The first way is to ask people. A photo goes in front of raters, they pick a number, and the average becomes the score. What this measures is real, but it is the raters' reaction, not anything about the face itself. Change the raters and the number moves.

The second way is to train a model on photos that people already rated, then let it predict a rating for a new photo. The number arrives fast and looks precise, but it inherits whatever the original raters preferred, and there is usually no way to see which part of your face pushed it up or down. Retrain the model and yesterday's score is gone.

The third way is to measure. Detect points on the face, compute distances and ratios between them, report those. Nothing is being predicted, so the same photo produces the same numbers today and next year, and each number points at a specific part of the face you can go look at.

This site takes the third path. That buys reproducibility and costs coverage: geometry can describe proportion, spacing and symmetry, and it cannot describe skin, expression, styling, or how you come across in motion. A test that claims to cover all of that from one still photo is doing the second thing and calling it the third.

There is a fourth source, and it is the one people notice first: two sites can measure identically and still hand you different tiers, because they drew their tier boundaries in different places.

How three well-known calculators split the same scores

The practical move is to stop comparing final numbers across tools and compare the underlying measurements instead. A ratio like eye spacing relative to face width means the same thing everywhere it is reported. A tier label does not.

These two numbers get searched constantly, and they come from rating communities rather than from any measurement. The habit started on a 0–8 scale, got rescaled to 10 for convenience, and the band labels travelled with it.

So when someone asks whether 6.5 is attractive, the honest answer is that 6.5 is a position on a convention, and the convention differs between the communities using it. The same face can be a 6.5 on one site and a 7 on another without either site making an arithmetic error — they drew their band lines in different places.

Which means the question "is 6.5 attractive" cannot be answered by this page or any other, because it is a question about where you stand among people, and what got measured was one photograph of one face. What this test hands back is the four proportions it read and the tier those proportions land in — and that tier mapping is provisional while it is still being checked against real photos.

It will not place you among other people. Ranking anyone requires a sample that stands in for the population you want to be ranked against, and what face-rating sites collect is people who decided to upload a selfie to a face-rating site. A number built on that sample describes the sample.

It will not tell you what to change, and it will not tell you whether changing it is safe. Reading pixels off one photograph puts a tool nowhere near the position a clinician occupies, so nothing here should be carried into a conversation about a procedure, an appliance or a diet.

It will not stay steady across bad photos. A steep camera angle or a strong side light changes the geometry that gets measured, which is why a rejected photo is a better outcome than a confidently wrong score.