AI Skin Analysis on Dark Skin: Does It Work on Deeper Skin Tones?

AI skin analysis works on darker skin tones, but usually less accurately, because most skin image datasets underrepresent deeper skin. In a 2023 study of selfies from 1,041 US women, an AI's grades for dark spots matched dermatologists' at r = 0.57 on the lightest phototype but only 0.16 to 0.17 on the two darkest. Pigmentation and redness are the hardest scores to trust; folds and sagging hold up better.

Below is where the bias comes from, which metrics suffer most on deeper skin and why, how to light and frame a selfie so your results are more consistent, and the questions to ask an app before you rely on its numbers.

Why can AI skin analysis be less accurate on darker skin?

AI models learn from the images they're trained on, and those images have mostly shown lighter skin. A systematic review of publicly available skin image datasets found 21 open-access sets with 106,950 images, but skin type data for only 2.1% of images and ethnicity for 1.3%, with most datasets that reported their origin coming from Europe, North America and Oceania. The authors described a substantial under-representation of darker skin types.

The effect shows up when models are tested. A Stanford-led team built a diverse, pathology-confirmed image set and found that state-of-the-art dermatology AI models performed substantially worse on dark skin tones. Dermatologists who label training data also performed worse on those images, which means the "correct answers" a model learns from can carry the same blind spot. Fine-tuning the models on the diverse images closed the gap between light and dark skin, which suggests the problem is fixable with better data.

Those studies looked at disease-detection models. Cosmetic skin analysis apps that score spots, pores and wrinkles face the same underlying issue. The camera adds another layer: a 2025 photography guide in JAMA Dermatology notes that many photographic technologies have historically calibrated color correction on people with light skin, producing inaccurate photos of skin of color before any AI sees them.

Which skin metrics are hardest to measure on deeper skin tones?

Pigmentation and redness are the hardest. Both depend on reading small color differences, and melanin changes how those differences look in a photo. Structural features such as wrinkles are easier because they depend more on shadow and shape.

The best direct evidence comes from a cross-sectional study of selfies from 1,041 women in the US, of different ages, ancestries and skin phototypes, graded both by an AI system and by 50 dermatologists. For five facial signs, the AI's grades correlated strongly with the dermatologists' (r of 0.75 or higher), cheek pores correlated moderately (r = 0.63) and pigmentation signs correlated weakly (r = 0.40), especially for the darkest skin tones. Age and ancestry had no effect on the correlations, and the authors concluded that skin tone still needed improvement.

The full paper shows how steep the drop is. The correlation for density of pigmentary spots was 0.57 for phototype I but 0.16 for phototype V and 0.17 for phototype VI, and the average across all seven signs fell to 0.68 and 0.63 for those two darkest groups. Diffuse redness scored well overall (0.79), but the reference atlas used for African American women had no redness scale, so redness wasn't graded for that group or for phototype VI at all. Most of the authors work for L'Oréal, which makes the system, so the study is worth reading as a company validating its own tool, though one that published its weak spots.

Why redness is hard to score

Redness shows less contrast against deeper skin, so a score built on reading pink tones can miss it. A review of photographs from atopic dermatitis trials highlighted challenges in evaluating erythema in patients with dark skin types. If clinicians find it difficult in standardized trial photos, a phone selfie in kitchen lighting is harder still. Our article on what a redness score can and can't tell you goes into the measurement itself.

Why pigmentation matters most on deeper skin

Pigmentation is also the metric people with deeper skin often care about most. Postinflammatory hyperpigmentation, the dark marks left after acne or irritation, tends to affect darker-skinned patients with greater frequency and severity, and dyschromias are among the most common reasons people in darker racial and ethnic groups see a dermatologist. So the score that is least reliable is also the one many people most want to track. Our explainer on how pigmentation analysis works covers how apps detect spots.

Does the Fitzpatrick scale measure skin color?

No. The Fitzpatrick scale (types I to VI) describes how skin burns and tans in the sun, and many apps and studies still use it as a stand-in for skin tone. The 2023 selfie study noted that using phototypes as discontinuous categories is likely a limiting factor for grading, whether by AI or dermatologists. If an app asks for your Fitzpatrick type or reports accuracy "for types V and VI," it's useful information, but it's a coarse grouping. See Fitzpatrick skin types I to VI explained for what the scale does measure.

How should I light a selfie for skin analysis on deeper skin?

Use soft, even light from in front of you and keep your setup identical every time. The JAMA Dermatology guide recommends soft, diffuse lighting, such as from a ring light, and a fixed distance when photographing darker skin, and notes that a royal blue background is often preferred in clinic photos because it adds contrast without distorting skin hue. Consistency matters as much as quality: in a smartphone study using color standards that represent diverse skin tones, changing the camera angle shifted both lightness and color, changing the distance shifted lightness, and the flash didn't reduce the variability.

1. Face the light. Stand facing a window in daylight, or use a ring light at eye level. Light from behind or directly above puts parts of your face in shadow, which an app can misread.
2. Use one light source. Mixing warm bulbs with daylight can shift how your skin color comes out from one photo to the next.
3. Skip the flash. In the study above it didn't make color more consistent.
4. Fix your distance and angle. Use a tripod or mark a spot, and hold the phone straight at eye level.
5. Use a plain background. A single-color wall keeps the frame simple; clinics photographing darker skin often prefer royal blue.
6. Scan bare-faced, at the same time of day. Makeup, tinted sunscreen and filters change what the app reads.

Our guide to the best lighting for skin photos has a fuller setup.

What should I ask a skin analysis app before trusting its scores?

Ask how the app was tested on skin like yours, and test its consistency yourself. A short checklist:

• Has it published accuracy by skin tone? A peer-reviewed study that breaks results down by tone, like the one above, is the strongest evidence. A marketing claim of "works on all skin types" isn't.
• Which scores does it validate? Strong results for wrinkles don't carry over to pigmentation or redness.
• Is it consistent for you? Take three scans a minute apart in the same light. If the pigmentation score swings widely, treat that score as noise.
• Does it show you what it measured? An app that explains region-by-region results is easier to sanity-check than a single number.
• Does it say what it can't do? Look for a clear statement that it isn't a diagnosis or a mole check.
• What happens to your photo? Check whether images are stored, for how long, and whether they're shared or sold.

Can an app check moles or skin cancer on dark skin?

A cosmetic skin analysis app can't, and you shouldn't rely on one for it. The AAD notes that skin cancer in people with darker skin often appears in areas with little sun exposure, such as the palms, soles, nails, mouth and genital area, and is often advanced by the time it's diagnosed. See a dermatologist for any spot that's new, changing, bleeding or that won't heal, including on the palms, soles and nails.

How can a selfie skin scan help on deeper skin?

The most reliable way to use any skin score on deeper skin is to compare yourself with yourself. Pori takes a selfie in about 10 seconds and reads your face region by region with a 468-point face mesh, returning an overall skin score plus Hydration & Moisture, Pores & Texture, Pigmentation & Tone and Anti-aging & Elasticity, with progress charts and a before/after comparison. Used with the same light and distance each time, those trends tell you more than any single result, especially for pigmentation. Photos are sent over HTTPS for analysis, aren't kept on servers after the result and are never sold; your history stays on your phone. The first scan is free, and Pro costs $9.99 a week with a 3-day trial for eligible Apple IDs, or $49.99 a year, in the US as of October 2026.

Pori isn't a medical device and doesn't diagnose conditions or check moles. For dark marks that won't fade, persistent redness or any changing spot, see a dermatologist, ideally one experienced with skin of color. For day-to-day care, see our skincare routine for dark skin.