How accurate is Skyyn?

Reading the same photo twice moves the Skyyn Score by about 2 points. Two photos of the same face taken minutes apart differ by about 3.5 points on average. Skyyn therefore reports a change only at 5 points or more against the middle of your last three scans, and even then about 1 comparison in 8 can cross that line when the skin has not changed.

These figures come from our own evaluation in October 2026, on scoring version 2. It used a public research set of repeated photographs of 19 people, covering light, medium and deep skin tones, taken in both good and poor lighting. It is a small sample, and the individual tests below used parts of it. We publish it because nobody can judge a score without knowing how much it wobbles.

What a selfie score can and cannot measure

A phone photo records how light reflects off the surface of your skin. From that, software can describe things a person could also see by looking closely: how even the tone is, how visible pores are, how smooth the surface looks, how bright the skin appears.

It cannot measure anything below the surface, anything about how your skin feels, or anything medical. Skyyn does not identify skin conditions and cannot tell you whether a mark is harmless. Apps that claim to detect disease from a phone photo are a different kind of product, held to a different standard, and Skyyn is not one of them.

How consistent is the score?

Before we change the scoring we run a repeatability check: three photos are each read three times, and the change fails if any Skyyn Score spreads by more than 4 points or any sub-score by more than 9. The point of that limit is that a re-read must never look like a real change.

How often does the app show a change that is not real?

We tested this directly. Take a new photo of someone, compare its score with the middle value of three of their other photos, and count how often the difference reaches the reporting threshold. Nothing about the skin changed between the photos, so every time it does is a false change. Fourteen of the 19 people had enough photos for this test.

Skyyn uses 5 points for the Skyyn Score and 10 points for sub-scores. So when your skin has not changed, a new scan still crosses the 5-point line about one time in eight. A given sub-score crosses its 10-point line about one time in twenty-five, but there are four of them, so on about 1 scan in 7 at least one sub-score crosses its line by chance. Treat a single sub-score move as a hint until the next scan agrees.

Comparing against a single previous photo instead of three roughly doubles the error: 25.9% for the Skyyn Score at 5 points. That is why the app compares each scan with the middle of your last three, and why one scan is never a trend. A move that holds for two or three scans in a row is far more trustworthy than any single reading.

How lighting changes the score

Lighting is the largest source of error, and it does not behave the way most people expect. A dim photo tends to score higher, not lower, because the analysis cannot mark down what it cannot see. When we darkened six well-lit photos by three stops, scores moved by 2.3 points on average. At five stops they moved by 4.0 points, upward in five cases out of six. Thirteen real photos taken in very dim rooms read about 5 points high on average.

Skyyn handles this in two ways. It never brightens or darkens the image towards a target; it only lifts a clearly under-exposed photo, by one stop at most, judged from the highlights on the face's own skin. And when a photo is too dim, too blurred or too unevenly lit to trust, the scan is marked low confidence and is not used to work out whether your score changed. In our evaluation 3 of 68 scored photos were flagged this way.

How to take a photo that gives a steady reading.

Does it work the same across skin tones?

We test this at three points and publish what we found, including the problem we had to fix.

• Getting past the camera check. Measured on 376 photos. For those that were front-facing and centred, 100% passed for light skin tones, 100% for medium and 98% for deep. Before October 2026 the figure for deep skin tones was 87%.
• What caused the gap. The face detector we use returned no face on 11 photos, all of people with deep skin tones and none of them badly lit. Skyyn now asks the detector a second time on a contrast-stretched copy of the frame, which recovered 9 of the 11. We also removed a brightness check supplied by the detector, because it tracked skin tone more than it tracked photo quality.
• Low-confidence flags. 3 of 68 scored photos were flagged, about one in each tone group. That is too few to compare groups; it shows only that no group is being flagged heavily.
• Photo-quality measurements. Sharpness is measured around the eyes with local contrast normalised, and clipping is measured on patches of skin, not the whole face box. Average brightness is never used as a quality measure, because it would penalise darker skin.

There is no logic in Skyyn that treats one skin tone differently from another. When a measurement turns out to depend on tone, we fix the measurement. With 19 people, each tone group rests on a handful of individuals, so we treat any gap as a reason to look closer, not as proof either way.

What happens when a photo cannot be scored

Skyyn would sooner refuse a photo than guess. A photo is turned away, with a note on what to change, if it is not a face, not a real photograph, shows more than one face, shows a face that is too small or covered, or is too poor to read. Our checks include a drawing, an object, a group shot, a distant figure and the back of a head, and each one must be refused.

Scoring versions

Scores from different versions are never compared, and old scores are never recalculated. What "Scoring updated" means on your chart.

What we have not measured

• We have not compared Skyyn Scores with a dermatologist's assessment, and we make no claim about agreement with one.
• We have not run a study showing that a rising score follows from any particular product or routine.
• Our evaluation used 19 people. It tells us about consistency; it is too small to say much about rare cases.
• The figures above apply to scoring version 2. We will update this page when they are measured again.

When to see a dermatologist instead. Skyyn is a personal skincare assistant, not a dermatologist, and its readings are not a medical diagnosis. See a professional for anything that is painful, bleeding, changing quickly or not going away. The American Academy of Dermatology's finder lists board-certified dermatologists in the United States.