Findation claims 107,000 shades across brands and Perfect Corp 89,969 skin tones, but every accuracy figure we found, from Il Makiage's 90% to Clarins's 96%, is the maker's own. Here is which tool fits where you start, and what the camera keeps.
SlipperSkin finding
Searching on 21 and 23 October 2026, we found no published independent test of any makeup shade-match or try-on tool's accuracy by skin tone; every accuracy figure we found is the maker's own, from Il Makiage's 90 percent (2019) to Clarins's 96 percent against makeup artists (2026).
The one that starts from what you already know. A shade match is only as good as the shades behind it and the light you are standing in, so the right tool depends less on its AI than on its catalogue and your starting point.
In short:
Nobody outside the makers has published a measurement. On 21 and 23 October 2026 we searched for a peer-reviewed study, a Which? or Consumer Reports test, or a journalist's systematic comparison of any commercial shade finder or try-on tool across skin tones, and found none. Several web pages describe their own "tests" of shade apps without naming who ran them or publishing the data, which is not something we can cite. What exists are the makers' figures:
To be fair to the makers, testing does happen; it is just not published. MAC's case study on Perfect Corp's site quotes the brand saying it does "scientific accuracy testing" of the shade matching at a corporate testing facility. Perfect Corp's own shade finder page offers a "200%" increase in sales, which is a business result, not an accuracy one. Il Makiage's 90 percent came, in Glossy's March 2019 report, from beta testers, and the brand did not explain where the figure came from.
The research next door is a reason to swatch anyway. A 2026 paper in Skin Research and Technology by Alipour, Burke and Courtney found that ITA, a colour angle widely used to infer skin tone from photos, classified Fitzpatrick skin types correctly only 22.9 percent of the time under uncontrolled imaging, below the 51.3 percent a guess of the commonest type would score, and failed on the darkest type, with lighting a named cause. That is a research metric, not any app's engine. And Riccio and Oliver (2022) found that social media beauty filters made race classifiers more likely to label faces of every race as white: filters, not shade finders, but the same kind of camera pipeline. For how far skin-analysis claims stretch, see how accurate AI skin analysis apps are.
Up to 107,000 on the makers' own counts, but each number counts something different. Findation's 107,000 are foundation shades across about 1,530 brands (its counter read 1,533 and then 1,526 on two loads of the same page on 23 October 2026). Perfect Corp's 89,969 are skin tones its matcher can tell apart, not products you can buy; a brand that licenses it still matches you to its own shelf. Lancome's often quoted 72,000 for Le Teint Particulier is 8,000 base shades times 3 coverage levels times 3 moisture levels, mixed at the counter, as Houstonia reported in 2018 and Cosmetics Business in 2017.
Yes, because a finder can only recommend from its own list. Findation is the multi-brand case: you enter shades you already wear well, and it finds equivalents from what other users who wear the same shades report, which it says it supplements with spectrophotometry in its own lab. It needs no photo at all. Brand finders work the other way round: Fenty's asks you to pick an undertone (cool, warm, neutral or not sure) and then reads a selfie, but its answer can only be a Fenty shade. Il Makiage's PowerMatch is a quiz of 18 questions in seven sections. None of that is a flaw; it is the brief. A finder built for a brand helps you choose within that brand, and Findation helps you move between them.
It depends on the tool, and the answer is on each maker's own page. Perfect Corp's YouCam privacy notice says that for most services neither your face photo nor your face feature vectors are stored in its database; a try-on that needs its servers deletes the photo after use; photos you ask it to email are kept for no longer than 3 years; and consenting to its collection of biometric information is a condition of using the app. It also says it does not and will not use facial recognition. The notice shows no effective date. Fenty's finder asks for consent to "obtain a scan of your face", including face geometry, before it starts.
| Tool | Uses a camera | What the page says | How long it is kept |
|---|---|---|---|
| YouCam Makeup | Yes, live or photo | Most services store neither face photos nor face feature vectors; cloud-processed try-on photos are deleted after use; consent to biometric collection is a condition of use; no facial recognition | Photos you email or save: up to 3 years |
| Fenty Beauty shade finder | Yes, selfie on your phone | May obtain a scan of your face, including face geometry, which it calls biometric data | Not stated on the tool page |
| Findation | No camera | You type in shades you already wear | No face image is involved |
| Clarins AI Shade Finder | Yes, at the counter | A series of images while the screen changes colour (in store only) | No policy text read |
Illinois's Biometric Information Privacy Act (BIPA), which defines a scan of face geometry as a biometric identifier, has produced a run of class actions over try-on tools, with mixed outcomes. Of the 6 rulings and settlements below, three let claims proceed, two ended in dismissal, and one settled. These are allegations and procedural rulings: a denied motion to dismiss means a case can go on, not that the company broke the law, and a settlement is not a finding.
From $5.99 a month, though the App Store lists 10 purchase options at once. Read on 23 October 2026, Premium runs $5.99 to $9.99 a month or $31.99 to $39.99 a year, and Pro is $19.99 a month or $79.99 a year (App Store listing). At the cheapest prices, a Premium year costs the same as 5.3 months and a Pro year the same as 4 months, so the annual plan pays off if you keep it longer than that. Which price you are shown depends on region and offer.
1. Light: face a window in daylight, no flash, no warm bulb overhead. Lighting moves the reading. 2. What is behind it: one brand's shelf, or many brands? A finder can only pick from its own list. 3. Undertone: does it ask, or guess? Fenty's finder asks you to choose before the scan. 4. The camera: read what is kept. YouCam keeps photos you email or save for up to 3 years. 5. Swatch before you buy: jawline, in daylight. No finder publishes an independent accuracy test.
Choosing a skincare scan instead? See the best AI skin analysis app. And for how far apart beauty apps can land on the same product, our 20-product ingredient study.
We read each maker's own page for its shade count, accuracy claim, price and privacy terms on 23 October 2026, and used trade reports only where the maker's page could not be read; those are dated. The search for independent accuracy evidence ran 10 queries on 21 October 2026 (study, arXiv, Which?, Consumer Reports, PubMed-style and journalist searches, plus each maker's own claim) and 2 more on 23 October 2026. PubMed and Google Scholar were not queried directly. Sephora's Color iQ page refused automated reading, so we describe it only in outline. Sephora Virtual Artist is not in the tree: a Sephora representative told a customer on Sephora's community forum in June 2022 that it was no longer available, and that forum has since closed. Court events come from legal commentary and trade reports, not dockets. Nothing here is a verdict on any product's quality; swatch before you buy.
As an Amazon Associate I earn from qualifying purchases.
It depends where you start. If you already wear a shade that works, Findation translates it across brands for free without a camera; it claims 107,000 shades. If you shop one brand, use that brand's own finder. In store, a device scan in good light is the next step.
Nobody has published an independent test. Searching on 21 and 23 October 2026 we found no peer-reviewed or consumer-group test by skin tone; every accuracy figure is the maker's own. Related research found a common photo-based skin tone measure failed on the darkest Fitzpatrick type under uncontrolled light.
Perfect Corp says its matcher draws on a database of 89,969 skin tones. Those are tones it can distinguish, not products; a brand that licenses the tool matches you to its own range.
Its privacy notice, read 23 October 2026, says most services store neither face photos nor face feature vectors, cloud try-on photos are deleted after use, and photos you ask it to email are kept for up to 3 years. Consenting to biometric collection is a condition of use.
On the US App Store on 23 October 2026, Premium was $5.99 to $9.99 a month or $31.99 to $39.99 a year, and Pro $19.99 a month or $79.99 a year. Prices vary by region and offer.
Yes, under Illinois's biometric privacy law. Charlotte Tilbury agreed a $2.925 million settlement (reported December 2024); claims against Estee Lauder were dismissed without prejudice in January 2024; cases against Shiseido, L'Oreal USA and MAC were allowed to proceed. None of these rulings is a finding of liability.
Pick a shade tool by what it can recommend and where you start, not by its accuracy claim, because no claim has been independently tested. Use daylight, check what the camera keeps, and swatch the shortlist on your jawline before you buy.