What Is My Face Shape? Free Face Shape Detector with Open Rules

What your result shows

If you are asking what is my face shape, the honest answer is a measurement plus a margin. A face-landmark model places 478 points on your photo inside your browser, the tool reads four numbers from them, and the result card reports the following.

A separate confidence value (also with one decimal) describes the photo, not the shape: it combines head pose, light and face size, and is multiplied by 0.95 because the top of the face is always an estimate. If the head is turned beyond 10° or tilted too far up or down, the card asks for a retake and withholds the shape, because a turned head changes width ratios more than real differences between faces do.

What is measured, and where

Every length is divided by the distance between your pupils (landmarks 468 and 473), so photo size does not matter. The eye line is levelled first, and a small head turn is undone with the model’s depth estimate.

Face width itself is the span between landmarks 234 and 454, which is why “widest at the cheekbones” is the reference for every shape below. A front photo cannot show profile features such as chin projection; those are listed as Not scored — needs a side photo rather than guessed.

The seven-shape decision table

Each shape is a prototype: a set of four target values. The tool finds the prototype closest to your four measurements, using tolerances of 0.06 for length ÷ width, 0.03 for forehead and jaw ratios, and 5° for the jaw angle. The last column shows a quick mirror check you can do without any software.

The jawline words follow the front-view jaw categories: angular below 128°, balanced from 128° to 136° and soft above 136°. Oval is centred on MediaPipe’s average face model, so an average face lands near it by design.

Be clear about what these numbers are. No peer-reviewed standard taxonomy of face shapes exists; oval, round, square, oblong, heart, diamond and triangle are a styling convention. The prototype values above are engineering estimates awaiting calibration against real photos, not measured population norms. The anthropometric literature gives the underlying measurements (face width, face height, mandible width), not these labels.

Why different face shape tools give different answers

Run the same photo through three tools and you may get three shapes. Nothing is necessarily broken; the tools are measuring slightly different things.

Where face length starts

This is the biggest source of disagreement. Length can begin at the nasion (the nose root), at the hairline, or at the upper forehead. Each start point gives a different length ÷ width, and that single ratio separates oblong from oval and round from square. Our tool uses the upper forehead, landmark 10, because the hairline is invisible to the model and varies with haircut, recession and styling. The value is an estimate, and we mark it as one.

Thresholds nobody publishes

Most tools do not say how long is “long” or how wide is “wide.” Without published cut-offs, two tools cannot be compared and you cannot tell whether a result is a close call. We publish all four prototype values and the tolerances above so anyone can check the arithmetic.

One label versus probabilities

A single label hides how close the second choice was. A face at 1.12 length ÷ width could fall on either side of a line drawn at 1.10 or 1.15. That is why we show a runner-up and a probability for all seven shapes, plus a confidence value for the photo itself.

Photo angle, lens distance and hair

A camera held low lengthens the lower face; a phone held close widens the centre of the face relative to the edges; a slight turn narrows one side. Hair across the forehead or jaw edge moves the outline points the width ratios depend on. Two photos of the same person taken a minute apart can therefore classify differently, and the photo gate and confidence value exist to flag exactly that.

How to read your result

Read the top two shapes and the gap between them. A large gap means one prototype clearly fits best. A small gap means you sit on a boundary, and either label, or a blend of the two, is accurate. The match percentage expresses closeness among the seven prototypes, so it can be high while the photo confidence is low; check both.

Then look at the feature table. If your forehead ratio is near 0.80 and your jaw ratio near 0.80, no single area dominates, which is why several shapes can score within a few points of each other. The shape name is a summary of those four numbers, and the numbers are the more precise description. None of the shapes is better than another; they are descriptions of outline.

If you need a second opinion, the jawline analysis goes deeper on jaw width and the projected angle that feed this table, and the golden ratio calculator shows the lower-face proportions separately.

Getting a reliable result

For the whole picture, the overall proportion score combines symmetry, thirds, fifths and golden ratio, and the mirror-symmetry analysis shows which features differ left to right.