Attractiveness
Test

It locates 478 facial landmarks in your browser, fits a midline through the 28 that sit on it, then measures four geometric components: bilateral symmetry at 35%, facial fifths at 25%, vertical balance at 25% and golden-ratio proximity at 15%. Those four distances are combined as a weighted root-mean-square, and your score is where that combined distance falls among all scans. Nothing is sent to a server, and only the resulting numbers are encoded into your result link.

It is reproducible, which is a narrower claim. Feed it the same photo and you get the same score every time, and the formula is printed on this page so you can check the arithmetic yourself. What no attractiveness test can do is predict how a particular person will react to your face: expression, voice, movement and familiarity all matter more in a room, and none of them survive in a still photo.

Exactly 50, because the score is a percentile by construction rather than a mark out of a hundred. That is also why we draw the whole distribution instead of handing you a bare number: the middle is genuinely crowded, and a few points either side of 50 separate faces that are geometrically almost identical. The 1-to-10 scale most attractiveness test sites use has drifted badly, since people call each other a 7 out of politeness, so a mathematically average 5 now reads as an insult. Percentiles do not have that problem.

No. The whole attractiveness test runs in your browser and the image never leaves the page, so there is no upload endpoint and no stored copy to delete. The only thing that travels is your score, packed into a seven-character result link. Even the share card is drawn on your own device and saved straight to your downloads — ticking “include my photo” changes what that local file looks like, not where it goes.

Because the measurement describes the photo, not your skull. Head rotation is by far the largest factor: six degrees off axis fabricates more asymmetry than most faces actually have, which is why past nine degrees this attractiveness test refuses to give a number at all. A raised or tucked chin is next, and it is the more dangerous of the two because it shifts vertical balance in one predictable direction rather than adding noise. That is what the photo grade at the top of every result is for. Compare two scores only when both photos graded the same.

The reachable number we show you is built from two things and no more: squaring up to the camera, and levelling your head. Both are measured in degrees on your own photo, so you can verify the claim by retaking it. Skeletal asymmetry is listed separately as plus or minus zero rather than folded into the total, because selling back a number that adult bone cannot reach is the cheapest trick in this category. No facial exercise moves bone, and any attractiveness test implying otherwise is selling something.

Yes, and that is the default. Every attractiveness test card carries your tier, your score and your four components but no photograph, and it is drawn on a canvas in your own browser before you ever see it. Your face appears only if you tick the box yourself, and even then the file stays on your device. The link you send out holds seven characters of encoded score and nothing else.

One specific tradition. The proportion targets come from Western anthropometry and cosmetic-surgery literature, which carry documented Eurocentric bias, so faces from other populations can land lower without being less attractive to anybody. That is why the bottom tier is called Unconventional rather than something worse: the number describes distance from a canon, and the canon is the thing with the limitation.