An attractiveness test that shows its working.
Most face-rating tests hand you a number out of ten and hide how they got there. Upload the same photo twice and many of them give you a different number. Canthion measures instead. It reads 24 facial proportions and your skin from one photo, compares each against published norms for your sex and reference group, and shows the working: the value, its error bar, and where it sits. Same photo, same result, every time. Nothing is uploaded.
What you get
The free overview shows eight of the reads in full, no account needed: canthal tilt, facial thirds, eye spacing, eye openness, nose width, mouth width, and two skin reads, redness and dark circles. Every other proportion and skin metric is computed in the same run — fifths, nasal index, jaw width, symmetry, texture, spots, shine, evenness — and shown in full when the report is unlocked.
The complete report adds your score out of 100 and the four parts behind it, your face shape and the masculinity to femininity axis, mock-ups of the recommended grooming drawn on your own photo, and a plan graded by the strength of the evidence.
How the test works, step by step
- Landmarks. Your browser finds 478 points on the face in the photo — the corners of the eyes, the wings of the nose, the edge of the lips, the line of the jaw. This runs in WebAssembly on your device; the photo is never sent anywhere.
- Proportions. From those points it takes 24 ratios and angles used in the anthropometric and facial-aesthetics literature. A ratio, not a distance, because a photograph has no scale: it cannot tell you your jaw is 11 cm wide, but it can tell you exactly how wide your jaw is relative to your cheekbones. Why millimetres from a photo are a fiction →
- Comparison. Each value is placed against the published mean and spread for your sex and the reference group you choose, and given a percentile and a grade from A to F. The grade measures distance from the optimum for that group, where the rating literature has found one, not distance from some universal ideal.
- Error bar. Every value carries the uncertainty of the landmarks it came from. A reading of 0.71 ± 0.04 is honest about what one photo can support; a bare 0.71 is not.
- Skin. The same photograph is read for redness, texture, blemishes, periorbital darkness, specular shine and uniformity of tone — the six skin qualities that move how a face is rated and that respond to a routine.
- Score. The complete report combines the parts into a number out of 100: harmony 40%, skin 30%, leanness and definition 15%, symmetry 15%. The weights are printed on the report because a score whose construction is hidden is not a measurement.
The honest part first
No test can tell you how one particular person will see you.
What a test can do is measure your face against a population and show where each proportion sits: which are strong, which are ordinary, which pull the whole read down. That is more useful than a single number pretending to be objective beauty, because it tells you what to work on.
It also means the score is not the useful part. Two people can land on the same number for opposite reasons, one with excellent proportions and neglected skin, the other the reverse, and the plan for each is different. Read the grades.
What a score of 60, 70 or 80 actually means
The score is a weighted average of graded parts, so it reads the way the grades do.
- 40%Harmony
- 30%SkinRedness, texture, blemishes, periorbital darkness, specular shine and uniformity of tone.
- 15%Leanness and definition
- 15%Symmetry
Around the middle of the range, most proportions sit within a standard deviation of the mean for your group and the skin readings are unremarkable. That describes most faces, because that is what a mean is. Higher scores come from proportions clustering near the optima and from clean skin. Lower ones almost always trace to two or three specific readings rather than to everything at once, and the report names them.
A 65 with skin graded D and everything else graded B is a face that will look different in three months. The same 65 built out of bone-level proportions graded C is a face that is fine and should stop measuring.
Scores from other tests do not compare to this one, or to each other. Each app weights different things, most do not publish the weights, and many will not give the same photo the same number twice. A 7.2 there and a 68 here are not one measurement in two units.
What a photo can and cannot rate
A front photograph reliably supports the proportions listed above. What it does not contain is bone angles: the gonial angle of the jaw, chin projection, the orbital vector, the E-line. Those are three-dimensional and the front view flattens them, so a test that quotes them off a selfie is quoting an estimate under a bony name.
Canthion measures what the front view supports. When a reading needs the profile it asks for the profile, and reads the chin and the jaw angle off the traced edge of it, the way the clinical literature does. The full list of what is not measured, and why →
Taking the photo so the rating is fair
Most of the disagreement between face-rating tools is not the tool. It is the photograph.
Light the face evenly from the front, so no shadow reads as a hollow. Put the camera at eye height and at least an arm's length back, because a phone held close widens the nose and narrows the jaw. Neutral expression, hair off the forehead, no glasses unless you always wear them. Shoot the same way when you re-test, and any change in the numbers is a change in your face.
Why the score is reproducible
Every value is computed the same way from the same 478 landmarks, so one photo produces the same numbers on every run, and you can check that by hash. If a rating moves when you upload one picture twice, it was not measuring the picture. The comparison page sets this against the rating apps and the consultation reports, and every formula is on the methodology page.
Common questions
How attractive am I — can a test really tell me?
Not the way the apps imply. What it can do is measure your proportions and your skin against norms, and show where each one sits.
Is it free?
The overview is free and needs no account. The complete report is a one-time $59, not a subscription.
Does it upload my photo?
No. The measurements run in your browser and the free scan uploads nothing. Open the network tab during a scan and watch.
Why does my score change on other face-rating apps?
Because many of them sample or guess, so the same photo lands somewhere different each time. Canthion returns the same result for the same photo, because every value is computed from the landmarks the same way on every run.
Is it accurate?
Each proportion is as accurate as the landmarks under it, and the error bar tells you how accurate that is on your photo. The comparison uses published population norms and names its source. What it will not do is claim accuracy for things a photograph cannot measure. Those are listed, with the reason, rather than estimated.
Does it work for women and men, and for every ethnicity?
Yes. Norms differ by sex and by population, so you pick the reference group you want to be compared against and every value is placed against that group's mean and spread. The same face read against two groups gives two different answers, and the report says which one you are looking at.
What should I do with the result?
Find the two or three lowest grades. Sort them into the ones a routine can move, which are skin, hair, brows, facial hair, glasses and leanness, and the ones it cannot. Work the first list in order of evidence. The complete report grades each recommendation by the research behind it and shows the result on your own photo. Then re-test the same way in eight weeks.
One photo. Nothing uploaded. The overview costs nothing and needs no account.
Related: looksmaxxing, the honest version · what is my face shape · the facial symmetry test · how Canthion compares to Qoves and the rating apps · what a photograph cannot measure.