FR-300
Allergens are a hard constraint
Not a warning badge, not a ranking penalty. A flagged ingredient removes the product from your results entirely — the system targets 100% recall on allergen detection.
Inclusive beauty intelligence
Colour science
that finally counts
every complexion.
CeySkin reads your skin the way a laboratory would, then finds makeup that actually matches it. All ten levels of the Monk Skin Tone Scale, held to the same standard of accuracy — the deepest exactly as closely as the lightest.
Coming to iOS and Android in 2026. Your face is mapped on your own device, and any photo sent for checking is deleted within 24 hours.
478 landmarks · on-device
Classification
Monk 07
CIELAB
L* 42.5 · a* 12.3 · b* 20.5
- 478
- MediaPipe face landmarks
- ΔE < 2.0
- Imperceptible colour difference
- 10 / 10
- Monk levels at parity
- < 5s
- End-to-end analysis
- 0
- Biometric records kept
The problem
Beauty tech was
trained on a
narrow face.
Shade matching tools routinely fall below 60% accuracy for deep complexions — the levels the industry calls MST 7 to 10. The models were not built to fail; they were built on datasets where those tones barely appeared.
You already know how that ends. Three shades bought in the hope that one of them lands. A parcel posted back. The quiet suggestion that your skin was the complicated part.
It was never your skin. It was the data the model learned from.
< 60%
Typical accuracy at MST 7–10
95%+
CeySkin target, every level
30%
Return reduction goal
The Monk Skin Tone Scale
Ten levels. One standard of accuracy.
Developed at Google as an open replacement for the six-point Fitzpatrick scale, the Monk scale gives deep complexions the resolution they were previously denied. Every level below carries a measured CIELAB reference.
Level
Monk 07
Lightness L*
42.5
Chroma a* / b*
12.3 / 20.5
Character
Warm
Reference values converted from the published Monk scale swatches under the D65 standard illuminant, per ISO 11664-4:2008.
The method
Four steps,
under five
seconds.
Nothing here is a black box. Each stage is a published, verifiable piece of colour science — which is precisely why the result holds at every level of the scale.
Capture
FR-101A guided overlay checks framing and luminance before the shutter fires. Frames outside a 30–220 luminance window are rejected, so a bad light never becomes a bad match.
Isolate
FR-102MediaPipe maps 478 landmarks on-device and samples only the Triangle of Truth — both cheeks and the forehead. Shadow, specular highlight, hair and blemish pixels are discarded before any colour is read.
Convert
FR-103Median RGB per cluster is converted to CIELAB under the D65 illuminant — a perceptually uniform space where equal numeric distance means equal visible difference. This is what makes deep tones measurable rather than approximated.
Match
FR-200Delta-E ranks every shade in the catalogue against your reading, allergens are removed as a hard constraint, and the model explains in plain language why each product surfaced.
In the hand
A match, and the
reason for it.
You get shades that will look right on you, and nothing that will react with your skin. Each one tells you plainly why it was chosen, with the full ingredient list sitting behind it.
- Close enough that your eye cannot tell the shade from your own skin — measured, not guessed at.
- Anything on your allergen list is gone before you see it. Not flagged, not ranked lower. Gone.
- Your matches stay on your phone, so the list still works with no signal and no shop nearby.
9:41100%
Your analysis
Monk 07
warmFoundation · 2 matches
Sienna Warm 340
Atelier Nº7
98
ΔE 0.8
Clear of your allergen profile
Deep Amber 12W
Maison Lune
94
ΔE 1.6
Clear of your allergen profile
Golden Umber 08
Excluded · contains linalool
What it does
Built around the constraints that matter.
Health first, then ethics, then colour. A recommendation that ignores the first two is not a recommendation worth making.
FR-301
Full INCI transparency
Every ingredient, named the way regulators name it. No proprietary blends hiding behind a fragrance line.
FR-302
Verified ethical filters
Vegan and cruelty-free claims held to sourced evidence, so the filter means something.
Privacy by design
Your face is mapped on your device
Landmark extraction runs on the handset, so the map of your face is never uploaded. No biometric template is created, no identity is inferred, and any photo sent for cross-checking is deleted within 24 hours.
FR-200
Cross-brand by default
A curated catalogue matched on colour science rather than on who paid for placement.
FR-400
Works where stores do not
Offline access to saved matches, plus a locator for the rare trip into town.
Who it is for
Designed from the edges inward.
Most products are built for the middle of the distribution and patched outward. This one starts with the people the middle leaves behind — and the middle is served automatically.
MST 7–10
Deep complexions
Existing tools drop below 60% accuracy at the deep end of the scale. Parity is the entire point of this build, not a later release.
Zero false negatives
Sensitive & reactive skin
When a reaction is the cost of being wrong, filtering has to be absolute rather than advisory.
Verified claims
Ethically-led buyers
Vegan and cruelty-free labels backed by evidence, so greenwashing has nowhere to sit.
Remote matching
No counter within reach
Two hours to the nearest shade range, and a return posted back if it misses. Accuracy at a distance changes the maths.
Early access
Be matched properly,
for once.
We are opening early access across all ten Monk levels, starting with the complexions shade finders have failed for longest. Leave your email and we will tell you the moment yours is ready.
One email at launch. No sharing, no selling, and deletion on request under GDPR and CCPA.
Monk 01 — 10 · equal accuracy