AI skin analysis

Redness is the average a* value, the red-green axis of Lab color, sampled across the flat zones of the face and compared with a neutral baseline. Flushing, sensitivity, post-acne marks and warm lighting all push it up, which is why the reading is a measure of how red the surface looks right now, not a diagnosis of why. Tone evenness is the spread of L*, the lightness axis, between forehead and cheeks: dark spots, patchy pigment and uneven flush widen it, and a uniform complexion keeps it narrow.

T-zone shine compares lightness on the forehead and nose with the cheeks. Sebum pools in those zones and forms bright highlights, so the gap between them tracks how oily the surface is at the moment of the photo; it rises through the day and drops after cleansing, so time since washing matters. Depth reports the ITA angle, a value derived from L* and b* that skin research uses to place skin on a continuous light-to-deep scale. It describes depth only and says nothing about undertone, which the undertone test on this site handles separately.

The SkinPal app runs a fuller analysis from the same kind of selfie and scores eight metrics: acne (a count of localized red spots, not a diagnosis), texture (fine luminance variation within each zone), hydration (a proxy read from how smoothly the surface reflects light, not a measurement of water content), redness, skin tone including dark spots, oiliness, skin age (a rough apparent-age reading) and skin type. Its real advantage is not the extra metrics but the timeline: the same face, the same conditions, scored week after week, so a change in routine shows up as a trend rather than a guess.

Clinic imaging systems such as VISIA are built around control. Your face rests in a fixed position inside a hood, the light is standardized, and the camera captures several passes: standard light, UV light that fluoresces sun damage and porphyrins from bacteria in pores, and cross-polarized light that suppresses surface shine so pigment and vascular redness can be separated. They also compare you against a reference database of people your age and skin type. That combination reveals things a phone camera in ambient light physically cannot see, and it repeats precisely from visit to visit.

A phone-based skin analyzer reads surface appearance in whatever light you have. It has no UV pass, no polarizer, and no positioning hood, so it sees less and repeats less precisely. What it has instead is availability: it costs nothing, takes ten seconds, and you can run it every morning. Those are different jobs. The machine answers "what is the state of my skin, including below the surface, right now" once or twice a year. The app answers "is my skin trending the right way under my current routine" every week.

Neither replaces a clinician. A machine report still needs a person to interpret it, and an app score still needs the same. Price is the other honest difference: a clinic scan is usually bundled into a consultation you pay for, while a browser tool costs nothing and asks for nothing. The sensible pairing is a clinic scan when you want depth or have a specific concern, and a consistent at-home scan between visits so you arrive with a trend rather than a memory.

Light is the whole game. Face a window during the day with the sun not shining directly on you. Diffuse daylight is close to neutral in color, which keeps redness and tone readings honest, and it falls evenly across the face, which keeps texture and evenness honest. Avoid overhead bathroom lights (they carve shadows under the brows and cheekbones that score as unevenness) and warm evening bulbs (they push every color reading toward red and yellow). Flash creates specular highlights that read as oil.

Camera position matters next. Hold the phone at eye level, about an arm's length away, and look straight into the lens with a relaxed face. Tilting the chin up or down changes how light lands on the forehead and chin and shifts several scores at once. Clear hair off the forehead and cheeks so the landmark zones sample skin and not hair. Take the photo with no makeup, no skincare freshly applied (moisturizer shines), and every beauty mode, smoothing setting, and filter switched off.

Then do the same thing every time. Same window, same time of day, same distance, same expression. A single skin analysis score is a snapshot of your skin and your lighting together; only when the lighting is held constant does a change in the number mean a change in the skin. Consistency does more for the usefulness of these tools than any improvement in the algorithm could.

Skin changes daily. A poor night of sleep dulls tone and deepens under-eye shadow. Hormonal cycles move oiliness and breakouts on a roughly monthly rhythm. A cold day outside then a heated room inside will flush the cheeks for an hour. Heavy exercise, alcohol, spicy food, a new detergent: all of it shows up on the surface within a day. That means any single set of scores contains a lot of noise that has nothing to do with your routine or your long-term skin health.

The signal is in comparison. Scan under the same conditions once or twice a week, and after four to six weeks the scatter averages out and a direction emerges: redness drifting down since you dropped the exfoliating acid, texture improving on a retinoid, oiliness unchanged despite the new cleanser. Compare like with like (same light, same time) and look at the slope of the line rather than the height of any one point. Two readings a week apart prove nothing; twelve readings over three months tell you what your skincare is actually doing.

This page gives you the snapshot for free. Tracking the trend with guided capture, consistent framing, and a history to compare against is what the SkinPal app is built for.

It cannot diagnose. Rosacea, eczema, perioral dermatitis, fungal acne, and a plain flush can all present as elevated redness in a photo, and they need different treatment. A dark spot can be a freckle, post-inflammatory pigmentation, a solar lentigo, or something that needs a biopsy, and no color statistic separates those. Skin analysis describes appearance; diagnosis requires a trained person who can ask questions, look closely, and touch the skin.

It cannot see below the surface. Sun damage that has not yet surfaced as visible pigment, early inflammation under a forming breakout, and the hydration state of deeper skin layers are invisible to a camera in ordinary light. It also cannot predict how your skin will react to an ingredient. A tool can tell you your barrier looks compromised; only patch testing tells you whether a specific product will sting.

If something on your skin is new, growing, changing color or shape, bleeding, or failing to heal after a few weeks, see a dermatologist regardless of what any app reports. The same applies to persistent painful acne, redness that does not settle, or any rash that spreads. Use these tools for what they are good at, which is tracking cosmetic metrics between the appointments, not instead of them.

If one metric stood out, there is a focused free tool for it. The skin type test (/tools/skin-type-test) works from behavior rather than a photo, so it makes a good cross-check against the type this scan assigned. The dehydrated skin test (/tools/skin-hydration-test) separates dehydration, a lack of water, from dry skin, a lack of oil, which the surface reading above cannot do on its own. The acne severity test (/tools/acne-severity-test) turns a red-spot count into a grade you can act on.

For tone, the skin tone finder (/tools/skin-tone-finder) reads depth and undertone in more detail than the single ITA figure here. The skin age test (/tools/skin-age-test) runs the neural age estimate on its own and compares it to your actual age. The pore visibility check (/tools/pore-visibility-check) breaks the texture reading down by zone. And before you buy a product to address any of this, the comedogenic checker (/tools/comedogenic-checker) will flag the ingredients most likely to clog pores.

For the reasoning behind all of it, the guides hub (/guides) covers skin conditions, ingredients, and routines in plain language. Each guide names the tool that measures what it discusses and links back here, so you can move between reading and measuring without losing the thread. The same rule applies throughout: cosmetic self-assessment, clearly stated limits, and a clinician for anything that looks like more than cosmetics.