Home / Blog / Reading the Numbers
Reading the numbers: symmetry, harmony, jaw
Open the PSL panel after a scan and you'll see five bars, each marked 0–100. They look authoritative, but a number on a screen isn't the same as a verdict. This is what each bar actually measures, what makes it move, and how much it really matters.
The pipeline in one paragraph
A single video frame is passed through a facial keypoint model — we use Google's MediaPipe FaceMesh, which produces 478 3D landmarks per face in around 25 ms. Those landmarks become distances and angles, and those distances and angles become five sub-scores. Everything happens in your browser; nothing is uploaded.
Symmetry — the easy one
We mirror the face along its midline (the vertical line through the nasal bridge and chin) and compare matching landmark pairs. The smaller the average distance after mirroring, the higher the score.
Two notes:
- Almost no face hits 100. The human face is naturally asymmetric — most adults sit between 65 and 85.
- Camera tilt, jaw clench, and asymmetric smiles can drop your score 10–15 points temporarily. Resting face for the scan avoids that.
Harmony — the rule-of-thirds one
Renaissance-era proportion theory divides the face into three equal vertical compartments: hairline → brow, brow → nose tip, nose tip → chin. Harmony scores how close yours come to equal.
Faces with a long mid-face (brow → nose tip) tend to score slightly lower; faces with very short lower thirds (nose tip → chin) tend to score lower too. There's nothing you can do about either short of cosmetic surgery, so this is the most "fixed" sub-score on the panel.
Jaw — the lighting trap
We grade two things:
- The gonial angle — the corner where the jaw turns up toward the ear. Sharper, lower angles read as more masculine; softer, higher angles read as more youthful.
- The lower-face width-to-length ratio. A narrower lower face for its height pulls the score up; a square shape pulls it slightly down.
The trap: jaw definition is the most lighting-sensitive metric on the panel. Soft overhead light flattens it; key light from above-and-side accentuates it. The same face can move 15 points in either direction without changing at all.
Skin — the volatile one
Skin grades local texture variance. We sample patches across the forehead, cheeks and chin and look at how uniform they are. Blemishes, redness gradients and visible pores all reduce uniformity.
This sub-score is the most reactive one on the panel. A clean frame in good light from a high-resolution camera scores 5–10 points higher than the same person on a 720p webcam in mixed lighting. Camera quality changes skin scores more than skincare does.
Canthal tilt — the famous one
Covered in detail in the dedicated canthal tilt article. Short version: the angle between the inner and outer eye corners. Slightly positive (outer above inner) usually reads as more dominant on camera, which the audience tends to reward.
How the bars combine into one number
The five bars are weighted (harmony and symmetry slightly more than skin and tilt), normalized to 0–100, and mapped to a 1–10 PSL band. The mapping isn't linear — the difference between PSL 7.0 and PSL 8.0 is much larger in raw score terms than between 5.0 and 6.0. That's by design: most people sit in the middle of the distribution.
What the panel doesn't see
A still frame can't capture animation. Two faces with identical PSL scores can read very differently in motion — micro-expressions, gaze direction, posture, vibe. That's why the actual mog battle is audience-judged, not score-judged. The numbers are a starting point, not a verdict.
Run a scan and see your own bars.
Open The Lab →