Methodology
How MIRU reads, and how well.
Any tool can print a number.
Fewer will tell you what the number is built on, or where it falls apart. This page is both, and the second half is the useful one.
The read, step by step
The photo is graded before your skin is.
Light, sharpness and contrast, checked first on your device. Warm bathroom light fakes redness and a soft focus hides texture. A weak capture gets flagged instead of quietly becoming a confident answer.
Then it is read once.
The active read examines only the image you selected. The privacy boundary is architectural, not aspirational.
Every signal says how sure it is.
Observed, Inferred, or Unknown, on every reading, every time. If the light hid your pores, pores comes back Unknown. A guess dressed as a measurement is the thing that breaks trust first.
Every signal collapses into one focus.
Ranking every concern at once multiplies the error in all of them. Naming the strongest keeps the reading honest and the next move obvious.
Validation, plainly
No study yet, so no number yet.
Fairness
Every skin tone, read on its own terms.
Redness on deep skin does not look like redness on fair skin.
It shows up as warmth and darkening, not pink. A model that averages the two calls it wrong in both directions.
So your reading is graded against your own tone group, never a blended average, and the validation above reports each Fitzpatrick type separately for the same reason.
When light or framing makes the tone call less certain, the reading says so before it says anything else.
What a photo can tell you
Tone patterns on the surface.
Darker and lighter regions, and how sharply they edge. That is real signal for visible pigmentation, read against your own baseline.
Redness, relative to your own skin.
A photo carries how red one area is beside the area next to it. MIRU reads that deviation, never an absolute colour, so it holds across skin tones.
Texture, as light and shadow.
Roughness and congestion change how light sits on the surface. The pattern is readable; its exact depth is not.
Shine, dryness cues, visible flaking.
Surface state photographs honestly in good light. This is why the quality gate cares so much about the light.
What it cannot
Collagen, elasticity, firmness.
These live beneath the surface, and a camera records reflected light from the surface. MIRU does not read firmness at all. It is not one of the signals, and no vertex is drawn for it. If firmness is the question, consult a dermatologist. A photo cannot answer it.
Anything subsurface: melanin depth, UV damage below the skin.
Clinic imaging uses UV and polarised light to see under the surface. A phone camera does not, so MIRU labels its pigmentation read as visible tone, full stop.
Hydration as a measurement.
Real hydration is measured by conductance on the skin. A photo shows its visible signature, so MIRU scores the look and asks the questionnaire about the feel.
True depth, in millimetres or at all.
A line and its shadow arrive as the same pixels. MIRU renders relief as visualisation and labels it as exactly that.
Stinging, tightness, how products feel.
Sensation does not photograph. These reach the read through your own words, and are weighed as reported, never as seen.
Every boundary above appears in the read itself: as a declined score, a labelled visualisation, or a question instead of a claim. A number printed past one of these boundaries is a guess wearing a scale.
Where MIRU is weakest
Warm indoor light exaggerates redness.
Tungsten and dim rooms both push skin toward pink. That is why the quality gate exists, and why it will refuse a photo you liked.
One photo is a moment, not a trend.
Two captures, two weeks apart, in the same light. That is the smallest honest unit of change. Anything faster is noise wearing a chart.
MIRU reads appearance, not pathology.
Moles, lesions and anything that is changing belong with a dermatologist. No app should stand between you and that appointment.
Take a reading and watch it do exactly this.
Personal guidance, not medical advice.