What Was Measured
Ollie's celebrity index holds 5,955 well-known living people with freely licensed photos from Wikimedia Commons. For each person, up to three photos were run through Google's MediaPipe Face Landmarker, which places 478 points on a face. Only straight-on photos were kept: head turned or tilted less than 10 degrees, and a face at least 100 pixels across at the cheekbones. 4,032 people had at least one photo that passed.
These are the same rules the face scan in Ollie Stylist uses on your camera, so the numbers below describe exactly what the tool measures.
How a Face Shape Is Calculated
Four widths are measured from the landmarks and divided by the distance between the cheekbones, so the result doesn't depend on photo size: face length (top of the forehead to chin), forehead width (between the temples), jaw width (lower jaw corners) and chin width (either side of the chin). Across 9,195 straight-on photos, the average face was 1.18 times as long as it was wide at the cheekbones, with the forehead at 0.93, the jaw at 0.90 and the chin at 0.58 of cheekbone width.
Each of the four ratios is turned into a z-score against those averages, then compared with seven prototypes, one per shape, written in the same z-score units. A round prototype, for example, is short with a slightly wide jaw and chin; a heart prototype has a wide forehead and a narrow jaw and chin. A Gaussian weight on the distance to each prototype becomes a probability, so every face gets a likelihood for all seven shapes.
One honest caveat: the prototypes were set by hand from how stylists describe each shape, not fitted to faces labelled by experts. They encode the standard definitions, and they can be refined if users report wrong results.
Read more: Face Shapes: The 7 Types and How to Tell Yours Apart
The Results: How Common Each Shape Is
Oval 22.0% (889 people), heart 18.0% (727), triangle 14.5% (586), round 12.2% (491), oblong 12.1% (486), square 11.9% (478) and diamond 9.3% (375).
Oval comes out most common partly by design: its prototype sits close to the average face, so faces near the middle land there. That's also why oval results are rarely confident. No celebrity's oval probability passed 50%, while 349 heart shaped faces did.
Get new guides by email
Men vs Women
Of the 4,032, 2,568 were men and 1,441 women (Wikidata gender; a few had none recorded). The shapes split very differently. Men: oval 21.2%, square 16.7%, triangle 15.7%, round 15.2%, oblong 13.8%, heart 12.9%, diamond 4.6%. Women: heart 26.9%, oval 23.6%, diamond 17.8%, triangle 12.6%, oblong 9.0%, round 7.0%, square 3.1%.
The pattern fits what's known about facial differences between the sexes: men's jaws are on average larger and squarer, which pushes faces toward square, round and triangle, while women's narrower jaws and chins push faces toward heart and diamond.
Read more: Celebrities With Heart Shaped Faces, Measured by AI
How Repeatable Is a Face Shape?
Not very, from a single photo. Comparing several photos of the same person, their ratios varied about two thirds as much as they varied between different people. Hair over the forehead, a slight head tilt, camera distance and expression all move the numbers. That's why the scan reports probabilities and scores haircuts against all shapes, and why celebrity examples on this blog use only faces the scan was confident about.
See the celebrities in each group: round, oval, heart and square.