The engine behind the report.
SKNscan Engine is our own scanning pipeline, built from the ground up for one job: reading how skin looks in a front-on phone photo and saying so plainly. This page is how it does that, step by step.
478
face mesh points
16
named analysis zones
12
live capture checks
6
signals, banded with confidence
Capture that earns the read.
A good measurement starts with a good photo. The phone does the work before the shutter, so the engine only ever sees a face it can read.
Live face tracking
The camera tracks the face in real time and checks twelve things at once: one face only, large enough in the frame, centred, facing the camera within ten degrees, phone at eye level, head straight, enough light but not too much, and eyes open.
Steady for over half a second
Every check must hold for 600 milliseconds before the shutter unlocks and the ring turns green. The instruction on screen tells the person exactly what to change, in plain words.
A six-frame burst
One tap takes six frames. The engine scores each for sharpness and exposure and keeps only the best, at full camera resolution.
Quality gates on the server
The photo is checked again for blur, brightness, clipped highlights and shadows, glare, uneven lighting between the cheeks, closed eyes and anything covering the skin. A photo that fails is sent back with the reason, never scored badly.
A map of the face before a single measurement.
The engine places a 478-point mesh on the face and draws sixteen named zones from it: forehead, between the brows, nose, chin, each cheek, under each eye, the outer eyes, beside the mouth, both nose-to-mouth folds and both mouth-to-chin lines.
Only skin is measured
Eyes, brows, lips and nostrils are cut out with a margin. Glare and deep shadow are masked. Hair, a fringe or a beard across a zone is detected by colour and that zone is left out rather than misread.
Lighting corrected first
Phone photos are never evenly lit. Before anything is measured, the slow fall-off of light across the face is modelled and removed, so a shadowed cheek is not read as a darker one.
Colour science, not pixels
Everything is analysed in CIELAB, the colour space used in dermatology and cosmetic research, where a difference in numbers matches a difference the eye can see.
Six signals, each measured its own way.
No two signals use the same method, because no two look the same on a face.
Nose-to-mouth and mouth-to-chin folds
Each fold is traced along its own expected path and sampled at 28 cross-sections, looking for the valley in lightness a fold casts. Depth and how far it runs decide the band. The strongest named fold is the one reported.
Visible lines
A ridge filter, the same family of filter used to trace blood vessels in medical imaging, finds line-shaped structure on the forehead, between the brows, at the outer eyes and beside the mouth. The result is line length per area of skin.
Under-eye colour
The under-eye band is compared with the cheek directly beside it on the same side, as a ΔE2000 colour difference and as shadow depth. Reported per side, so a darker left under-eye is said to be the left.
Visible redness
Redness is the a* channel of CIELAB. The engine reads how far each patch sits above the skin around it, so a naturally warm complexion is not mistaken for flushing.
Tone evenness
How much lightness and colour vary across each zone, measured with a robust spread that one freckle cannot swing.
Colour contrast
A multi-scale difference filter picks out small areas that sit clearly darker or lighter than the skin around them, and reports how much of the zone they cover.
Bands and confidence, never a score.
Each signal becomes low, mild, moderate or strong. Alongside it sits a confidence the engine works out for itself.
Confidence is earned
A reading starts at full confidence and loses a step for each thing that makes it harder: soft focus, glare, uneven light between the cheeks, an over- or under-exposed face, a zone the engine could only partly see, or makeup the person told us about.
Makeup is respected
With makeup or fake tan ticked, the colour signals show "not read with makeup on" rather than a guess. Folds and lines, which makeup does not hide, are still measured.
Why there is no number out of 100
A single score from one phone photo would be false precision. A band with a confidence is what the measurement can honestly support, and it is what a practitioner can confirm in the room.
The AI writes. The engine owns every number.
Two separate systems, by design. The measuring is deterministic code. The explaining is a language model with a short leash.
What the writer is given
The bands, the zones, the confidence, the person's stated goal and concerns, and the treatments your rules selected. Never the photo. Never a raw number it could round up.
What it is forbidden to say
No diagnosis or condition. No score, percentage or ranking. No skin age, hydration or collagen claims. No promise that a treatment will fix anything. No treatment your rules did not pick. Nothing that shames or hints at age.
Your clinic has the last word
Every report can be read, edited and approved by a practitioner before it is sent, and a scan can be sent back for a retake. The report says a practitioner confirms everything in person, because one does.
Rules decide the treatments. Yours.
A treatment appears on a report only when a rule you can read says so: this signal, at least this band, from this age, never during pregnancy or breastfeeding. Under 25s are never offered an injectable by name.
Your menu, your names
Start from a UK-compliant list and edit freely. The report can only name what you have switched on.
Strongest area first
The plan is ordered by what stood out most, with up to three treatments per area and your booking link behind every button.
A stranger's face is handled like one.
Consent per purpose
The scan, the report email, the AI preview and marketing are separate switches the person sees before the camera opens, and each answer is recorded.
Photos on our own servers
Camera metadata is stripped on arrival. Photos are never served from a public address; the application checks who is asking every time one is shown.
Contact details encrypted
Names, emails and phone numbers are encrypted in the database, with a separate key for search.
Deleted on request, with a record
A deletion request removes the photos, the scan and its jobs, and writes an audit entry. The link to ask is on every report.
What it reads, and what it leaves to you.
- It reads a photograph. Six signals from one front-on photo, each with a band and a confidence. A cosmetic appearance snapshot, not a medical assessment and never a diagnosis.
- Light matters. Hard light from one side exaggerates folds; a soft photo hides them. The capture guide pushes people towards daylight and a bare face, and the report says when a photo made a signal harder to read.
- Jowls and sagging are not measured from a single front-on photo. A practitioner assesses laxity in the room.
- Built and tuned in a working clinic. SKNscan Engine was developed at Sorrell & Co and tuned on real patient photos there before it was offered to anyone else. In its first month it ran 77 scans that led to 27 consultations.