How VISTASY works

Written so that anyone can check what we do and why. Last updated October 2026.

The photo

Before any analysis we check the photo: exactly one face, large enough in the frame, looking at the camera, no glasses, even light, sharp. If it fails we ask for another photo and say why. This matters more than the model: most score differences between two selfies come from lighting and angle.

The scores

We align the face and cut fixed regions: forehead, under-eye, T-zone, cheeks. Each concern is scored from the region where it is visible, on a 0 to 100 scale where 0 means nothing visible and 100 means extreme. A vision model reads the regions three separate times; the median is your score and the spread between readings is shown as confidence. When something cannot be judged from the photo, we say so instead of guessing.

Scores are cosmetic observations, not medical findings. We do not name diseases. If something looks severe or atypical we suggest seeing a dermatologist.

The ingredients

For every concern we keep a knowledge base of active ingredients with the evidence behind them. Every entry cites PubMed studies that we verified exist, and each ingredient-concern pair carries a grade: A randomised trials or meta-analyses in humans, B smaller human studies, C laboratory or anecdotal evidence. Entries are drafted from the literature and reviewed by a cosmetic chemist and a dermatologist before they are marked validated. EU regulatory limits and bans are hard facts in the system, not opinions.

Reading a product

We parse the ingredient list printed on the product. In the EU and the US ingredients are listed in descending order of concentration down to 1 %; below that, order is free. We estimate where that 1 % line sits from ingredients that are capped near it, so we can tell whether an active is present at a meaningful level. Some actives, like retinol or peptides, work below 1 %, and we account for that. When a brand states a percentage, we use it.

We also estimate how much of each active a product holds. Three facts bound it: an ingredient listed above the 1 % line is at least 1 %; it cannot be more than the ingredient listed before it; and everything adds up to 100 %, so the fifth ingredient, for example, can never exceed a fifth of what is left. We then tighten the range with what products on the market typically contain and with legal limits. The result is a range, such as "about 1 to 5 %", compared with the range studies support for that ingredient: enough, may be enough, or likely too little. It is an estimate from the label, never a lab measurement, and it assumes the label follows the listing rules.

The routine

Your top concerns, weighted by score and confidence, point to target ingredients, weighted by evidence grade. Pregnancy, sensitivity, allergies and EU bans remove ingredients before anything is ranked. Products are scored on how well their ingredient list matches the targets, with penalties for fragrance, declared allergens, drying alcohol and ingredient conflicts within the routine. The store's catalogue is only applied at the end as an availability filter. If the best ingredient for you is not sold where you are, we say so and tell you what to look for elsewhere.

Ranking is never influenced by brand, margin or sponsorship. Who pays us, and what nobody can buy.

Sarah

Sarah answers questions using only your analysis, your routine and the ingredient notes. She cannot invent products or studies, and she will tell you when she does not have data.

Your data

• Your photo is stored in our Europe region. To read it, we send it to AI models and to our own image models in London, UK. Our London copies are deleted as soon as they have been read. Moving every step into the EU is planned.
• You choose how long your photo is kept. By default it is kept 30 days, so your next scan can be compared with it. With "delete after scoring" it is deleted right after your results, and within 2 days at the latest. Scores and your routine are kept so you can track progress.
• We record your consent and your confirmation that you are 18 or older with each scan.
• Accounts sign in with an email link or, during the private preview, a password. You can delete a single scan, or your whole account and its photos, from the app at any time.
• Anonymised, aggregated trends are only produced from people who opted in, and only for groups of at least 50 people.
• We never sell individual data.

What we are still working on

A validation study comparing our scores with dermatologist grading across skin tones, with results published here. Moving the photo analysis entirely into the EU. A review of every ingredient entry by a cosmetic chemist and a dermatologist.