The Science Behind Facial Harmony Scores

“Facial harmony” is one of those terms that gets thrown around without much rigour. It sounds scientific, but in most consumer apps it's a black box that produces a number designed to make you feel good. At Objektiv we treat it differently — as a measurable composite of geometric features, each anchored to a rubric, each contributing a known weight to the final score.

What “harmony” actually measures

Facial harmony, in our model, is the agreement between the major spatial relationships of a face: the rule of thirds (forehead, mid- face, lower face), the rule of fifths (horizontal eye spacing), jaw definition relative to the cheekbone, chin projection relative to the lip line, and bilateral symmetry. None of these is sufficient on its own. A perfectly symmetrical face with weak chin projection can still feel off; a strongly asymmetric face with great proportional balance often reads as charismatic rather than wrong.

Why we use a real bell curve

A score is only useful if it can distinguish people. Most apps compress everyone into the 7–9 range, which means the scores carry no information. Objektiv anchors 5.0 to the population average. About 68% of users sit between 4 and 6. A 7 is genuinely above average; an 8 is rare; a 9 demands explicit justification from the model and is rejected server-side if the rationale doesn't hold up. The point isn't to be punishing — it's to make the score load-bearing.

Anti-inflation rules

Vision models trained on internet data have a strong prior toward flattery. Left unchecked, they'll happily call every face an eight. Our system prompt enforces a counter-prior: the model must spend its tokens on what's improvable, not on what's already working. It can't award high scores without naming the specific geometric feature responsible. And we validate the result before it reaches your dashboard — scores above 8 without documented justification are rejected and re-rolled.

What the score is not

A harmony score is not an attractiveness score, not a worth score, and not a fixed property. It measures geometric agreement at a single moment, in a single photo, under specific lighting. Hair, beard, posture, expression, and angle all change it. Two analyses a week apart can produce different scores for the same person, which is a feature: it forces you to think about what you actually control, not about a number tattooed on your face. The companion articles What Is a Color Season and Why Does It Matter? and How AI Can Help You Find Your Best Hairstyle go deeper on two of the variables that move it the most.

From score to action

The score itself is the boring part. The interesting part is the breakdown — which sub-component is dragging the composite down, and what intervention would move it. A 5.4 with weak jaw definition and strong thirds is a different person, with a different action plan, than a 5.4 with strong jaw definition and a long lower third. The dashboard surfaces both, ranks the recommendations by time-to-impact and cost, and tells you which interventions are reversible (haircut, beard) versus non-reversible (cosmetic procedures we generally discourage).

Read more

See how the platform fits together on the About page, or check the pricing page for tier details.