The Kurl
Can AI analyse my hair from a photo? What it reads and what it misses
Yes, within limits you should know before you trust the answer. A clear photograph shows a model your curl pattern, how much scalp shows through at the part and the crown, where your colour stops and your own hair begins, and how worn the last few inches are. Light bounces off all of those, so a camera records them.
What a photograph cannot hold is everything you learn by touching hair. How wide one strand is. How quickly it takes in water. Whether it springs back when you pull it. A tool that reads your porosity off a selfie is guessing, and most of them do not mention that.
This post draws the line between the two lists, says what makes a photograph usable, and gives you a test you can run on any reading you are handed.
"Hair analysis" means three unrelated things
The phrase "hair analysis" is used for at least three different services, and mixing them up leads people to tools that cannot do what they hope.
An AI photo reading, the subject of this post, describes what a camera can see: curl pattern, visible density, colour, condition. A commercial hair mineral test, sold directly to consumers, chemically analyses a cut sample of hair and is a separate, disputed method not endorsed by major medical bodies for diagnosing nutrient levels or food sensitivities. The British Dietetic Association states plainly that using hair analysis for allergy testing is unproven and has no scientific basis. And forensic hair analysis, comparing a hair from a crime scene against a suspect under a microscope, is a third field again, with its own troubled history: the FBI has acknowledged that its examiners gave erroneous testimony in at least 90% of microscopic hair comparison cases reviewed from before 2000. None of that is about AI photo tools, and all of it is a reason to ask which "hair analysis" a service offers before trusting the name alone.
What is in the picture
A photograph is a record of light. Hold your phone at arm's length and the width of your part fills a few hundred pixels, which is enough to see that scalp is visible and nowhere near enough to measure anything about a single hair.
Clinics do not work that way. Trichoscopy presses a magnifying camera against the scalp and examines hairs at 20 to 1000 times magnification, according to a 2024 systematic review in the Journal of Clinical Medicine. Density is counted as hairs per square centimetre, and the normal range depends on who you are: a 2017 study in Skin Appendage Disorders measured 230 hairs per square centimetre at the frontal scalp in Caucasian participants, 174 in Hispanic Americans and 160 in people of African descent.
A phone cannot do what a trichoscope does. The useful question is what is left over, and the answer is more than most people expect.
What AI can read from a photograph of your hair
Curl pattern is the easiest thing in the picture, provided the hair is dry and has no product in it. Straight, S-shaped, spiral, zigzag: all of that is outline, and outline is what cameras are best at.
Visible density at the part and the crown comes next. This is coverage, not a count: how much scalp shows along the part line, and how much shows at the crown when you tip your head forward. The crown is the one that matters, because it is the view you never see of yourself.
The regrowth line is readable if your hair is coloured, because there is a band where your own hair takes over. Scalp hair grows at about 0.35 mm a day, roughly 1 cm a month, so 3 cm of regrowth puts your last colour about three months back. That is arithmetic a model can do from one picture, and it changes the advice: hair with a three-month line has two different conditions on one head.
Rough, worn ends show too. Split ends themselves are small, and what shows in a photograph is the effect of them: a ragged outline at the bottom instead of a clean one, short broken hairs standing away from the length, and ends that catch light unevenly.
Shine at the roots is visible because oil flattens the surface at the scalp, and a flat surface reflects in one direction. Paired with the answer to when you last washed, it becomes information. Shine six hours after a wash means something different from shine on day three.
Colour condition is the last one. Banding, brass, and the point where bleach stops are all visible, and all of them change what belongs in a routine.
What it cannot read, no matter how good the photograph
Strand width, as a number, is out of reach. Fine, medium and coarse are separated by tens of micrometres. A 2026 review in the Journal of Clinical and Aesthetic Dermatology looked at ten AI hair analysis tools and found the ones that measure thickness use dedicated trichoscopes at 15 to 200 times magnification. A model can put you in a wide band from how your hair behaves in the picture. It cannot give you a number.
Porosity is a behaviour, not an appearance: how fast hair takes water in and how fast it lets it go. Damaged hair is more porous, and the 2015 review Hair Cosmetics: An Overview puts it plainly, that hair is porous and damaged hair is intensely so. So a photograph of damaged ends is a hint. It is not a reading. The porosity tests you can do at home all involve water and time, which is the part a camera has no access to.
Elasticity is found by stretching a wet hair between two fingers. The same review reports that wet virgin hair stretches to 30% beyond its length and returns, while stretching between 30% and 70% causes permanent change. That is a test with your hands, and it tells you things a picture never will, which is why elasticity deserves its own check before a big change.
Smell is the one people forget. A scalp with a yeast problem often smells before it looks like anything. Your stylist notices in the first thirty seconds. A camera records nothing.
A diagnosis is the last item, and this one is a category error rather than a technical limit. The American Academy of Dermatology describes a dermatologist as a medical doctor who specialises in treating the skin, hair, and nails. A photo reading can say that scalp is showing at your crown. Only a doctor can say why.
The photographs decide the answer more than the model does
This is the part people skip, and it changes the result more than which AI is behind it.
Take three angles: front, a small turn to one side, and the top of your head. The top one is the one almost nobody takes, and it is the only one that shows the crown.
Use daylight, near a window. A bathroom ceiling light sits above your head, which puts the crown in its own shadow and darkens the part line you are trying to show. Face a window instead, with the light coming at you and not down on you.
Keep the hair unstyled and dry. Product coats the surface and changes how it reflects. Wet hair is worse still, because wet hair clumps together and shows more scalp than the same hair dry, so a wet photograph reports thinner hair than you have.
Second-day hair is better, if you can manage it. Freshly washed hair is at its best and least typical. Day two is closer to the hair you live with, and the routine is being written for the hair you live with.
The 2026 systematic review of AI in trichology names this problem directly: differences in imaging device, magnification, lighting and polarisation between datasets may introduce dataset shift. If that matters between research datasets, it matters between your bathroom and your kitchen window.
How the Kurli AI Specialist does it
Four questions, answered by tapping: how much time you have in the morning, how often you wash, when you last washed, and what is bothering you. The last one is optional, because plenty of people cannot name what is wrong, which is why they came.
Then three photographs: your face, a small turn to one side, and the top of your head. The angle check runs in your browser and the camera takes the shot itself once your head is at the right angle, so you do not have to judge the angle yourself.
What comes back is a reading of the hair in those pictures (hair type, texture, density, scalp, damage, any colour or heat treatment, and a plain summary), one change to start with, wash-day steps with a reason attached to each one, what to do between washes, a seven-day calendar marking which days are wash days, the ingredients to look for and to avoid with a line on why each matters for you, what to buy described by what the label says rather than by brand, and your own photographs marked up. Follow-up questions cost nothing extra and your photographs go back with the question. If something in the pictures belongs with a doctor or a stylist instead of a routine, the plan says so.
One payment buys one routine. There is no subscription, and there is no charge if the routine fails to write. A full worked example is published in the open, which is the fastest way to judge whether a reading like this is worth anything to you. The Kurli AI Specialist is the thing that writes it.
Where the research on AI hair analysis stands
Two reviews published in 2026 give the clearest picture, and neither is a sales page.
The systematic review in Digital Health found real strength in structured, repeated tasks: strong agreement for automated alopecia severity scoring, and 96.1% sensitivity with 88.2% specificity separating scalp psoriasis from seborrheic dermatitis. It also found that the evidence base is dominated by retrospective single-source datasets with limited external validation, that only one of the fifteen included studies validated performance across diverse ethnicities, and that no included study showed AI assistance improving clinical decisions or patient outcomes.
The review of ten commercial tools in the Journal of Clinical and Aesthetic Dermatology is blunter about the products: a scarcity of peer-reviewed validation, reliance on manufacturer-provided data, and training-set diversity that was often left unspecified.
A model is strong on the narrow task somebody checked, and weak on every claim nobody checked. That is the whole story, and it applies to us too. The Kurli team
How it compares with a stylist putting hands in your hair
A stylist has three inputs no camera gets. Weight, felt by lifting a section. Spring, felt by pulling it and letting go. And the scalp itself, felt through the fingertips as tight, loose, dry or flaking. Those matter. They are how an experienced person tells fine hair with a lot of it from coarse hair with less.
What a photo reading gives back is different rather than better. It shows you the top of your own head, which you have never seen. It produces a written record you can reread in March when you have forgotten what was said in January. And it is available at eleven at night when no salon is open.
The comparison people get wrong is treating them as rivals. Use a reading to arrive at the salon knowing what to ask, and the twenty minutes in the chair go further. The longer version of that argument is in what AI hair analysis can and cannot see, and the wider question of what any consultation should give you sits in the hair consultation guide.
The test to run on any reading you are given
Open the reading and your own photographs side by side. Go line by line and mark each statement one of two ways: I can see that, or where is that.
Curl pattern, visible scalp at the crown, a colour line, ragged ends: those should land in the first pile, and you should be able to point at them. Porosity, strand thickness in micrometres, a named condition, a cause: if those appear with no hedging, they belong in the second pile.
A reading with a short second pile is doing its job; one with a long second pile is describing hair in general and putting your name on it. Trust least the reading that never says what it is unsure about. Uncertainty is the most honest thing a model looking at a photograph can report.
Common questions
How to take the best hair pictures?
Stand facing a window in daylight, on dry hair with no product, and take three shots: your face from the front, a small turn to one side, and the top of your head. Switch off any beauty filter or portrait mode, because those smooth the fine edges that carry the information about your ends. Day-two hair shows how your roots and ends behave on an ordinary day; freshly washed hair hides it. Ask someone to take the crown shot if you cannot reach.
Is hair analysis legit?
It depends which service is meant. A photo reading is legitimate for what light shows: curl pattern, visible scalp at the part and crown, a colour line and worn ends. Consumer hair mineral tests that claim to read nutrient levels or food sensitivities from a cut sample are not endorsed by major medical bodies, and the British Dietetic Association says hair analysis for allergy testing has no scientific basis. Forensic hair comparison has its own troubled record. Ask what is being read and how before you trust the label.
What does a mineral hair analysis test for?
It claims to measure mineral and metal levels in a cut sample of hair and to read nutrient status or food sensitivities from them. Major medical bodies do not endorse it for that, and the British Dietetic Association states that hair analysis for allergy testing is unproven and has no scientific basis. It has nothing to do with a photo reading, which looks at the visible condition of hair on your head. For a vitamin or mineral level, a blood test through a doctor is the reliable route.
Is an AI hairstyle app legit?
Judge it by two tests. First, does it name things you can point to in your own photograph? A reading that says your ends are rough when you can see rough ends is a real observation; one that announces your porosity is guessing, because water uptake cannot be photographed. Second, does it say plainly where your photograph goes, whether a copy is kept and how to delete it? An app that passes both is legitimate for what a picture can show, and no more than that.
Can AI suggest a hairstyle?
Yes, and a suggestion is where it is most reliable, because a hairstyle suggestion is advice, not a finding. A model can see your curl pattern, your length, how much hair you have and the shape of your face, and those are the inputs a stylist uses too. Whether the suggestion suits your morning is still your call.
Are AI hairstyle apps safe to use?
Check three things before uploading a photograph of your face: where the image is sent, whether a copy is kept, and how you delete it. A privacy policy that does not answer those three plainly is the warning. Photographs of faces are personal data in most countries, so a service that will not say where yours goes is one to skip.
Can AI fix hair in a photo?
An image model can redraw hair in a photograph, which is editing rather than analysis. That is useful for seeing a style on your own face before you book it, and it changes nothing about your actual hair. Keep the two apart: a picture of you with a fringe is a preview, not evidence that a fringe will fall that way.
How would AI style my hair?
It works from what a photograph shows: your curl pattern, your length, how much hair you have and the shape of your face, then suggests shapes that tend to suit that combination and can draw them onto your picture so you can react. What it does not know is your morning: how many minutes you have, whether you use heat, and how your hair behaves by day two. A suggestion is advice, so treat the rendered picture as a preview and check the shape against the time you have.
Can you add hair to a photo?
Yes. An image model can draw more hair onto a photograph, fill a parting or add a fringe, and the result can look convincing. It is a picture, not a forecast. Adding hair in a photo tells you nothing about whether your own hair can grow or be styled that way, which depends on your density and condition. If the reason you want more hair in the picture is that your parting looks wider than it used to, that is a question for a dermatologist, not an editing tool.
How can you tell what kind of hair type you have?
Wash your hair, use no product, let it dry with no heat, then look at a single dry strand in daylight. Straight through its length is type 1, an S bend is type 2, a spiral you could wrap around a pencil is type 3, and a tight zigzag is type 4. Do it on clean hair, because product and heat both flatten the pattern you are trying to see.
How can I tell if my hair is fine?
Roll one clean strand between finger and thumb. If you can barely feel it, it is fine; if it feels like a thread, it is medium; if you can feel it clearly as a round object, it is coarse. Fine describes one strand. Thin describes how many strands you have. People often say one and mean the other.
What happens to my photographs after an AI hair reading?
That depends on the service, and you should read the answer before you upload. With the Kurli AI Specialist the angle check runs in your browser and the camera takes the shot only when your head is at the right angle. The three photographs are sent to Google Gemini to write the plan, Kurli keeps a copy so results can be checked, they are never sold or published, and emailing hello@kurli.me has them deleted.
Can AI cut hair?
No. An AI model can generate an image that shows what a cut might look like, or write instructions describing one, but it has no way to physically cut anything. Any tool that shows you 'cutting' your hair is producing a preview image, not performing or guiding a cut, and the actual cutting still needs a person with scissors.
Can AI change hair colour?
In a photograph, yes. An image model can redraw your hair a different colour so you can see roughly how it might look, the same way it can preview a different length or style. That preview does not tell you what the actual chemical process would do to your specific hair, which depends on your current colour and condition.
How much does a hair analysis cost?
It depends entirely on which kind of 'hair analysis' is meant, since the phrase covers several unrelated services at different prices, from free AI photo tools to paid trichologist appointments to laboratory mineral tests. Ask what specific service and price you are being quoted before comparing options, since the name alone does not tell you.
Does hair analysis work for food intolerance?
No. Hair contains no antibodies, and food intolerance and allergy testing relies on markers found in blood, not hair. The British Dietetic Association states plainly that hair analysis for allergy testing is unproven and has no scientific basis. A test that uses a hair sample to diagnose food intolerance is not a credible diagnostic method.
What does a hair analysis test for?
It depends which service you are being sold, because the phrase covers unrelated things. A photo reading looks at what is visible: curl pattern, how much scalp shows at the parting and crown, the regrowth line, oil at the roots and wear at the ends. A laboratory hair mineral test claims to measure mineral and metal levels in a cut sample, and major medical bodies do not endorse it for reading nutrient status. Ask which kind you are getting and what it will hand you at the end.
What is hair analysis in forensic science?
A different field entirely from anything covered here: comparing a hair found at a scene against a suspect's hair under a microscope. It has a troubled record. The FBI has acknowledged that its own examiners gave erroneous testimony in at least 90% of microscopic hair comparison cases reviewed from before 2000, which is why the technique is now used with far more caution.
See it on your own face
Upload a selfie and Kurli recommends the hairstyles that suit you, then renders each look onto your photo.