AI Face Shape Detector: What Is My Face Shape?

This AI facial analysis measures forehead, cheekbone and jaw width against face length, analyzed on your device, and returns your class, plus the margin to the next class. A class describes the outline in one photograph. It does not rank you.

The Short Version

Each of the seven shapes is a different relationship between four measurements: face length, forehead width, cheekbone width and jaw width. The detector reads those from the landmark mesh and reports which pattern your numbers sit closest to, along with how close the call was.

How Can I Identify My Face Shape?

This detector runs the same comparison on a live photo. A browser-based landmark mesh finds the four points, measures each span, and applies the classification rules published below. You can run the identical comparison yourself, with a tape measure and five steps.

1. Pull your hair back off your forehead and both temples.
2. Measure straight across the widest part of your forehead.
3. Measure across your cheekbones, just below the outer corners of your eyes.
4. Measure jaw angle to jaw angle, at the widest point below your ears.
5. Measure from your hairline to the point of your chin.

A phone, a tape measure and the classification rules below return the same class this detector does.

How This Detector Measures Your Face, and What It Does About the Hairline

This detector runs a 478-point MediaPipe landmark mesh in your browser, fits a horizontal eye line to correct for head tilt, and reads four spans from the corrected mesh. Forehead width runs across the widest visible span of your forehead. Cheekbone width, the bizygomatic width, runs across the widest point of your face, just below your outer eye corners. Jaw width runs gonion to gonion, the two points where your jaw turns up toward your ears. Face length runs from your visible hairline to menton, the lowest point of your chin. Each measurement is reported with both endpoints named, so you can check it against your own photo.

The classical face length starts at the trichion, your natural hairline, and no landmark model can see a hairline through hair. Every detector on this search result either guesses that point or quietly substitutes a forehead landmark instead. This one publishes the substitute it uses, a fixed visible-window convention, and lets you place a marker on your own hairline to see how the class changes.

The three ratios that decide your class

Cheekbone width is the reference each ratio uses, because it is the most visible span on a straight-on photo. A high length-to-cheekbone ratio pushes the class toward the longer shapes. Whichever of jaw and forehead measures wider than the other decides which of those longer or shorter shapes it becomes.

The Seven Face Shapes, and the Rule That Decides Each One

Seven classes cover this taxonomy: oval, round, square, heart, diamond, oblong and triangle, ordered here by search volume rather than by preference. Each one is defined by which of the three measurements is widest and by how long the face reads against that width. No class in this order ranks above another.

None of these seven rules stands alone. A face longer than it is wide still reads as round when the width difference between forehead, cheekbones and jaw stays small, because the length ratio and the width order are checked together, not one after the other. That is also why two people with the same length-to-width ratio land in different classes: the second test, which measurement is widest, is doing real work.

Rectangle, long, and pear: the names that mean the same shapes

The Rules This Detector Classifies By

This detector assigns a class with two published tests run together: a ratio test comparing face length to cheekbone width, and a width order test ranking forehead, cheekbone and jaw width against each other. A jaw angle test separates the angular classes, square and triangle, from the curved ones. Every threshold behind those tests is fixed under one version, Shape v1.0, and published in full on the measurement points, the thresholds and the version.

A result counts as a genuine tie when the winning class's rule score sits within a published margin of the runner-up's. That margin is fixed on the same methodology page and printed again on every result card, so the number a visitor sees in the result region matches the number printed there. The thresholds are chosen first and published before this page goes live, and a later change carries a dated changelog entry, because a class that flips six months later deserves a reason.

What If You're Between Two Face Shapes?

A margin is the distance between your winning class and the runner-up, and it is what a class name alone can never show. When that distance falls under the published tie threshold, the result is a genuine two-class tie, not a misclassification. Both guides apply, and the styling advice from two neighboring classes mostly agrees anyway, because the underlying measurements are close. This is the section every boundary question on this topic is really asking about: oblong against oval, round against oval, heart against diamond. None of those visitors were misclassified. Their face sits near a line, and the margin says exactly how near.

What Anthropometry Calls Face Shape, and What It Publishes

These are anthropometric studies of specific national samples. They measured an index rather than our seven classes, and they report distributions, not verdicts about any one face. Two studies below name their population and sample size for exactly that reason.

The styling taxonomy this page uses has no published population distribution of its own. And the one time a face shape typology was tested against an independent measurement, Williams' claim that face shape predicts tooth form, the association was not significant, tested biometrically in 1987 [4] and again with digital tracings on 200 subjects in 2019 [5].

Face Shape for Men and Women

Measured facial dimensions do differ by sex. In a 200-subject index study, men's facial height, bizygomatic breadth and facial index were all significantly greater than women's [2]. But the distribution of the five anthropometric phenotypes across that same sample did not differ significantly by sex, chi-square 6.07, 4 degrees of freedom, p equal to 0.194 [2], measured by the same dimorphism approach used across this literature [12].

What Is the Most Attractive Face Shape?

Here is what the evidence says. Shape symmetry, averageness and sexual dimorphism, measured from images and tested together, found that averageness predicted attractiveness ratings in both male and female faces, and femininity predicted them in female faces, while masculinity and symmetry did not [6]. That is the strongest single result in this literature, and it does not mention outline categories at all.

Morphometric masculinity told the same story a second way. A discriminant function that classified over 96% of faces correctly by sex showed no relationship with rated attractiveness, while skin colour did [7]. A measurement built specifically to separate male and female face shape still failed to predict how attractive either group was rated.

The review position agrees: averageness, symmetry and sexual dimorphism are the studied dimensions of facial attractiveness [8][9], and none of them is which of six or seven outline categories a face falls into. Your class does not set how attractive you are. It tells you which proportions a style choice is working with.

Can AI Detect Face Shape? And What Makes One Detector Better Than Another?

People classifying face shape by eye show very low agreement, between different viewers, within the same viewer, and even across repeated presentations of the same photograph to the same person, and training in the method did not improve face-matching accuracy [1]. A model that measures the same four points every time and compares them against a fixed rule does not have that problem.

That also answers what makes one face shape detector better than another: the one that names which points it measured, which rule it applied, and how close the call was, not the one with the most confident font. This detector publishes no accuracy percentage, because no validation study exists yet to justify one. See how we measure, and where the measurement fails.

What This Score Cannot Tell You

You now have a class and the four measurements behind it. What is left is what that class is worth, how to check it, and what would change it.

You Have a Shape Now, What Does It Actually Mean?

Your class is a summary of four measurements, forehead width, cheekbone width, jaw width and face length, sorted through published rules. Three questions matter now: what the label is actually worth, how to confirm it before you act on it, and what could change it. The three sections below take them in order.

What a shape label is actually worth

A class is a bucket, and the margin is how confidently your measurements landed in it. Neither one is a score. A wide margin means your face sat clearly inside one class; a narrow one means it sat close to a line between two. This page has no score to interpret, so the module that usually explains what your score means points instead to where the site's actual score vocabulary lives: what a 1-10 attractiveness score really means. A descriptive class and a 1-10 attractiveness score answer two different questions.

How to confirm your shape before you act on it

1. Retake the photo with your hair fully back off your forehead and both temples.
2. Use your rear camera at arm's length or further, with the lens level at eye height.
3. Keep your chin level and your expression neutral, with your mouth closed.
4. Compare the four measurements themselves, not just the class name, between the two photos.

A gap smaller than your printed margin is noise from the photo, not a new face. A gap that flips your class is the system telling you your face sits close to a boundary, which is a correct result, not an error.

What changes your face shape, and what only changes the photo

The photograph moves your result fastest and most. Hair over your temples, a tilted chin, camera distance and a smile all shift the measured outline immediately, and none of them touch your face [13]. Soft tissue moves it slowly: facial width correlates with body weight, at r equal to 0.378 in men and 0.291 in women in one 476-subject index study [3], and facial adiposity is readable from a photograph on its own [11]. Your skeleton moved it once, a long time ago. The growth trajectory that set your facial proportions runs through adolescence and is largely complete by your early twenties [10], and no routine reaches it now.

What Your Shape Changes: Hairstyles, Glasses, and Beards

Once you know your class, three styling routes use it as an input, each running on its own balance rule rather than a style list.

• Hair: balances the length of your face against its width.
• Frames: contrast your face's dominant line instead of repeating it.
• Beard shape: changes the jaw outline this detector just measured.

Face shape is one input among several. Hair texture, density, lifestyle and your own preference matter just as much, and this page does not pretend a class is a prescription. Glasses, beard styling and colour analysis all use the same class as an input; hairstyles chosen by face shape gets its own guide once it publishes, and until then the seven class guides above carry the route.

What Else the Same Scan Measures

The same landmark mesh that found your face shape runs the composite that scores every metric at once. Analyse your whole face reports each measurement with its own unit and its own tolerance, rather than folding face shape into a single number.

The namesake instrument on this site is the attractiveness test, which turns those measurements into the 1-10 score this page deliberately does not produce. This page does not re-explain face shape there, and it does not re-explain that score here.

Not medical advice. This is a measurement of a photo, not a diagnosis, a clinical assessment, or a verdict on you.