The science behind every score.
AiSkin estimates visible skin concerns from a single photo using computer vision and a multimodal model. This page explains how scores are derived, the dermatology grading scales that inform them, and the limitations you should know.
From pixel to score.
A single, well-lit frontal photo passes through facial landmark alignment, region segmentation, per-region feature extraction and a calibrated scoring model that maps to eight dermatological channels.
A single frame or upload is normalized for exposure and white balance.
Facial landmarks register the face so regions are comparable across angles and devices.
Per-region features feed eight calibrated concern channels, each scored 0–100.
Scores roll into an overall skin score, a skin type, and tailored guidance.
What each of the 8 concerns measures — and where it falls short.
Every channel is a visible proxy, not a clinical measurement. Understanding the limits keeps expectations realistic.
Visible comedones, inflammatory papules and pustules, estimated severity against IGA-style banding.
Cannot distinguish active inflammation from residual marks (PIH) in a single image; subtype resolution is approximate.
Diffuse erythema and localized flushing across cheek and perinasal regions via chromatic analysis.
Sensitive to lighting warmth and recent exertion; not a rosacea diagnosis.
Surface flaking, fine dehydration lines and dullness used as proxies for barrier water loss.
A photo cannot measure transepidermal water loss directly; scores are visible proxies.
Surface sheen and pore prominence in the T-zone relative to cheeks.
Recent cleansing or product use alters reflectance; best on a bare, rested face.
Focal hyperpigmentation including sun spots, melasma patterns and post-inflammatory marks.
Cannot separate melasma from PIH or lentigines reliably without dermoscopy.
Dynamic and static lines on the forehead, periorbital and nasolabial regions.
Expression during capture changes line visibility; scores favor a neutral face.
Periorbital shadowing and infraorbital hollowing via contrast analysis.
Heavily affected by shadow, fatigue and allergies, not just skin biology.
Surface smoothness, pore visibility and micro-relief uniformity.
Resolution-dependent; fine texture needs high-quality, in-focus capture.
Grounded in established dermatology scales.
Channel banding draws on widely used clinical grading systems so the language of severity stays familiar to practitioners.
Acne severity banding (clear to severe).
Photoaging type I–IV for wrinkle and damage grading.
UV reactivity and baseline pigmentation classification.
What AiSkin is — and isn't.
- · An educational AI assessment of visible skin features.
- · Best on well-lit, in-focus, frontal faces without heavy makeup.
- · A tool to track relative change over time across comparable captures.
- · Designed to inform skincare guidance, not to diagnose disease.
- · A medical device, and not a substitute for a clinician's diagnosis.
- · Able to detect conditions beneath the surface (e.g. deep nodules).
- · Reliable under unusual lighting, motion blur, or partial faces.
- · Validated for every skin tone equally — we continually audit fairness.
AI-generated skin assessment. AiSkin provides educational skincare insights and is not a substitute for diagnosis or treatment by a qualified healthcare professional.
Peer-reviewed literature informing the model.
A curated reading list of dermatology and computer-vision research. These are the works of their original authors; AiSkin does not claim them as its own.
Links open a PubMed search for each citation so you can locate the primary source. AiSkin is an educational tool; always consult a qualified dermatologist for medical concerns.