Face analysis in 2026: the complete guide

The four pillars of a real face analysis

Any serious face analysis engine reports on four families of metrics.

• Geometry — thirds, fifths, ratios, angles, symmetry
• Skin — texture, tone evenness, pore visibility, hyperpigmentation
• Dimorphism — how strongly the face reads as masculine or feminine
• Harmony — how the sub-metrics compose into a coherent whole

Geometry — the math of your face

Landmarks are placed on the face, normalized for pose, and reduced to ratios. Facial thirds, fifths, the golden ratio, canthal tilt, chin projection, and midface ratio are the six that carry the most predictive weight in almost every dataset.

Skin — the highest-ROI category

Skin quality is the fastest-moving lever in any face analysis report. Retinoids, sunscreen, sleep, and a caloric deficit will move a skin sub-score more in 12 weeks than any surgical intervention will move geometry.

Dimorphism — masculine vs feminine features

Dimorphism scores how strongly a face expresses sex-typical features: brow ridge, jaw width, and midface length for men; softer jaw, larger eyes, and shorter chin for women. High dimorphism correlates strongly with 'attractive within your sex' ratings across cultures.

Harmony — the composite that matters

A face with perfect individual ratios can still read as off if the parts do not fit together. Harmony is a learned metric — the model is trained to score coherence across the whole face rather than reward individual traits.

How to read your report

Ignore the headline number. Look at the two lowest sub-scores. Those are your roadmap for the next quarter. The gap between your highest and lowest sub-score is a better predictor of perceived attractiveness than the composite.