirds 15%, five-eye 10%, eye spacing 10%, width-to-height 10%, jaw to cheekbone 10%, and the remaining four at 5% each.
Reference values: five-eye 5.0, eye spacing 1.0, width-to-height 1.9, jaw over cheekbone 0.80, canthal tilt 4°, nose over mouth width 0.70, upper over lower lip 0.70, face length over width 1.45. Thirds are compared against the MediaPipe canonical face model.
Why we convert: the highest mesh vertex is not the hairline, the side points sit slightly outside the skin contour, and the inner eye corner includes the caruncle. Measured raw, the canonical face scores 5.9 on five-eye and 1.43 on eye spacing. We use the official canonical face model as a baseline to convert each measure back onto the textbook scale, and the conversion factors are in the source.
The total: 50 + weighted score × 0.48, giving a range of 50–98.
Where it is accurate
- Rescan the same photo and the score is identical, unlike asking a chatbot about an image and getting something different each time.
- The proportions themselves are measured precisely: landmark error is typically under 1% of face width, so symmetry, thirds and five-eye are reliable geometry.
- Face shape uses explicit thresholds you can verify yourself, see how to tell your face shape.
Where it is not
- A 2D photo has no depth: nose bridge height, cheekbone projection and chin position cannot be measured, and those matter a great deal to how a face reads.
- Skin, expression and presence are not in it: the score describes bone proportion and nothing else.
- Reference values are not truth: five-eye and thirds are widely used references, but preferences differ across cultures and eras, and plenty of universally admired faces are not symmetrical.
- Shooting conditions move it 3–5 points: a wide lens close to your face enlarges your nose and widens your face, tilting your head changes the thirds, and side lighting shifts the landmarks. Different photos of the same person disagreeing is expected.
So how should you use it
Treat it as a signal for which single thing is worth your effort, not as a ranking. The "most worth working on" line and the face-shape advice on the result card are the genuinely useful parts. The number itself is just a number.