AI Reads Wrinkles, Pigmentation and Redness From a Face Photo

A 2025 study in the Journal of Cosmetic Dermatology: one AI model graded skin type, pigmentation, redness and wrinkle severity from ordinary face photos with 85% accuracy.

Medical spas and aesthetic clinics see more clients every year, and many skin assessments are now done by non-dermatologists. A 2025 study in the Journal of Cosmetic Dermatology set out to give them an objective helper.

The researchers built a dataset of 3,662 colour photographs of faces, each graded by a dermatologist on five scales: Fitzpatrick skin type, hyperpigmentation, redness and wrinkle severity. They then trained and compared 15 different AI set-ups.

The best model read all of these at once from a single face photo, with an average accuracy of 85% across every scale. It was most confident at the clear ends of each scale, exactly where a treatment decision is easiest to make.

This is the kind of reading a consultation starts with. A face photo in, a structured picture of tone, redness and lines out, so the conversation moves straight to which treatment or product fits.

Independent research on AI skin analysis in general, not a test of DiemSkin.