In using Stable Diffusion for a DnD related project, I've found that it's actually weirdly hard to get it to generate people (of either sex) that aren't attractive - I wonder if it's a bias in the training materials, or a deliberate bias introduced into the models because most people want attractive people in their AI pics
It's trained on professionally taken photos. Professional photographers tend to prefer taking photos of attractive subjects.
Yeah all of my professional photographers hate me.
That's true, but it's not like ugly people don't get photographed - ultimately a professional photographer is going to take photos of whoever pays them to do so. That explanation accounts for part of the bias I think, but not all of it
If I would get pictures taken by a photographer I would not allow them to be used as training data. I don't even like looking into a mirror. Maybe that's part of why there are less ugly people pictures to train with.
I would guess that ugly people are less likely to commission photos.
They were created via a prompt, that prompt probably included some tags to make them more attractive. It's often standard practice to put tags like "ugly" and "deformed" into the negative prompts just to keep the hands and facial features from going wonky.
There are no elderly women, no female toddlers, and so forth either. Presumably just not what whoever generated this was going for. You can get those from many AI models if you want them.
Battleship coordinates, (B10). Also (I4) looks a lot like my niece. I really think it depends on your definition of "average" though. But as @fubo indicated. There are 0% black people in this photo. There's some vaguely Asian, roughly Middle Eastern looking, sort of South American, and whatever that is going on in (M8). But there are distinctly zero black people pictured.
Now I'm curious. What's average looking to you if none of these aren't?
These people are (almost) all young, attractive and have flawless skin
Thats what pops into your head first?
It's one thing that strikes me as kinda odd.
But then, the other day I was messing around with an image generation model and it took me way too long to realize that it was only generating East Asian-looking faces unless explicitly instructed not to.
Every model is going to have something as its "average case." If you want the model to generate something else you'll have to ask it.
Models generally trend towards one thing. Its hard to create a generalized model, from a mathematical standpoint. You just have to say what you want.
Yep, but you can't just say "diversity plz".
Matrix prompt with | white | black | brown | asian
Wow, you sure that's AI? That's really impressive!
wow cant even tell the difference.
Can you imagine this technology being used to impersonate a president? It would be bedlam, but fortunately there are no girl presidents!
I believe they are already doing that to make fake nude leaks for celebrities. It'll be scary.
Impressive and a bit horrifying. Still, it seems to have a big problem recreating hands. A dating profile with a set of profile pics that doesn't feature a hand will now become suspect... until it cracks hands, too.
Plenty of models can do hands just fine
This AI has a type
That is some how cool and creepy all at once. I guess it gives a macro-view of what we consider attractive.
Not really. OP doesn't mention how they were generated, and how the pictures were selected.
It could just reflect the kinds of photos that went into the image generator, and the biases that were in its training data, unless the photos were chosen based on an attractiveness survey or some such.
Most of these look like the same person with different hair and makeup, so it could just be a very specialized network.
It gives a view of what whoever generated these images considers attractive.
A16, C7, still not quite there with hands. đŦ
L15 also, wtf is going on there
I can't decide if B8 or L15 is my favourite.
She has only one buttock đ¤Ŗ
The one right below it has 6 fingers also
We all have trouble drawing hands hahahah.
Truth.
Shh they'll find us
L15 has a unibutt
And sadly her friend is looking at it and laughing. Some friend.
Finally, some positive representation in media for people with diverse bodies / disability! /s
There's a ton of fucked up hands in there. It also seems to struggle with handbags. It's kinda fun to try to find one that isn't flawed in some way, I haven't found one that isn't.
The hands are the giveaway in an otherwise convincing image almost every time.
Check out the fand on A7
real females remixed
Most real women don't look like twenty year old thin white models.
But clearly that was the dataset used here.
A lot of them look like they're 16
There are repeats too
The cop probably has pointy elbows.
This is wild.
Is there a source or article to read or something?
At least 15 of those women are mila kunis
There is that one brunette on the right, reminds me of a movie/video I saw where the same AI woman kept popping up.
Tell me they didn't allow American source material without telling me they didn't allow American source material.
Oh, is that why none of them are black?
Or even average looking.
In using Stable Diffusion for a DnD related project, I've found that it's actually weirdly hard to get it to generate people (of either sex) that aren't attractive - I wonder if it's a bias in the training materials, or a deliberate bias introduced into the models because most people want attractive people in their AI pics
It's trained on professionally taken photos. Professional photographers tend to prefer taking photos of attractive subjects.
Yeah all of my professional photographers hate me.
That's true, but it's not like ugly people don't get photographed - ultimately a professional photographer is going to take photos of whoever pays them to do so. That explanation accounts for part of the bias I think, but not all of it
If I would get pictures taken by a photographer I would not allow them to be used as training data. I don't even like looking into a mirror. Maybe that's part of why there are less ugly people pictures to train with.
I would guess that ugly people are less likely to commission photos.
They were created via a prompt, that prompt probably included some tags to make them more attractive. It's often standard practice to put tags like "ugly" and "deformed" into the negative prompts just to keep the hands and facial features from going wonky.
There are no elderly women, no female toddlers, and so forth either. Presumably just not what whoever generated this was going for. You can get those from many AI models if you want them.
Battleship coordinates, (B10). Also (I4) looks a lot like my niece. I really think it depends on your definition of "average" though. But as @fubo indicated. There are 0% black people in this photo. There's some vaguely Asian, roughly Middle Eastern looking, sort of South American, and whatever that is going on in (M8). But there are distinctly zero black people pictured.
Now I'm curious. What's average looking to you if none of these aren't?
Older and fatter for a start.
Are you american?
What country do you live in btw?
France, and 75% of people are not fatter
I wouldn't be so confident buddy. 25.3% of French people are obese. Not that lower then America which is at 28.8%. You ain't much better.
No way
These people are (almost) all young, attractive and have flawless skin
Thats what pops into your head first?
It's one thing that strikes me as kinda odd.
But then, the other day I was messing around with an image generation model and it took me way too long to realize that it was only generating East Asian-looking faces unless explicitly instructed not to.
Every model is going to have something as its "average case." If you want the model to generate something else you'll have to ask it.
Models generally trend towards one thing. Its hard to create a generalized model, from a mathematical standpoint. You just have to say what you want.
Yep, but you can't just say "diversity plz".
Matrix prompt with | white | black | brown | asian
"Not just these women, but all pictures of all women everywhere. Women are a myth!"
Just like birds.
That's a lot of white chicks.
Really it's like 6, copy pasted over and over
Yeah, a lot of the faces look very very similar!
So?
Points to a limited dataset including mostly white people
Without knowing the prompt vs the model, it's impossible to say which side of things is responsible for the lack of variety.
Many modern models are actually very good at reversing sample biases in the training set.
Good point
The astronaut one is funny
casually takes off pressurized helmet on the moon
Where is it? I looked everywhere, couldn't find it.
6th row from the bottom towards the middle, left of the cop
They mostly all look like that one girl from that show, whatever it was.
Mila Kunis?
My god, soon they will be making them nsfw!
They what!? Oh my
Oh, they definitely are already
For example, NSFW obviously:
https://lemmynsfw.com/post/518229
Even the genitals are amazingly accurate (nsfw) - https://imgur.io/gallery/H5nPH
Wow, you sure that's AI? That's really impressive!
wow cant even tell the difference.
Can you imagine this technology being used to impersonate a president? It would be bedlam, but fortunately there are no girl presidents!
I believe they are already doing that to make fake nude leaks for celebrities. It'll be scary.
Impressive and a bit horrifying. Still, it seems to have a big problem recreating hands. A dating profile with a set of profile pics that doesn't feature a hand will now become suspect... until it cracks hands, too.
Plenty of models can do hands just fine
This AI has a type
That is some how cool and creepy all at once. I guess it gives a macro-view of what we consider attractive.
Not really. OP doesn't mention how they were generated, and how the pictures were selected.
It could just reflect the kinds of photos that went into the image generator, and the biases that were in its training data, unless the photos were chosen based on an attractiveness survey or some such.
Most of these look like the same person with different hair and makeup, so it could just be a very specialized network.
It gives a view of what whoever generated these images considers attractive.
A16, C7, still not quite there with hands. đŦ
L15 also, wtf is going on there
I can't decide if B8 or L15 is my favourite.
She has only one buttock đ¤Ŗ
The one right below it has 6 fingers also
We all have trouble drawing hands hahahah.
Truth.
Shh they'll find us
L15 has a unibutt
And sadly her friend is looking at it and laughing. Some friend.
Finally, some positive representation in media for people with diverse bodies / disability! /s
There's a ton of fucked up hands in there. It also seems to struggle with handbags. It's kinda fun to try to find one that isn't flawed in some way, I haven't found one that isn't.
The hands are the giveaway in an otherwise convincing image almost every time.
Check out the fand on A7
real females remixed
Most real women don't look like twenty year old thin white models.
But clearly that was the dataset used here.
A lot of them look like they're 16
There are repeats too
The cop probably has pointy elbows.
This is wild.
Is there a source or article to read or something?
At least 15 of those women are mila kunis
There is that one brunette on the right, reminds me of a movie/video I saw where the same AI woman kept popping up.
Tell me they didn't allow American source material without telling me they didn't allow American source material.