baselit.app / blog / skin analysis app
[ skin analysis · ai skin analysis · skin score ]
What does an AI skin analysis app actually see?
Short version: An AI skin analysis reads a selfie and describes what is visible on the surface: how even the tone looks, how fine the texture appears, where the skin shines, where it looks dry. That usually becomes one number between 0 and 100 plus a handful of sub-scores. It is a cosmetic snapshot, not a medical assessment.
Almost all of the value sits in repetition. One scan says very little; eight weekly scans under the same light show a direction. What an app cannot do: see below the surface, judge a mole, or diagnose anything. That is what a dermatologist is for.
“Skin analysis” now covers three fairly different things: a device on a counter in a shop, a selfie tool on a beauty brand’s website, and an app on your own phone. The underlying technique overlaps; the question each one answers does not. This page sorts that out, explains how a photo becomes a number, and states the limits before anyone else has to.
What is a skin analysis?
A record of what is currently visible. Not a verdict, not a grade, and explicitly not a diagnosis. A facialist does it with a lamp and a trained eye. A phone does it with a vision model on a photo.
What gets read is roughly the same either way: how even the tone looks, how fine or rough the surface appears, where the face shines, where it looks dull or tight. None of that is below the skin. All of it is surface, which happens to be exactly where skincare works.
How does an AI skin analysis app work?
In four steps that barely differ between products:
- Capture. One or several selfies in a row. Several frames beat one, because a single frame can be soft or badly lit.
- Normalisation. The image is scaled and levelled before anything is scored. Skip this and you mostly score the lamp in the room.
- Separate sub-scores. Not one grade but several. In Baselit there are five: clarity, evenness, texture, hydration and radiance.
- Summary. The sub-scores are folded into one figure, usually on a 0 to 100 scale.
The interesting part is not the number, it is what happens next. An analysis that only hands you a score leaves you alone with a score. It becomes useful when it turns into a concrete order of steps you actually apply, and when you look again a few weeks later.
How is a skin score from 0 to 100 calculated?
By weighting and combining the sub-scores. The scale itself is a convention: 0 to 100 reads easily, and that is the whole reason it exists.
Two things separate a number you can work with from one that just looks tidy. First, stability: the same photo has to produce the same value, otherwise you are tracking chance. Second, a noise floor: small swings between two scans are normal and should not be reported as change. There is a longer version of this on how skin scores work.
The only comparison group that counts is you. A skin score across two people is about as meaningful as comparing two photos taken in two different flats. Across your own scan today and your own scan eight weeks ago, it is genuinely useful, as long as the light and the pose stayed put.
What can a skin analysis app see, and what can it not?
| Question | Answer | Why |
|---|---|---|
| How even does my tone look? | Yes | Pure surface information, readable from a photo. |
| Where does my face shine? | Yes | Shine is light on a surface, so it is visible. |
| How fine does my texture look? | Yes | Visible, as long as the frame is sharp enough. |
| Is anything changing over eight weeks? | Yes, this is the strength | Comparing you to you is the most solid thing an app does. |
| How hydrated is my skin really? | Impression only | From an ordinary selfie this is an impression from visible cues, not a probe measurement. |
| Is this mole fine? | No | That is a clinic question. Cosmetic apps do not check moles. |
| What skin condition do I have? | No | A cosmetic app does not diagnose, and should not pretend to. |
| What is happening below the surface? | No | A photo carries surface. Anything deeper needs other instruments. |
App, in-store device, or dermatologist?
Three tools for three questions. None replaces the others, and the split is simpler than it sounds:
| Route | Answers | Limit |
|---|---|---|
| App on your own phone | “Is anything moving since I changed my routine?” | Light and pose are not controlled, so the capture takes discipline. |
| In-store device analysis, for example at a beauty retailer counter | “How does my skin look today under controlled light, and what does someone recommend?” | One appointment, not a trend. The advice is tied to a range of stock. |
| Brand tool such as L'Oréal Skin Genius | “Which products from this brand suit my selfie?” | The suggestion space is that brand’s own catalogue. |
| Dermatologist | Everything medical: pain, bleeding, anything that will not settle or changes quickly, and every mole. | Not a tool for a weekly look in the bathroom. |
This is not a ranking. A counter device produces cleaner images at its appointment than any selfie will, because the camera and the light are fixed. An app produces worse single images and twenty of them across five months. If you want to know whether a change worked, you need the twenty.
How accurate is an AI skin analysis?
Accurate enough for direction, too coarse for decimals. The biggest error source is not the model, it is the bathroom: the same face looks different by a window, under a ceiling light, and in a mirror cabinet. Change the room and you measure the room.
The second error source is timing. Straight out of bed, after exercise, and the day after a short night are three different starting points. Which makes the most useful rule about any skin analysis app boringly practical: same spot, same time, same distance.
How do you take a selfie that stays comparable?
Face the window rather than turning your back to it. No flash, no backlight, no coloured lamp behind you. Light is the single biggest source of noise in any skin analysis.
No make-up, no freshly applied moisturiser, no sunscreen. A film on the surface changes shine and tone, and therefore the result.
Straight on, neutral expression, hair back, glasses off. A face shot from below is a different face to the model.
Note the window, the time and the posture. Use the same combination next time. Only then does the trend line compare skin instead of lighting.
Do not scan daily. A weekly gap across eight weeks gives you a line; a daily gap gives you noise.
What do you do with the result?
An analysis on its own changes nothing. A number becomes useful when three things follow from it: a short list of products, a fixed slot for each of them morning and evening, and a date to look again.
The usual mistake straight after a skin analysis is buying six bottles at once. After that you cannot tell which one helped and which one made your face interesting. One strong active plus gentle support is a complete routine; which pairs genuinely clash is on which skincare ingredients you should not mix.
How Baselit does the analysis
No marketing, just what is built:
- A selfie scan from several frames, normalised before anything is scored. The same photo returns the same value.
- One value from 0 to 100 plus five separate sub-scores: clarity, evenness, texture, hydration and radiance.
- Face zones, not only numbers. For a weaker sub-score the scan names the zones involved (forehead, nose, cheeks, chin, jawline, eye area) and outlines them on your own selfie, never colouring the skin itself. Shine is reported as its own observation, separate from radiance.
- A weekly re-scan and an eight-week trend, with a noise floor of plus or minus three points so a coincidence is not reported as change. The routine’s focus only shifts when three scans agree.
- The scan turns into a routine with fixed AM and PM slots and a reason attached to every single step.
- Mia, the coach in the app, reads the current scan and the current routine before every reply. Send a photo of a product and she says whether its actives clash with your routine and which slot it belongs in.
- Photos stay on the device.
- A clear pointer to a clinic when a result comes back pronounced, instead of talking on.
And the limits we name ourselves: hydration from an RGB selfie is an impression from visible cues, not a probe measurement. Moles are not checked. Even lighting is a precondition. The app is iPhone only; there is no Android build and no web version.
Common questions
What is a skin analysis?
How does an AI skin analysis app work?
How is a skin score from 0 to 100 calculated?
How accurate is an AI skin analysis?
What can a skin analysis app not do?
Is a skin analysis app better than a device in a shop?
Does a skin analysis app replace a dermatologist?
How do I take a selfie that is worth comparing?
How often should you scan your skin?
What do you do with the result of a skin analysis?
Baselit is a cosmetic skin app, not a medical device. This page describes how skin looks on the surface and how to compare that over weeks. It is not a diagnosis and does not replace seeing a doctor.