Saundarya AIBack to the site

Guide

AI skin analysis on darker skin tones

AI reads darker skin less reliably than lighter skin, and the research saying so is not disputed. The size of the gap depends on what is being asked: judging a disease from a photo is where models fail badly, while reading surface traits like oiliness or pigmentation is a far easier problem.

Written by the Saundarya AI team. General cosmetic guidance; not reviewed by a doctor and not medical advice. How we write our guides

Last updated

What the research found

Reviews of diagnostic dermatology models have repeatedly found lower accuracy on deeper skin. One widely cited figure has a skin-image model scoring around 17 per cent on the darkest Fitzpatrick type against roughly 70 per cent on lighter skin. The cause is training data, not skin.

Those numbers describe models asked to name a disease from a photograph. That is a much harder task than the one a cosmetic app performs, and the two should not be quoted as though they were the same finding.

Why cosmetic analysis is a different problem

Telling oily skin from dry, or finding a patch of uneven tone, depends on contrast within one face rather than on comparing that face to a library of diagnosed cases. It is a measurably easier task, and it degrades far less on deeper skin than disease classification does.

This is not an argument that cosmetic tools are unbiased. It is a reason to ask what any given app is actually doing before you accept a number from it, and to treat diagnosis claims with much more suspicion than skin-type claims.

What to ask before you trust an app

Ask whether it publishes accuracy by skin tone, whether it shows the evidence for each finding, and whether it claims to diagnose anything. An app that reports one overall accuracy figure with no breakdown has most likely not tested across the range of skin it is being used on.

  • Does it publish results across all six Fitzpatrick types, or one blended number?
  • Can you see the evidence for a finding on your own face, or only a score?
  • Does it stop short of diagnosis, and say so plainly?
  • Does it recognise concerns common in deeper skin, such as post-inflammatory hyperpigmentation?

Where Saundarya AI stands

Saundarya AI reads surface traits and builds a cosmetic routine. It does not diagnose conditions and does not claim to. Every finding is shown with the visual evidence behind it, so a result that does not match what you can see in a mirror is one you are meant to reject.

We do not publish accuracy figures broken down by Fitzpatrick type. Until we do, treat the report as cosmetic guidance you can check against your own face, not as a measurement.

Get a routine for your own skin

Saundarya AI is an AI skin analysis and grooming app that reads a single selfie and returns a personalised report: face shape, skin type, detected skin concerns with the visual evidence behind them, a morning and night skincare routine, product picks at three budgets, and an 8-week improvement plan.

Generic advice is a starting point. A scan reads tone, texture and pigmentation on your own face on your own face, shows the visual evidence and a confidence score, and builds a morning and night routine around it. The first full analysis is free.

This is general cosmetic guidance, not medical advice. See a dermatologist for skin that is painful, spreading, scarring or not improving after 8-12 weeks.

Related guides