Can an AI app diagnose my skin condition?

No — and you should be wary of any tool that says it can. A diagnosis is a clinical judgement made by a qualified professional who can examine you in person, take a history, feel the lesion, and order tests like a biopsy when needed. An app sees a photograph. It can describe what's visible and suggest what that pattern is often associated with, which is genuinely useful for orientation — but that isn't a diagnosis.

Then what is a skin assessment actually for?

Orientation and language. It helps you understand what you're looking at, gives you vocabulary for the thing on your skin, points you toward general care that's usually appropriate, and — importantly — flags when something warrants an in-person opinion. It's a first step that makes the next step clearer, not a replacement for that step.

How accurate is the AI behind it?

dermatrix.life is built on MedGemma, an open, general medical model from Google. MedGemma models have been reported at roughly 72% on a dermatology multiple-choice benchmark — respectable for a general medical model, and nowhere near a guarantee. It isn't a dermatology-specialised classifier, it isn't a medical device, and roughly-three-in-four on multiple-choice questions is not the same thing as being right about your skin.

Can it tell me whether a mole is skin cancer?

No. Nothing automated should be trusted for that, and we won't pretend otherwise. Any mole or spot that is new, changing, asymmetric, growing, itching, bleeding, or not healing needs to be examined by a doctor in person. If that's your question, skip the app and book the appointment.

Can AI Diagnose a Skin Condition? (An Honest Answer)

"AI skin diagnosis" is one of the phrases people actually type into a search bar. We see it, and we deliberately don't use it — because the honest answer to the question behind it is no.

That deserves a real explanation rather than a disclaimer, so here it is.

What a diagnosis actually is

A diagnosis is a clinical judgement, made by a qualified professional, using far more than a picture:

An in-person examination — under good light, from angles, sometimes with a dermatoscope.
Touch. Is it firm or soft? Fixed or mobile? Tender? Rough? A photo cannot tell you this.
A history. How long has it been there? What's changed? What medications are you on? What runs in your family? What did you try already?
Context and testing. The rest of your skin, your general health, and when needed a biopsy, a swab, or blood work — the only things that can actually settle certain questions.
Responsibility. A clinician is accountable for that judgement and can follow it up.

An automated tool has exactly one of those inputs: an image. Calling its output a diagnosis wouldn't just be inaccurate — it would encourage people to act on it as if it settled something.

What an automated assessment can honestly do

Quite a lot, actually — as long as we're clear about what kind of help it is:

Describe what's visible in plain language, without jargon.
Give you the vocabulary. Knowing that the thing you have is called a "comedone", "milium", or a "plaque" makes everything else you read — and your eventual appointment — dramatically more useful.
Orient you. What families of conditions look like this? What generally helps that kind of thing?
Point you to sensible general care — the same evidence-based basics we write about in /learn.
Flag when to stop reading and go see someone. Arguably the most valuable output of all.

That's the honest job description: a private, informational first step that makes your next step clearer. Not a verdict.

What's under the hood, and why we say what we say

dermatrix.life runs on MedGemma — an open, general medical AI model published by Google (technical report; model card). Three things follow from that, and we'd rather state them up front:

It's a general medical model, not a dermatology specialist. It wasn't built as a dedicated skin classifier, and we don't present it as one. On a dermatology multiple-choice benchmark, MedGemma models have been reported at around 72% — a reasonable showing for a general model, and plainly not a number you should stake a health decision on.
We're independent of Google. We use the open model. dermatrix.life is not made, endorsed, or reviewed by Google.
There is no clinician in the loop. The assessment is fully automated. Nobody reviews your photos before you get your result — which is part of why it's private and inexpensive, and also precisely why it can't be a diagnosis.

We go deeper into the accuracy question in Is AI Skin Analysis Accurate?, and into the mechanics in How AI Skin Analysis Works.

Why photographs are a genuinely hard input

Even a perfect model would be limited by what a photo carries:

Lighting and color shift. Phone cameras auto-correct, and redness and pigment are exactly what gets adjusted away.
Depth is invisible. Raised, flat, or indented can look identical straight-on.
Scale is ambiguous without a reference — 2 mm and 2 cm photograph the same.
Skin tone changes presentation. Many conditions are described and illustrated primarily on lighter skin; redness reads very differently on deeper skin tones, and datasets have historically under-represented them.
Some things simply don't photograph. Itch, pain, burning, how it feels to touch.

You can meaningfully improve what a tool has to work with — see what photos to take for a skin assessment. But no framing beats the ceiling: a photograph is a slice of the information a clinician uses.

Where an AI tool is genuinely the wrong tool

Skip the app and see a doctor if you're asking about:

A mole or spot that is new, changing, asymmetric, growing, itching, bleeding, or won't heal. This is the single most important carve-out on this site — see skin cancer warning signs and when to worry about a mole.
Anything painful, rapidly spreading, blistering, or infected-looking — warmth, swelling, pus, fever.
A dark streak under a nail, or a non-healing sore anywhere.
A rash with systemic symptoms — fever, joint pain, feeling unwell.
A baby or young child you're unsure about.
Anything that's already been treated and isn't improving.

"Get this looked at in person" is a legitimate result, and often the most valuable one.

The short version

Can AI diagnose a skin condition? No. Can a well-built, honest assessment help you understand what you're seeing, learn the right words, take reasonable care of it, and recognise when to book an appointment? Yes — and that's a real, useful thing that doesn't require overclaiming.

We'd rather tell you what this can't do and be trusted on what it can.

Curious what an honest, informational read of your skin looks like? A dermatrix.life skin assessment reviews photos you upload — over an encrypted connection, to a private bucket, deleted within 24 hours — and returns a private, plain-language summary. It's informational only, not a diagnosis, fully automated with no clinician review, and never a substitute for seeing a professional.