How the face shape detector works
- 1
Add a front-facing photo
We check the angle, the light and that there is exactly one face before measuring anything. If the head is turned or tilted too far, the photo is rejected so the measurements stay reliable.
- 2
478 points, on your device
A face model of about 4 MB runs in your browser and finds points along the outline of your face, brows, eyes, nose and jaw. The photo is never uploaded and never leaves your tab.
- 3
Five distances, one shape
Face length and the forehead, jaw and chin widths are each divided by cheekbone width, then compared with six shape prototypes. You get the closest shape and how clear the call is: a clear match, the closest match with the next shape, or between two shapes.
What we measure
We take five distances from fixed landmark points on a 3D face mesh, so the result does not depend on where you would place a tape measure. Each distance is turned into a ratio so photo size and zoom do not matter.
Face length to width
DeviceFace length runs along the midline from the top of the face mesh (mid-forehead) to the bottom of the chin, divided by cheekbone width. Hair is not detected, so the top point sits on the upper forehead rather than the hairline. Measuring along the midline removes in-plane head roll from the height reading.
Cheekbone width
DeviceThe distance across the face between the two cheekbone landmarks (234 and 454). Face length and the forehead, jaw and chin widths are each divided by this number, so photo size and zoom do not skew the result.
Forehead to cheekbone width
DeviceThe width across the upper forehead (landmarks 54 and 284), divided by cheekbone width. A wide forehead paired with a narrower jaw and chin points toward a heart shape, while a narrow forehead points toward diamond.
Jaw to cheekbone width
DeviceThe width across the lower jaw (landmarks 172 and 397), divided by cheekbone width. A wide jaw points toward square, while a narrower jaw points toward heart or diamond.
Chin width
DeviceThe width across the lower chin (landmarks 176 and 400), divided by cheekbone width. It helps tell a broad chin, as in the square prototype, from a narrower one, as in heart and diamond.
How five distances become six face shapes
When people ask what their face shape is, they are usually trying to match their proportions to one of six styling categories: oval, round, square, heart, oblong or diamond. Styling guides use these categories as shorthand for where the face looks widest, where it tapers, and how long it looks relative to its width. Traditional guides ask you to measure with a flexible tape or trace your reflection, and the answer can change with where you place the tape on the forehead or jaw.
Our detector replaces the tape measure with fixed geometry. The MediaPipe Face Landmarker places 478 3-dimensional landmarks on your face in your browser tab. This kind of model infers an approximate 3D mesh of a face from a single camera image (Kartynnik and colleagues, 2019). From the mesh, the engine takes five straight-line distances: face length from the top of the mesh to the bottom of the chin, cheekbone width, forehead width, jaw width and chin width. Dividing face length, forehead width, jaw width and chin width by cheekbone width gives four scale-free ratios.1

Pink dots mark detected facial features; pink lines connect the measured widths, face length and eye corners.
Those four ratios are compared with six shape prototypes, each expressed against our mesh norms. Oval is slightly long with a slight taper. Round is short with a wider jaw. Square has a wide jaw and a broad chin. Heart has a wide forehead, a narrow jaw and a narrower chin. Oblong is long. Diamond has a narrow forehead, jaw and chin. The engine measures how far each of your ratios sits from the norms, compares that pattern with each prototype, and reports your closest shape with how clear the call is. Because the six shapes share one comparison, a typical face's top shape holds only about half of it, so we show words rather than a percentage that would read as a poor match.
Because the mesh is fitted to a canonical face, real faces span a narrow band of ratios on it. In the middle 80% of our calibration photos, face length divided by cheekbone width runs from 1.085 to 1.246. The mesh norm is 1.163: a typical face is about 1.16 times as long as it is wide. This is not the 1.0 to 1.8 range that tape-measure guides use, so compare your readout with the mesh range rather than with an older chart.
Why mesh ratios differ from tape-measure guides
If you have measured your face with a ruler and then tried an online detector, the two answers can differ. One reason is where the top of the face is defined. A tape measure may start at the hairline, and a hairline changes with hairstyle, bangs or hair loss, so the length reading changes with it.
Our tool measures face length from the top of the face outline, in the middle of the forehead, down to the bottom of the chin. It is not the hairline. Measuring along the midline removes in-plane head roll from the height. Because the landmarks are fixed points, the result is repeatable: the same photo always returns the same numbers.
Number of photos
- Median:
- 1.16
- Middle 80%:
- 1.085–1.246
Research on facial proportions also shows that real faces rarely match rigid textbook formulas. Farkas and colleagues (1985) tested nine neoclassical proportion canons in 153 young adults and reported large variability in facial feature size. They concluded that the canons do not represent average facial proportions. That is one reason we tell you how clear the call is, and name the next shape when it is close, rather than giving a single label.2
Our reference bands come from 133 real front-facing photos from Pexels. They provide a mesh reference, not population norms or a test of classification accuracy. On these photos, the results were round 34, oblong 23, heart 23, oval 22, diamond 17 and square 14. No shape accounted for more than about 26% of the set. So the result is a guide, not a precise classification. If your result says between two shapes, for example oval / oblong, your proportions sit near the border between them, and it may be worth reading the styling notes for both.
Reading forehead, jaw, and chin proportions
Face length to width tells you whether the face is more compact or more elongated. The forehead, jaw and chin ratios describe the outline. Each is a comparison with cheekbone width, so a jaw counts as wide or narrow relative to the cheekbones, not on its own.
In our engine, cheekbone width is the anchor (1.00) for the horizontal comparisons. Our mesh norms are 0.858 for forehead to cheekbone width, 0.802 for jaw to cheekbone width and 0.333 for chin width. When your forehead ratio is above 0.858 while your jaw and chin ratios are below their norms, the outline is widest at the top and tapers toward the chin, which points toward heart. When forehead, jaw and chin ratios are all below their norms, the cheekbones are the widest part, which points toward diamond.
Oval
Gently tapered jawRound
Relatively short faceSquare
Wide jaw and chinHeart
Broad forehead, narrow jawOblong
Relatively long faceDiamond
Cheeks wider than foreheadFace shape is only one cue among many in how people see a face. Little, Jones and DeBruine (2011) reviewed research on face preferences and listed several characteristics that influence judgements, including symmetry, shape cues, averageness and skin colour or texture. They also reviewed individual differences such as visual experience and familiarity. In everyday use, face shape is best treated as a practical aid for choosing styles, not as a rulebook.3
Common styling advice uses hair, brows, glasses and grooming to balance the outline rather than change it. For example, people with a broad jaw often try softer lines around the temples or rounded frames, and people with a wider forehead and narrower chin often look for volume near the jaw or frames with more weight at the bottom. These are conventions rather than rules, so use them as starting points.
How camera distance and head tilt affect your result
Even with fixed landmarks, a single 2D photo carries perspective distortion from the lens and from the angle of your head. A camera held very close to the face can change the apparent proportions of the face, so a photo from a little further away is a better starting point.
Before any ratio is calculated, a quality check rejects photos with no face or more than one face, a face that is too small, a photo that is too dark, and a head turned (yaw) or tilted (pitch) too far. Turning or tilting the head changes how much of the jaw, forehead and cheekbones the camera sees, which can shift the measured widths, so keep the head straight.
Front and level
Turned or looking up/down
Too dark
More than one face
MediaPipe Face Landmarker accepts still images, decoded video frames or a live video feed, and the approach behind it runs at high speed on mobile GPUs (Kartynnik and colleagues, 2019). Speed does not remove perspective error from the photo itself, so step back and keep the lens at eye level before you take the picture.1
For the most reliable reading, pull your hair back off your forehead and temples, tuck it behind your ears so the jaw is visible, and keep your expression relaxed. Once you have a result, use the lines drawn on your photo to check that the top point sits on your upper forehead and the side markers sit on your cheekbones, jaw and chin.
The six face shapes
Most faces sit between two shapes, which is why we always show a runner-up. Treat the shapes as styling shorthand, not as boxes.
| Shape | How it measures | What people usually look for |
|---|---|---|
| Oval | Slightly long, with a slight taper from forehead to chin. | Keeping the balance the face already has. |
| Round | Shorter face with a wider jaw and a soft outline. | Styles and frames that add some height or angles. |
| Square | Wide jaw and a broad chin. | Softer lines around the jaw, or keeping the strong jaw as the feature. |
| Heart | Wide forehead, narrow jaw and a narrower chin. | Volume near the jaw and chin to balance a wider forehead. |
| Oblong | Clearly longer than wide. | Styles that add width at the sides instead of more height. |
| Diamond | Narrow forehead, jaw and chin compared with the cheekbones. | Softening the cheekbones, or adding width at the forehead and chin. |
What this detector can't tell you
- Face shape is a styling convention, and different guides draw the lines between shapes in different places.
- A beard along the jaw, hair covering the forehead or temples, or a tilted head can change the outline we measure.
- Close-up photos can distort proportions through perspective. A photo taken from a little further away is a better starting point.
- Mesh ratios sit in a narrower range than tape-measure guides (1.085 to 1.246 for length to width in the middle 80% of our calibration photos). Our reference bands come from 133 real front-facing photos, not a representative population sample; no shape accounts for more than about 26% of these photos.
- The label says how close you are to the six shapes in this photo. It is not a measure of attractiveness, health, or personality.
Tips for an accurate result
Face the camera
Keep the lens at eye level, your head straight and your chin level, so the forehead and jaw are not seen at an angle.
Hair back
Pull hair away from your forehead, temples and ears so the outline and jaw are clearly visible.
Even light
Face soft daylight from a window, without harsh shadows hiding one side of your jaw or cheekbones.
Neutral expression
Relax your mouth and cheeks. A wide smile can change the shape of the mid-face.
Questions
References
- Kartynnik, Y., Ablavatski, A., Grishchenko, I., & Grundmann, M. (2019). Real-time facial surface geometry from monocular video on mobile GPUs. arXiv:1907.06724. arxiv.org/abs/1907.06724
- Farkas, L. G., Hreczko, T. A., Kolar, J. C., & Munro, I. R. (1985). Vertical and horizontal proportions of the face in young adult North American Caucasians: Revision of neoclassical canons. Plastic and Reconstructive Surgery, 75(3), 328–337. doi.org/10.1097/00006534-198503000-00005
- Little, A. C., Jones, B. C., & DeBruine, L. M. (2011). Facial attractiveness: Evolutionary based research. Philosophical Transactions of the Royal Society B, 366(1571), 1638–1659. doi.org/10.1098/rstb.2010.0404