The numbers behind the scores.
Stella reads 21 signs of how your skin looks from one guided scan. We tested it two ways: against itself, and against dermatologists.
Cosmetic appearance only. Not a medical device, and not a diagnosis.
0.94
Stable
When nothing has changed, the score doesn't either. Average reliability of a person's trend across all 21 scores.
1.4 points
Precise
A 30-person study can detect an average change of about 1.4 points on the 0 to 100 score, and 2 points or less on 18 of the 21.
92%
Read the way dermatologists read it
On six signs, Stella agrees with dermatologists 92% as closely as they agree with each other.
Measured against how dermatologists agree.
Two dermatologists looking at the same skin rarely give identical grades. So the fair test isn't whether Stella matches one of them exactly, but whether it agrees with them about as closely as they agree with each other.
On six signs, Stella agrees with the dermatologists 92% as closely as they agree with each other. Against the two with no financial interest in Stella, 96%.
The six are redness, pigmentation, fine lines, breakouts, dark circles and post-breakout marks. The other 15 scores were tested for how steady they are, not against dermatologists.
How we tested it.
Against itself
People scanned at home, on their own phones, on different days. The app checks every scan before it counts: the face framed and facing the camera, and the light and focus good enough.
Against dermatologists
- Unseen photos. None had been used to build the AI.
- Answers locked first. Stella's scores were sealed before grading began.
- Graded blind. Each dermatologist worked alone.
One of the three is Dr Alice Rudd, Stella's co-founder. The other two work at her dermatology practice and hold no financial interest in Stella. None was paid for grading.
21 scores of how your skin looks, each from 0 to 100.
| Lines and folds | Fine lines, deep lines, smile lines, marionette lines |
|---|---|
| Volume and contour | Sagging, jowls, eye bags, puffiness, tear troughs |
| Tone and colour | Redness, pigmentation, sun spots, freckles, melasma, dark circles |
| Blemishes | Breakouts, post-breakout marks |
| Surface | Texture, pores, oiliness, hydration |
How steady each score is.
Tested at four levels, each adding something that could move the score. 1.0 is perfect.
- The same photo, read againIdentical
- Two frames, milliseconds apart0.97 to 1.00
- One scan to the next, a day or two apart0.85
- A person's trend over repeat scans0.94
See all 21 scores
Trend reliability by score. Excellent is 0.90 or higher; Good is 0.75 to 0.89.
| Score | Trend reliability | Grade |
|---|---|---|
| Fine lines | 0.98 | Excellent |
| Deep lines | 0.98 | Excellent |
| Post-breakout marks | 0.98 | Excellent |
| Breakouts | 0.98 | Excellent |
| Sagging | 0.97 | Excellent |
| Marionette lines | 0.97 | Excellent |
| Smile lines | 0.97 | Excellent |
| Jowls | 0.96 | Excellent |
| Tear troughs | 0.95 | Excellent |
| Eye bags | 0.95 | Excellent |
| Puffiness | 0.95 | Excellent |
| Redness | 0.94 | Excellent |
| Pigmentation | 0.91 | Excellent |
| Sun spots | 0.90 | Excellent |
| Freckles | 0.90 | Excellent |
| Hydration | 0.90 | Excellent |
| Melasma | 0.88 | Good |
| Pores | 0.82 | Good |
| Dark circles | 0.81 | Good |
| Texture | 0.80 | Good |
| Oiliness | 0.64 | Changes hourly |
How to read these numbers.
- The dermatologist comparison covers six of the 21 signs. All 21 are tested for repeatability.
- Stella ran this study. The method paper has been submitted to Cureus for peer review.
- Stella reads how skin looks. It doesn't diagnose.
- Claims about a product come from group results, never one person's scans.
Your photos.
The dermatologists saw cropped regions only, with no names or account details.
Read the full method.
Stella Skin AI Benchmark
White paper, available on request.
Book a call for the white paper
Questions
How reliable are Stella's skin scores?
When nothing has changed, the score doesn't either. A person's trend repeats with an average reliability of 0.94 across all 21 scores, where 1.0 is perfect, measured on repeat scans taken at home on people's own phones. The same photo always returns the same score.
How does Stella compare with dermatologists?
Three dermatologists graded the same photos without seeing Stella's answers. On six signs, Stella agrees with the dermatologists 92% as closely as they agree with each other. Against the two with no financial interest in Stella, 96%.
What is the smallest change Stella can detect?
In a 30-person study, an average change of about 1.4 points on the 0 to 100 score, and 2 points or less on 18 of the 21 scores.
Which signs of skin does Stella score?
21, each from 0 to 100, in five groups: lines and folds, volume and contour, tone and colour, blemishes, and surface, from fine lines and sagging to redness, pigmentation, breakouts, pores and hydration.
Has the research been peer reviewed?
Not yet. Stella ran this study, and the method paper has been submitted to Cureus for peer review.
Does Stella diagnose skin conditions?
No. Stella reads how skin looks. Its scores are cosmetic, it is not a medical device, and it doesn't diagnose.
Appearance-level observations, not therapeutic advice. Not a medical device.