A facial symmetry test that tells you what it measured.
Most symmetry tests mirror one half of your face and leave you to stare at the result. Canthion measures it instead. One straight-on photo, 478 landmarks, a midline fitted through the centre of the face, and every paired point on the left checked against its mirror on the right. You get three things back: a value, where that value sits against healthy adults, and how much of it the detector itself would have produced on a face that was perfectly symmetric. Nothing is uploaded.
How the test works
- A midline is fitted, not assumed. Twenty-three landmarks sit on the centre of a face: the top of the forehead, the bridge of the nose, the tip, the base, the centre of both lips, the chin. A line is fitted through all of them. A slightly tilted head therefore does not, on its own, produce asymmetry.
- Every paired landmark is reflected across it. The outer corner of the left eye is mirrored across the midline and the distance to the outer corner of the right eye is measured. The same for every left–right pair the mesh provides.
- The distances are averaged. The mean, as a percentage of the distance between your pupils so that photo scale does not matter, is your symmetry value. Lower means more symmetric.
- The noise floor is printed beside it. The landmark detector has a published accuracy of about 4% of inter-ocular distance per point. Run on a perfectly symmetric face it would still produce a non-zero value. That value is shown next to yours, and the gap between them is what you read.
The measurement is deterministic. The same photo gives the same value every time, and you can verify that by hash. A different photo of the same face, taken with the head turned a few degrees, will not.
What a number means, and what it does not
Healthy adults measured this way with 2D landmarks typically land around 3.2% of interpupillary distance, with a standard deviation of about 1.2. Your value is placed against that range and reported with its error bar.
Below the range means the face is highly symmetric at the resolution the detector can see. Inside the range is typical, and that is where most people are. Above it is above-typical asymmetry at this resolution. Before you read anything into a high value, two cautions.
First, head yaw. Turning the head even a few degrees moves every left–right pair by more than most real asymmetry does. A value that came out high on a photo where you were not quite square to the camera is measuring the turn, not the face. Retake it square and the number will drop.
Second, the noise floor. Much of any symmetry score taken from a photograph is landmark jitter. So the floor is printed, and a value sitting close to it is described as "at the detector's limit" rather than as a finding about you. A test that hides this is reporting its own error as your asymmetry.
Is my face symmetrical?
No, and neither is anyone's. Mirrored-half composites, the kind most symmetry apps generate, look uncanny for exactly that reason: real faces are never bilaterally identical. So the question worth asking is not whether, but how much, and whether that much is visible to anybody else. A value inside the typical range is not a flaw to fix.
Where asymmetry is clearly above typical and holds up across several square photographs, it is usually located rather than spread across the whole face. One brow sitting higher. One corner of the mouth. One eye more open than the other.
The full report measures each of those on its own (brow position, mouth corner tilt, eye openness), so you can see where your value is coming from instead of taking one number for the whole face.
Does it matter?
In the research, the effect of symmetry on rated attractiveness is real and small. Skin, grooming, hair and body-fat level all dwarf it. Symmetry is measured here because people ask for it, and because a number beats a mirrored composite.
It is not a lever. Nothing non-surgical changes bilateral bone symmetry, and the report says that rather than selling you an exercise for it. What can visibly balance a face that reads uneven is smaller than that: brow grooming, a haircut with a deliberate side, and for men the way a beard is shaped. Those are what the plan recommends when the numbers justify them.
How to take the photo
- Square to the camera. This is the one that matters most for this test: even a few degrees of turn inflates the value.
- Camera at eye height, at least an arm's length away.
- Even light from the front. A side shadow moves where the outline is found on one side only.
- Neutral expression, mouth closed. A half-smile is asymmetric in nearly everyone.
Then run the free scan. Symmetry is computed on the free scan and shown in full, with its noise floor and its place against the typical range, in the complete report.
Common questions
How is facial symmetry measured?
A midline is fitted through the centre landmarks, each paired landmark is reflected across it, and the mean distance to its partner, as a percentage of interpupillary distance, is the value. Lower is more symmetric.
What is a normal facial symmetry score?
Around 3.2% of interpupillary distance for healthy adults measured this way, with a standard deviation of about 1.2.
Is my face symmetrical?
Not perfectly, and that is normal. The test tells you how much, against the typical range, and how much of that is the detector's own noise.
Why does the test show a noise floor?
Because landmark jitter and head turn account for much of any photo-based symmetry score. The floor is what the detector would produce on a perfectly symmetric face, shown so the gap between it and your value is what you read.
Does symmetry make a face more attractive?
A little, in the research. Far less than skin, grooming, hair and leanness. It is measured because people ask, not because it is a lever.
Does the test upload my photo?
No. The test runs in your browser and the free scan uploads nothing.
One photo. Nothing uploaded. The free scan needs no account.
Related: how the asymmetry measurement is defined · what is my face shape · the attractiveness test · the honest guide to looksmaxxing.