A 478-point face mesh
The browser places facial landmarks on the selected photo before any ratios are calculated.
Upload one clear, front-facing photo to get a primary and secondary face-shape match from 478 measured points. No signup and no upload.
No signup. Your photo never leaves your device.
The guide is measured from this sample photo.
Generated sample · measured with this face tool
Every scan runs on your own device. We load the same 478-point face mesh used on our home page — 468 facial landmarks plus ten iris points — and place those points on the photo you pick. You can turn your head slightly; before measuring, we level the image along your eye line so a tilted selfie does not skew the result.
From that grid we take 16 pairs of points that sit symmetrically on the left and right of your face — at the brow, outer eye, cheekbone, jaw angle and chin. Each pair gives us a width ratio; each point gives us a vertical position ratio. That is 16 width ratios and 16 vertical ratios, 32 numbers in total, and they are the only input to the classification.
Those 32 numbers are standardised against the mean and standard deviation of a labelled training set, then ranked by a five-class linear classifier. You get a primary shape and a second-closest shape — nothing else.
We measure ratios, not centimetres. No number on this page tells you your face is 14 cm wide; widths are expressed relative to each other, which is also why the result does not change when you move closer to the camera.
The browser places facial landmarks on the selected photo before any ratios are calculated.
A small sideways tilt is levelled first, so a slightly tilted selfie can still be read consistently.
Sixteen mirrored widths and sixteen vertical positions feed the classifier.
Five of these seven shapes are ones this detector will name for you. Diamond and triangle are real, widely used styling terms, but we do not return them as automatic results — the chart tells you which is which, and the section below explains why.
| Shape | How to recognise it | Auto-detected | Flattering haircuts | Frames |
|---|---|---|---|---|
| Oval | Length a little greater than width; cheekbones widest | Detected | Most cuts work; avoid heavy fringe | Almost any frame |
| Heart | Forehead and cheekbones wider than jaw; chin tapers | Detected | Chin-length bob, side part | Bottom-heavy frames |
| Oblong | Noticeably longer than wide; widths stay similar | Detected | Layers and width at the sides | Tall / deep frames |
| Round | Length and width close; jaw soft, no hard angles | Detected | Height on top, volume off cheeks | Angular frames |
| Square | Forehead, cheekbones and jaw similar; jaw angular | Detected | Soft waves, side-swept fringe | Round / oval frames |
| Diamond | Cheekbones widest; forehead and jaw narrower | Reference only | Volume at the forehead | Frames wide at the brow |
| Triangle | Jaw wider than cheekbones | Reference only | Volume above the ears | Top-heavy frames |
For left-right differences within an outline, use the facial symmetry test.
People describe the same tool in a lot of different ways — face shape test, face shape quiz, face shape finder, face shape analyzer, face shape calculator and face shape identifier — and most of those searches land on the same result page.
They all mean roughly the same thing: hand the tool a photo and let it name your shape. A test or quiz usually implies questions; there are none here, because the measurement happens on the image rather than on your answers. A calculator usually implies numbers you type in — that is the manual method in the next section. Analyzer and detector describe the same process.
There is no separate page for each of those words and no second result. One photo, one classification, one page.
Four measurements and one angle, taken in front of a mirror with your hair pulled back. This is the manual calculator method, and it is separate from the detector’s landmark ratios.
Use the same units for every measurement, then compare ratios rather than chasing an absolute number. For example, a face length of 18 cm divided by a cheekbone width of 12 cm gives 1.50; if the forehead, cheekbone and jaw widths are all close to 12 cm, that combination reads as longer and fairly even in width, which is a useful first clue for an oblong face.
Measure from the centre of the visible hairline to the bottom of your chin, keeping the tape straight down the centre. Divide this by cheekbone width: for example, 18 cm ÷ 12 cm = 1.50, which suggests a noticeably longer face when the other widths are similar.
Measure across the widest part of the forehead from temple to temple, keeping hair pulled back. Compare it with cheekbone and jaw width: a forehead close to 12 cm in an 18 cm by 12 cm example supports an even, elongated outline, while a much wider forehead points toward a different shape.
Measure across the widest part of the cheekbones, usually below the outer corners of the eyes. Use this as the reference width for your ratios; in the example, 12 cm is the widest or near-widest middle-face measurement. Compare face length with cheek width in the golden ratio face calculator.
Measure the lower face from one jaw angle to the other, just below the ears, rather than measuring only one side. A jaw near the same 12 cm as the cheekbones keeps the outline relatively parallel; a much narrower jaw suggests taper, while a wider jaw suggests a stronger lower-face shape. The jawline rating test measures jaw width and angles separately.
Note how sharply each jaw corner turns toward the chin, and compare both sides in the same relaxed position. A softer angle can support a round or oval impression, while a clearer corner can support square; use it with the length and width ratios instead of treating one angle as decisive.
We could print a simple accuracy claim on this page. We do not, because we could not stand behind the number. The face rating page explains the separate limits of a photo score.
The classifier was fitted on a public set of 500 labelled photos — 100 per shape, five shapes. Six annotators passed a qualifying test before labelling, and only images where at least four of six agreed were kept. That is enough to fit a geometric model. It is not enough to claim a population accuracy figure: the set leans on female celebrity photos, contains repeated identities and has unclear original image licences. We use it to fit the model and nothing else — it is not redistributed with the product and does not appear on this site.
Diamond and triangle are the two shapes we cannot reliably return. The major public face datasets we reviewed do not carry a consistent popular seven-shape label; the seven-class paper we found does not publish its data, and its test split contains only five diamond and eleven triangle examples. Rather than hand-write thresholds and return a shape we cannot defend, we return five and describe seven.
If your proportions read closer to diamond or triangle, that is a real possibility this tool will not name — the geometry is there, the labelled data is not. Your result will tell you when that happens.
Heart: try chin-level volume and bottom-heavy frames. Oblong: add width at the sides and choose deeper frames. Oval: most cuts and frame shapes can work as a starting point. Round: add height or angular contrast if you want to lengthen the outline. Square: soft waves and round or oval frames can balance clear jaw corners.
Use these as styling ideas, not rules. Hair texture, density, hairline, fit, maintenance and personal taste still matter more than a label.
478-point geometry analysis runs in your browser. Your free photo is never uploaded.
The detector reads the picture you give it. Four small choices make the outline easier to measure and the result easier to repeat.
Keep your head level and look straight ahead. A small turn can make one cheek look narrower and shift the ratios that separate oval, round, square and oblong.
Leave at least an arm’s length between your face and the phone. A close wide-angle selfie can enlarge nearby features and make the outline look shorter or wider than it does at a normal viewing distance.
Use soft light from in front of you so both cheek edges and the jaw stay visible. Hard side light can turn a shadow boundary into a false face edge.
Move hair, hats and reflective glasses away from the forehead, cheeks and eyes when possible. On a phone, keep the lens near eye level, tap to focus and repeat the same setup for comparisons.
Quick answers before you test another photo.
Upload a clear front-facing photo. The detector compares face-length, width and taper ratios and returns a primary and secondary match from five automatically detected categories.
Yes. The scan is free, requires no signup and runs in your browser.
No. Your face landmarks and proportions are calculated on your device. The selected photo never leaves your browser and is cleared when you remove it or close the page.
It is a useful styling estimate, not an exact anatomical measurement or a population accuracy claim. The classifier was fitted on a consensus-labelled five-class reference set, and photo pose, hair, expression, lighting and lens distance can still change what it sees.
Diamond and triangle are useful styling terms, but the available labelled data does not support reliable automatic classes. We return five better-supported categories and explain the two reference-only shapes in the chart.
They are different words for roughly the same result page. A quiz usually suggests questions; this detector uses your photo and measured ratios instead.
It can still work, but covered or reflective features can move the detected points. For the clearest result, keep your eyes, cheeks, jaw and as much of the upper outline visible as possible.
Yes. It runs in a modern mobile browser. Use a front-facing photo with the camera near eye level and enough distance to avoid a close-up lens distortion.