What's In A Face? Testing 15k faces

What's In A Face? Testing 15k faces

and other things

Aella

Oct 30, 2026

106

50

11

Share

What do more abused faces look like? It turns out you can tell by looking.

I made

StudyOfUs

, a site where you rate faces, submit your own face to get rated, and get a huge amount of data about the kinds of people you prefer and the kinds who prefer you. But this

also

left me with a huge dataset of people’s submitted selfies, and their answers to a ~70 question survey. So, I pulled the clearest, camera-facing photos and averaged some faces.

Before reading this post: if you want to take a quiz to see how accurate you are at guessing the traits,

do so here

.

Seem obvious to you? 83% of people

guessed

correctly, so it’s obvious to most people too.

Women who report being abused in childhood, also are more likely to tilt their faces down in photographs - and it shows up clearly in the composite.

Other traits that predicted likelihood of women tilting their faces downward - or holding cameras higher? - were being nonbinary, trans, or overweight:

Keep in mind that the faces are exaggerated 3-5x to make the differences more obvious. I asked my AI to search for published standards in exaggeration level for composite samples this big, but it said “Typical psychology composites use tens to hundreds of faces per group” and that there wasn’t really a precedent for this level of composite.

Some face differences seem to show up primarily via expression:

Look how happy the elite are!

The biggest smiling gaps are among men and trans women, with women having less of a smiling distinction between traits. The biggest smiling gap (which is measurable! not just me eyeballing the photos!) is men with more money, followed up by men with higher education and men having a more upper-class childhood.

Men who’ve done sex work, and who are anxious/depressed, and who have

low

BMI, also smile less:

I wondered if the sex work gap was money-confounded - like maybe poor people just smile less, and poorer men are more likely to be sex workers - but no, sex work men (n=246) smile less even holding wealth steady.

One of my favorite comparisons is height. I asked people how tall they were, and this is how their faces differed:

You can see it! It’s crazy how much you can see the tallness in the face itself. This was also one of the most accurately guessed ones - when I showed the (male) pairs to people and asked them to guess, 96% got it right.

People were also super good at guessing which people reported exercising more and being bad at dating:

People were

worst

at figuring out who believes in astrology:

The SOU survey asks about far more than the stuff I’m showing in this post. Out of 54 traits, 24 seemed reliable: I split each group in half at random, and the two halves had to show the same difference more consistently than in 99% of random shuffles.

BPD didn’t quite pass, but since people meme on it so much you can see it anyway:

Politics and narcissism were trickier for people to guess, but it’s still guessable:

And then we have a collection of sex-related stuff. In general, people who watch less porn, have sex with more people, and are more straight, look more conventionally attractive:

Kinkiness is interesting though - I could easily see the faces being a

type

, where they are different but neither is obviously more atractive than the other.

But what if we try to look at actual attractiveness? Here’s faces divided by if people rated themselves as more attractive than others:

And here it is divided by other people actually rating those people, or ‘true’ attractiveness (adjusted by BMI, since BMI correlates decently with lower perceived attractiveness:)

I did give people a fast big-5 survey, just 3 questions per spectrum. This does seem enough to see visible differences, although I had to pick the extremes of each end which is a bit noisier than our above faces:

Different faces in the personality split were less significant than most of the other traits above, and vary a lot; for example, men’s neuroticism my AI considered an uncertain difference, while it was pretty confident about women’s extroversion.

And lastly, let’s check age for fun:

Other pairs I didn’t include and that didn’t pass the significance threshhold yet are:

being spanked as a kid, if you had a sexually liberated vs repressed upbringing, your birth order, how old you were when you first had sex, if you’re submissive or dominant, if you’re monogamous or poly, if you’ve hired a sex worker, your religiousness, your attachment style, if you have kids, and your relationship status.

In general, 15% of men and 29% of women submitted a photo where their face was too obstructed to use - looking away from the camera, holding a hand up, etc.

Everyone else fell in between: 18% of enby amabs, 22% of trans men, 25% of trans women, and 25% of enby afabs had face obstruction or turning too much to be able to use their photo.

Among men, gayer ones used more obscured photos (18% vs 14% of straight men), more overweight men (17% with a BMI of 27 or over vs 13% under 22), and kinkier men (16% of kinkier men vs 13% of vanilla-ish men). Men who grew up higher-class obscured at 12%, vs 15% of men who grew up lower-class.

Interestingly, the difference holds primarily for men. Women mostly did not differ in how much they avoided the camera when split by trait.

Reminder - men were also more likely to be differentiated by smiling. This is painting a picture where choices you make in a photo - head angle, facial expression - seems to be more a meaningful reveal for men than it is for women. Maybe women are more laser-focused on what makes them hot, which is a unifying force across photo types, while men allow more genuine signals about themselves to leak through?

Other gaps are less gender-differentiated, though. Most of the asymmetrical sample below is just that the women’s sample was smaller, and so I could be sure about fewer of the differences compared to men’s:

I’m pretty surprised at how… cliche everything is? Richer, more sexually successful, mentally healthy people just… show it on their aggregate faces. Is it tans from spending time outside? Smiling because they have friends to catch them in the middle of an activity? Relaxed muscle tension from being so functional and cooler than the rest of us?

In some sense, it feels like nearly all the faces above are just two types of faces - the healthy and the unhealthy ones. Is there a general healthiness-unhealthiness factor we can check?

Turns out there is - if you compress all the stuff we’ve looked at, all the items correlate into a general ‘doing pretty good’ vs ‘doing pretty bad’ cluster. If we merge it all together, here are their faces:

Knowingless is a reader-supported publication. To receive new posts and support my work, consider becoming a free or paid subscriber.

106

50

11

Share

Previous