About
We are building
the thing that
should already
exist.
Matching a foundation to a face is a solved problem in colour science. It has been solved since the 1970s. It is only unsolved in practice, and only for some people.
CeySkin exists to close that gap — not by inventing new science, but by applying the existing science evenly, and refusing the shortcuts that made it uneven in the first place.
Why the gap exists
Nobody decided
to build it
badly.
A model learns the distribution it is shown. Show it a dataset where deep complexions are rare, and it will be confident and wrong about them — without anyone ever writing a line of code that intends harm.
The same is true of the scale the industry inherited. Fitzpatrick was designed in 1975 to predict how skin responds to ultraviolet light — how easily it burns. It was never a description of colour, and it compresses the entire deep end of humanity into its last two types. Build a shade finder on top of that and the resolution simply is not there to succeed.
None of this requires malice, and pointing at malice would let the actual causes off the hook. What it requires is a different scale, a perceptually uniform colour space, and a refusal to average away the cases that are hardest to get right.
Which means the fix is not clever. It is just deliberate — and measurable enough that we can be held to it.
What we will not do
A product is what
it refuses.
Values are easy to publish and hard to keep. These are written as refusals because a refusal can be checked — either the thing happens or it does not.
We will not keep your face
No facial recognition, no biometric template, no account in your name. The app is keyed to a device identifier, and any photo sent for cross-checking is deleted automatically inside 24 hours.
We will not sell your place in the results
Shades are ranked by measured colour distance from your skin. There is no paid placement, no house brand pushed to the top, and no reason for the ordering other than the arithmetic.
We will not soften a safety filter
A declared allergen removes a product outright. Not a warning label, not a lower rank. Where recall and precision conflict, recall wins — a missed allergen is a reaction, not a metric.
We will not pass on a claim we cannot verify
Vegan and cruelty-free filters show only what is backed by evidence. A brand asserting its own virtue is not evidence, and forwarding it unchecked would make the filter worthless.
We will not hide the method
The colour space, the sampling geometry, the conversion and the thresholds are all published on this site. Anyone can check the reasoning, including the people it has historically failed.
We will not let a model overrule a measurement
Where the vision model and the CIELAB arithmetic disagree beyond two levels, the arithmetic wins. Models inherit the imbalance of what they were trained on. Measurement does not.
How we work
Built on things
anyone can
read.
Nothing at the core of this product is proprietary. Every standard below is published, most are free to use, and all of them can be checked by someone who does not work here.
That is deliberate. A claim about fairness that rests on a secret method is not a claim anyone can hold you to.
Monk Skin Tone Scale
Google, open licenceTen levels rather than Fitzpatrick’s six, with the additional resolution placed where it was missing — the deep end.
ISO 11664-4:2008
CIELAB colour spaceA perceptually uniform space, defined publicly decades ago. Equal numeric distance, equal visible difference.
MediaPipe Face Landmarker
Apache 2.0Runs on the handset, costs nothing to license, and can be inspected by anyone who wants to know how the landmarks are found.
INCI nomenclature
Regulatory standardIngredients named the way regulators name them, so an allergen cannot hide behind a friendlier synonym.
WCAG 2.1 Level AA
W3CThe accessibility bar, treated as a requirement to be met rather than a rating to aspire to.
Who can use it
Accessible is not a later release.
An app about being left out cannot itself leave people out. These are requirements in the specification, not items on a backlog.
- AA
- WCAG 2.1, every screen
- Full VoiceOver and TalkBack support, not a partial pass on the main flow.
- 4.5:1
- Minimum text contrast
- And 3:1 for large text, checked rather than assumed.
- 44pt
- Smallest touch target
- 48dp on Android. Nothing important hidden behind a target you have to aim at.
- ∞
- Time to complete a step
- No timed interactions anywhere. Retake the photograph as often as you need.
Who is building it
A small team,
working in the
open.
CeySkin is built by a small engineering team at Helixz Solutions.
We are not a beauty brand, and we do not sell cosmetics. We have no shade range to protect and no inventory to move, which is precisely why the ranking can be honest — the results are ordered by measured colour distance because we have no commercial reason to order them any other way.
What we do have is a specification with numbers in it: 95% accuracy at every level of the Monk scale, no more than five points of variation between the lightest and the deepest, and total recall on declared allergens. Those figures decide whether this ships. They are published on the science page so that they can be quoted back at us.
Working with us
We are interested in brands willing to publish shade data properly, and in researchers working on tone-inclusive measurement.
Testing with us
Early access opens across all ten levels, weighted towards the complexions current tools serve worst.
Hold us to it.
Early access opens across all ten Monk levels. If we miss the standard we have set, we would rather hear it from you than not hear it at all.