Face shape: how AI detects yours

Face shape is a categorical description of the outline of the head seen from the front, defined by the ratios between face length, forehead width, cheekbone width, and jaw width, plus the angularity of the jawline. The six classic categories are oval, round, square, heart, diamond, and oblong — and most real faces are blends of two or more.

The six classic face shapes

Face-shape classification comes from hairdressing and portraiture tradition rather than from anatomy textbooks, but the categories map onto measurable proportions. The six most widely used are:

Oval

Face length is noticeably greater than cheekbone width — a ratio around 1.4 to 1.5 is commonly cited — with cheekbones as the widest point and a jaw that tapers gently to a rounded chin. The forehead is slightly wider than the jaw. Oval is often treated as the reference category because its proportions sit near the population average.

Round

Face length and cheekbone width are close to equal, with a length-to-width ratio near 1.0 to 1.2. The widest point is at the cheeks, the jawline is soft and curved rather than angular, and the chin is rounded. Round faces read as shorter and fuller even at the same body weight as other shapes.

Square

Forehead, cheekbone, and jaw widths are all similar, and face length is close to face width. The defining feature is the jaw: a wide, angular gonial region with a flat or squared chin. The outline approximates a rectangle with nearly vertical sides.

Heart

The forehead is the widest region, the cheekbones are slightly narrower, and the face tapers strongly to a narrow, sometimes pointed chin. A widow's-peak hairline is often mentioned alongside this shape, though it is not required — the width taper from top to bottom is what defines it.

Diamond

The cheekbones are clearly the widest point while both the forehead and the jaw are narrow, producing a rhombus-like outline. Face length is usually greater than width. Diamond is the least common assignment in most classification surveys.

Oblong

Sometimes called rectangular: forehead, cheekbones, and jaw are similar in width — as in a square face — but face length is markedly greater, commonly cited as more than about 1.5 times the width. The face reads as long and narrow with relatively straight sides.

Defining proportions at a glance

These thresholds are approximations, not standards — different classification schemes draw the lines in slightly different places, which is one reason two quizzes can give you two different answers for the same face.

How AI classifies face shape, step by step

An AI face shape detector replaces the tape measure with landmark geometry. Refrakt's pipeline works roughly like this:

1. Landmark detection. A neural network maps hundreds of points onto the face, including the full jaw contour from ear to ear, the hairline region, the zygomatic (cheekbone) area, and the chin.
2. Pose normalization. The head is rotated to a level, front-facing reference so that tilt or a turned head doesn't shorten or widen the apparent outline.
3. Outline extraction. The landmarks along the facial border are connected into a continuous contour — effectively the same outline you would trace on a mirror.
4. Ratio computation. The system measures face length (trichion or hairline estimate to menton, the lowest chin point), bizygomatic width (cheekbone to cheekbone), bigonial width (jaw angle to jaw angle), and forehead width, then computes the ratios between them.
5. Jaw-angle measurement. The gonial angle — how sharply the jaw turns from the vertical ramus toward the chin — separates angular shapes (square, oblong) from curved ones (round, oval).
6. Classification. The ratio set is compared against the definitions above. Instead of forcing one label, a well-designed system reports proximity to each category — for example, 60% oval, 30% square — because that is what the geometry actually shows.

Why most faces are blends

The six categories are convenient bins drawn over continuous measurements. Face length, cheekbone width, and jaw angle each vary smoothly across the population, so most faces land between prototypes rather than on one: a face can have oval proportions with a square jawline, or round width ratios with a tapered chin. Surveys that ask multiple raters to classify the same faces regularly produce disagreement for exactly this reason.

That is why a percentage-based output is more honest than a single label. If your face is 55% oval and 40% oblong, both descriptions are partially true — and picking one over the other is a rounding decision, not a discovery.

What face shape is actually useful for

Face shape carries no information about health, character, or attractiveness — highly rated faces exist in every category. Where the label earns its keep is as a framing reference for choices that interact with the facial outline:

• Hairstyles. Stylists use shape to decide where volume and length balance the outline — for example, adding height for rounder faces or softness around an angular jaw. These are aesthetic conventions, not rules.
• Glasses. Frame-fitting guides typically contrast frame geometry with face geometry: angular frames on softer outlines, rounded frames on angular ones, and frame width matched to face width.
• Beard and makeup framing. Contouring and beard lines are often planned relative to the jaw and cheekbone widths that define shape.

What face shape is not useful for is ranking. Systems that assign worth or "tier" to a shape are reading social preference into a neutral geometric description. If you want measurements with meaning, look at symmetry and proportion — and even those describe variation, not value.

Frequently asked questions