Is AI Skin Analysis Accurate? What a Camera Can and Can't See

This week someone on Reddit described a clinic face scan that rated their skin “very good”. It missed the marks, small scars and under-eye lines they could see in the mirror, and their face was shiny with serum and sunscreen at the time.

So is AI skin analysis accurate? Close, most of the time, but not exact.

And what’s on your skin when the photo is taken matters more than most people expect.

What a skin scan is actually reading

A camera doesn’t touch your skin. It reads the light coming off it.

Shine from serum, moisturiser or sunscreen bounces light straight back at the lens. That bright layer smooths over fine texture, which is why small scars and fine lines can disappear on a shiny face.

An international panel on AI in aesthetic medicine recommends starting from makeup-free skin for exactly this reason (Frank et al., J Cosmet Dermatol 2024). The same logic applies to anything that leaves a sheen.

Bare, clean, dry skin gives the most honest reading.

How close do these tools get?

In one study, an AI model graded eight facial features in photos of 1,099 people, checked against five dermatologists (Lee et al., Skin Res Technol 2026). It matched the dermatologists’ exact grade about half the time, and was within one grade 93% of the time.

Photos from people’s own phones held up well against studio camera photos, except for pigmentation, where accuracy dropped. The authors also noted the model was sensitive to image quality.

A 2026 review of facial assessment tools points to the same weak spots: photo set-ups that vary from study to study, and training data that doesn’t cover every skin tone well (Sun et al., Aesthetic Plast Surg 2026). It’s also worth knowing that much of the published checking is done by the companies that make the apps.

Close to a dermatologist’s grade most of the time is useful. It isn’t the same as a dermatologist looking at your skin.

Tiny scars are hard for everyone

Small scars are where flat photos struggle most. Researchers developing 3D scar imaging noted that standard photos are significantly affected by lighting and shadow, and that scar depth is hard to judge even in person (Petit et al., Skin Res Technol 2018).

Dermatologists find it hard too. When dermatologists assessed the same patients’ scars in person, all of them found only about half the scars, and agreement was poor for scars under 2 mm (Kang et al., J Drugs Dermatol 2016).

So if scarring is what you care about, a dermatologist looking at your skin in person, with light coming from the side, is still the better judge. A camera facing you straight on flattens the very shadows that show where a scar dips.

One scan is a snapshot

Lighting, angle and whatever is on your skin all move the result. That’s why two scans a day apart can disagree, and why a single scan shouldn’t be taken as the final word on your skin.

What a scan does well is comparison. Take the same kind of scan in the same conditions every week or two, and the question becomes whether something moved, not what one picture says.

Give it the right amount of time. Clarity and marks tend to shift over 4 to 12 weeks, and pigmentation over months (how long each kind of change takes).

If nothing has changed after about three months of a steady routine, see a GP or dermatologist.

How to get a fair reading at home

Start with bare, clean, dry skin: no serum, moisturiser, sunscreen or makeup. Use the same spot, in daylight, at the same time of day.

Use the rear camera if you can, with no filter and no beauty mode. Keep your distance and angle the same each time.

The low-tech version is a photo once a fortnight in those same conditions. It won’t score anything, but it’s far better than memory.

Photos and mirrors often disagree, and here’s why.

Connecting to Stella

Stella scores appearance on every scan, including redness, pigmentation, texture, breakouts and the marks breakouts leave. It adjusts for lighting and camera differences, and in repeat testing, back-to-back scans varied by 5 to 8 per cent.

That’s the range to keep in mind when you compare. A small wobble between two scans is noise, and a steady trend over weeks is the signal.

Stella does not diagnose. A scan is a record of how your skin looks, not a medical assessment.

For a fair test of a single product, here’s the full method.