Get the top HN stories in your inbox every day.
retrac
I had some fun with StyleGAN a couple years ago. I should probably have written this up, or something. Some others I showed it too thought I should. But I never took it too serious. I was just bored over a couple weekends when I had access to some beefy GPUs. And y'know, it's quasi-porno.
I opted for dudes not gals. Part personal preference, partly because I happened to get my hands on an enormous dataset of hot shirtless guys taking selfies.
The initial results were rather poor: https://i.imgur.com/npzDcCL.jpg
That's around when I realized how important the training data really is. Too much variation in pose. So I trained a model that detected the face, nipples, and belly button, and aligned them vertically with that. Then I briefly skimmed over the dataset manually, and deleted anything that just varied too much from the median, in a rather arbitrary way.
Results were better: https://i.imgur.com/bS7ERC6.jpg
So I just let it churn away for a week, heating up the basement. Some final results with narrow truncation and style transfer: https://i.imgur.com/rhnWnpK.jpg
One thing I find intriguing is how bland those men are, really. Oh, beautiful, but too classically beautiful. In a sense, StyleGAN is a compression algorithm, and highly truncated outputs are something like a median of the inputs. So the above reflect a sort of "average" idealized image, that contains the traits most commonly found, in the media I trained it with.
jancsika
In your example link here:
https://i.imgur.com/bS7ERC6.jpg
you show one fellow with a smartphone stabbed through the chest, plus a headphone cord fused into his forearm veins.
When you "let it churn away for a week," which of the following is more likely:
1. Smartphone Stab Wound Guys decrease in likelihood
2. More realistic-looking smartphone stabs
In either case it doesn't exactly inspire confidence in self-driving cars.
jiggawatts
In a future Tesla shareholders meeting:
“Version 10.2 can detect children crossing the road with 97% accuracy, and collides with them dead on at the same rate. Those well publicised close calls are now a thing of the past!”
spiderfarmer
To be fair if a kid crosses a road dangerously in front of human drivers there are a lot factors at play that impede with the accuracy of those drivers. No system is perfect and we shouldn’t expect that.
cyanydeez
The good news: we kill less people.
The bad news: when we do, it's unfathomable how anything but a computer could do this
anigbrowl
Sounds like it might be time for another instalment in the franchise https://en.wikipedia.org/wiki/Tetsuo:_The_Iron_Man
undefined
tailspin2019
Superb
fire
This would make for a hilarious blog post
metadat
cirgue
Because when presented with a bunch of pictures of hot shirtless dudes, thinking “how can I apply cutting edge technology here?” simultaneously isn’t most peoples first instinct (regardless of your gender identity or sexual orientation), but is immediately relatable if you’re a technology nerd. It’s both unexpected and familiar, and that’s the magic formula for humor.
forgingahead
The community can't have too few blog posts / ML experiment writeups, so do write a blog post on your experiences, it would definitely be interesting.
chimeracoder
> partly because I happened to get my hands on an enormous dataset of hot shirtless guys taking selfies.
I feel like this is burying the lede. I'm very curious where you sourced this training data.
dzhiurgis
Wabi-sabi is a thing. Friend noted years ago how kids who are super aware of their weird teeth or hair colour turns out to be their most beautiful trait later.
Buttons840
Those first images remind me of the "brainlet" memes you often see in high-quality shitposts (something I appreciate more than the average HN reader I think).
scrollaway
SFW and hilarious :) https://pornpen.ai/view/4m7Sa4OnVP4yMPNqikmC
anothernewdude
rickstanley
Look at that reflects
seanp2k2
koolba
They’re even more wonderful to stare at when there’s a pair of em’.
Boobies are beautiful as well: https://en.wikipedia.org/wiki/Booby
userbinator
I suppose that could be considered "contamination" in the training data.
thematrixturtle
Inevitable successor to this: a TikTok-style adaptive porn feed that learns exactly what you like and starts generating porn customized to your kinks and preferences.
dgs_sgd
Porn addiction is already a serious problem for many people. Combine an enhanced version of this model with TikTok style delivery and I’m very fearful of the end result. It’ll be the equivalent of crack in terms of the dopamine rush.
sbf501
We may experience the first ever dopamine burnout. How much can the brain handle?
Remember the film "Brainstorm" from the 1980's? Where the guy almost dies because he blasts a real-life full brain capture of sex into his own brain in a loop for days?
We'll probably learn if there is a thing as too much dopamine.
Melting_Harps
> Remember the film "Brainstorm" from the 1980's? Where the guy almost dies because he blasts a real-life full brain capture of sex into his own brain in a loop for days?
No, I had to go look for it and it was on youtube (NSFW) but is this the scene [0] you're talking about? This was likely what inspired the braindance features in Cyberpunk 2077, it's pretty much the 1980s iteration (with a massive tape) on what looks like a futuristic VHS cassette and a phone for some reason.
Well, it wasn't exactly just dopamine, but there was woman who pleased herself to severe level [1] with 30 min orgasms that impacted her entire life, it's a become a meme on JRE [2].
Sidenote: this is probably the only film where Chris Walken wasn't like he is portrayed now, even Kings of NY had that vibe about him but here he is kind of... normal?
0: https://youtu.be/cOGAEAJ4xJE?t=2969
1: https://www.zmescience.com/other/feature-post/brain-pain-cen...
floren
Reminds me of an Arthur C Clarke story, Patent Pending: https://en.wikipedia.org/wiki/Patent_Pending_(short_story)
jcims
I cringe at the thought of watching it again and dismantling it from the very high place it currently stands on my all-time favorite movies list. Mostly just for the concept, I can't even remember the plot.
tianshuo
Sorry but to say, TikTok for P*rn has already been made, works for iOS and Android. Probably already has millions of users, with a percentage of them uploading original content. Not giving the name or link here for obvious reasons.
LewisVerstappen
It’s Time to Build
wildpeaks
"If you build it, they will come" takes a new meaning
caf
As does "normalization of deviance".
pmayrgundter
clap clap clap
moomoo11
That would be a cause for the models on TikTok to unionize because that’s taking their jobs away.
kbenson
The really smart models will put together an AI to generate images of them using only their own images as training data, and then they can sit back and profit.
Eventually we'll be watching generated personalities on twitch playing video games, and then we'll know the end is nigh. There is no infinity, eventually the snake chokes on its own tail...
Barrin92
>Eventually we'll be watching generated personalities on twitch playing video games
that won't happen because the entire value proposition is in the personal connection to the streamer. (the modern derogatory term is 'parasocial relationship').
just like nobody watches chess computers or starcraft bots play nobody is going to watch bots play games. in fact to stay on the topic of the thread, people are making millions on onlyfans because they realized chatting with their viewers is much more valuable than generic pornography.
zyx321
We've already got vtubers playing robot characters, it feels like it's only a matter of time before someone attempts to make a 'true' AI vtuber. Though I suspect we're still a while away from AI being able to complete arbitrary video games.
Zuider
Those generated images, fixed permanently in the stage of early adulthood could still be produced when the model is well past retirement age.
XorNot
There's a business model in there though: the algorithms need well tagged poses and images, and one of the biggest issues with sex work is that it is used to attack people in other areas of their lives (even in a perfect society for this, the desire not to be directly personally associated as a part of your identity would probably still exist).
If you can retain paying customers, and then pay performers for a supply of source material (which won't, unaltered, ever end up in customer hands) then there's a new business model somewhere in there.
woudsma
And then sells your preferences to the highest bidder. A bit like the Westworld HBO series.
6510
For the general public you mean. I'm sure Bezos already has a feed of "urinate in pants".
Why the downvote? At 15 $/h his 162.7 billion could buy 1.24 million years worth of bathroom breaks. You want to argue other motivations are at play here?
markdown
So, just a feed of Liam Neeson? https://cheezburger.com/13330949/liam-neesons-unapologetic-p...
dymk
I'm having some trouble finding it through search, but if I recall, this project was submitted to HN at some point.
bkanber
Was it this one? https://news.ycombinator.com/item?id=9847283
If so, it existed for a while but shut down years ago.
dymk
Yes, I think this is it. It wasn't generated porn, but it would hone in on what you liked similar to Tic Tac.
collegeburner
hmm yes that will definitely not totally fuck up the dopamine systems of young men and make an already badly addicted generation hopelessly dependent on this shit
LewisVerstappen
We lost that battle a decade ago
superkuh
Porn "addiction" and all that hand wringing is just FUD from the moral panic types and the skeezy media "doctors" that prey on them. In fact non-substance "addictions" aren't recognized at all. "Gambling disorder" is presently the only condition in the subsection of "Non-substance-related disorders" in DSM 5.
Addiction mediated by chemicals that directly increase perceptions of incentive salience through mesolimbic dopaminergic activation are real addictions.
Perceptions of sensory stimuli come in through the normal paths do not directly hijack any part of the brain like addictive chemicals do. Downstream of the stimuli there may be reward learning but it's through normal channels, not directly turning up incentive salience. You actually have to like/enjoy something first before reward learning happens.
wyager
Is there a DSM-5 condition for outsourcing all your rational facilities to the DSM-5?
DiggyJohnson
Carry on doing your thing then.
MisterMower
They’re working on changing that, too.
kurupt213
Thats a long description of habit forming
undefined
tagami
/snark Welcome to the metaverse
stavros
Better than Instagram, at least.
collegeburner
no, its gonna have people combining dark pattern skinner box shit from insta and tiktok with the addictive nature of porn. and its gonna be way worse. yall wouldn't believe me if i told you how common straight up porn addiction is in young kids, it's fucking bad already.
edit: @stickfigure i'm can y'all please quit trying to tar me with this brush, it's bullshit. i'm not arguing about masturbation i'm saying excessive consumption of porn when people have limitless access to weird ass shit that doesn't reflect the real world is causing problems. don't try to make me some puritan or whatever because literally nothing i said suggests that.
EddySchauHai
If people wanted to work on this, I genuinely believe that could be huge.
postalrat
Eventually this and similar technology will be able to generate unlimited porn of all types. Including CP. Should or even could it be made illegal?
To me that kinda feels like if it was made illegal then you are basically acting as the thought police.
Edit: are people just uncomfortable talking about this or do they think downvoting is going to somehow prevent it?
nikkwong
The question of whether or not machine generated CP is ethical or not is a confusing one to me, since the crux of problem of CP is that it's violating the rights of a child. If you get rid of that fact, CP becomes a 'victimless crime', in a sense, and it becomes hard to pinpoint what the actual perpetration is. (Ignoring the fact that a CP model has to be trained on real images, which destroys this argument outright, for now).
Nobody wants to be a pedophile, but if you are unfortunately hardwired that way the impulses are impossible to ignore—similar to any other sexual impulse that 'healthy' adults have.
I think the goal is how to create the least amount of harm for children, and it's theoretically interesting to consider the possibility that machine generated CP could be used to ultimately end the real-life CP black market.
I'm sure this won't ever happen because nobody wants to touch this subject with a 10-foot pole; and our species tends to favor solutions that are idealistic rather than rational in many ways. But the European approach to criminal justice is something like: "humans are not perfect, and we can't fix them, so let's try to engineer solutions which make living together more pleasant". Nobody likes an icky solution like this, but the alternative is a lot of children suffering (although I'm not saying those two options are perfectly substitutable).
On the other hand, it could be a completely stupid idea outright. Just an interesting thing to think about.
Alex7845
>(Ignoring the fact that a CP model has to be trained on real images).
that's not how this tech works.
Did you think actual guinea pigs had to be trained to do household chores for these pictures to be generated? (dailywrong.com/new-course-teaches-guinea-pigs-household-chores/)
I had a lot of fun creating pictures of llamas in space. I'm pretty sure no llamas have ever walked on the moon.
nikkwong
I think it is how the tech works; though. The model knows what a guinea pig looks like, and it knows what household chores looks like, so it can concatenate the two. I don't think the model knows how children look like unclothed, and I don't think it can posit that by looking at nude human adults. But I could be very incorrect there; and these techniques will certainly evolve over time.
mc32
Question is does it reduce demand from the real world or does it serve as demand multiplier in the real physical world?
nikkwong
I think these are the right questions to ask; and these types of things would need to be studied in order to pose a potential solution. But I can't imagine any modern day institution beginning to pose these questions without facing public outlash. The sensationalized headlines that would spin out of this would probably stop any research in this area in its tracks.
In that way our species is very primitive. We're trigger happy in our outrage and often unable to focus on a larger shared goal, much to our own demise.
Viliam1234
I think the research shows that availability of normal porn reduces rape. (Don't have a link at hand, sorry.) So I would guess the effect would probably be the same here.
wraptile
It's not entirely a new issue, see the discussions around "loli erotic japanese manga" etc. Though I agree with you, nobody is touching this tech anytime soon (at least publically) even though you can easily imagine safe applications like MRI with AI generated images to study exact brain patterns for potential treatments and so on.
girvo
I don’t think you can ignore the fact that the model would have to be trained on child sexual abuse material, though. That’s pretty important.
nikkwong
I view that currently as a technical limitation that will likely be lifted as the technology progresses. Our species has been hard at work on the field of AI-image generation for only a decade, and we have thousands of years to go. So I would posit that this barrier is probably a short-term one.
bhedgeoser
What about writing a blog post about raping and killing people. We have to draw the line somewhere and for CP the line is at the first step.
ignoramous
Unfortunately, don't think folks watching CP would stop at that. Porn is an addiction (or compulsion?) and I don't see how CP makes it any better for folks with pedophilic tendencies.
To think that porn reduces sexual tendencies may be true (in the short to mid term), but eventually, as the addiction intensifies, those hooked have the urge to seek more novel, more aggressive versions of it (in reel or for real).
https://en.wikipedia.org/wiki/Effects_of_pornography#Sexual_...
version_five
Otoh, I have read articles worrying that young men aren't going out looking to meet women like they used to, because they are satisfying their desires though porn.
deepdriver
Not at all contradictory. Opioid addicts need higher and higher doses for their fix. In the worst case they too drop out of society, committing crimes such as theft and prostitution to service their addiction.
hoseja
Oh no the engine of society is losing fuel to burn through, how awful. How will the necrarchy ever afford their servants if the subjects have dodged the exploitation of their reproductive urge.
undefined
psyc
Add reefer into the mix, and you've got a recipe for madness.
Throwaway556655
I don't know if you're being sarcastic here or not, but I've mixed porn with stimulants and the result is definitely pretty. I've went through 16-hour masturbation sessions and have done things I would havje never imagined. The thoughts still haunt me. Porn addiction is not a joke.
whiddershins
It is a terrible idea to compare pedophilia to drug use.
One might be a victimless crime.
The other never can be.
8f2ab37a-ed6c
Is that any different from the argument about violent videogames back in the day? Or Dungeons & Dragons before that?
komali2
I wish people would explain why they're downvoting these ethics-related comments so aggressively, because I'm interested in these conversations and not sure why other users find them irrelevant.
dang
The guidelines specifically ask you not to post like this: https://news.ycombinator.com/newsguidelines.html. It's completely off-topic and, what's worse, repetitive.
It's frustrating not to be able to know the reasons for downvotes, but it's the nature of the beast. Sometimes people think we should require explanations, but you can't change the way people are by demanding they jump through a hoop.
If you see unfairly downvoted comments, give them a corrective upvote (https://hn.algolia.com/?dateRange=all&page=0&prefix=true&sor...). Other users will hopefully do that too.
A lot of the time, that happens and then posts like this end up as uncollected garbage in the thread, complaining about something that doesn't exist anymore.
noduerme
I think it demonstrates why it's so easy for technology to run ahead of ethics. No one wants to hear a public safety announcement when they get the keys to a new car.
OTOH, I think the risk of CSAM from this kind of thing is overblown, because either (1) it wouldn't actually involve real children in any way, or (2) the model would have to be train on actual CSAM, which there are already laws against.
If no humans are harmed by it at any stage, then how's it any different from the CGI murders all over movies and TV? In the absence of harm, the only reason to consider it a crime is that it's disgusting. But we don't prosecute people for having sick thoughts, just for acting on them. If you take the child and abuse out of the equation, the only argument left is that seeing CSAM (artificial or not) is a "gateway drug" to commiting child abuse. If that's the case then why shouldn't we prosecute people for watching Dexter, on the grounds that they're on their way to becoming serial killers?
fragmede
I'd bet that downvoting is a kneejerk response to the claim "porn is an addiction".
pid-1
Developers like pretending technology is always good and/or inevitable, I think this sort of question is uncomfortable to many folks.
undefined
dang
We detached this subthread from https://news.ycombinator.com/item?id=32573719.
dragonwriter
The GP comment made largely unsupported fact claims, rather than ethical ones.
psyc
Ethics? Please. It's moral panic. The putatively ethical folks having a "harmless" academic discussion about which nice things everybody is or isn't allowed to have, according to them.
int_19h
Many countries already ban painted or rendered CP.
US did that for a few years, as well, until the Supreme Court struck it down: https://en.wikipedia.org/wiki/Ashcroft_v._Free_Speech_Coalit...
So as far as the output goes, there's already some kind of precedent in most countries. Or are you asking about the model that performs the generation?
mgdlbp
What? That's about things that _aren't pornography_.
US convicted for written obscenity just in 2013: https://www.techdirt.com/2013/05/20/court-finds-fantasy-stor...
int_19h
The Miller test applies to all pornography, not just CP. If you look around, you can find federal cases about "transportation of obscene matters" for adult stuff, e.g.:
https://www.networkworld.com/article/2276921/porn-producer-s...
goolz
I’ll bite. The issue is it will take all of 5 seconds for an enterprising CP producer/admin/consumer to mix AI content into their stash and then claim they’re being persecuted when the feds come knocking.
It’s like lolicon, sure, you can sit there and pretend it’s a 2000 year old demon warrior… but she is still dressed, posed and drawn to mimic a small child.
Don’t get me wrong, it’s a conundrum, that’s for sure. I hate the idea of the thought police. I love free speech. But in the end, I believe any material, whether it is “art” or computer generated, should be barred if it contains images that can be construed as a child in a sexual act (ancient art gets a pass?). Lolicon does not stop pedophiles from being pedophiles. Nor would this. It would simply act as a gateway and a scapegoat imho.
rpigab
There's prolly gonna be a big controversy when pedos start using it that way and proof is posted on mainstream websites. I've just discovered lexica.art and I love all picture generator AI tools, but when stumbling upon some pictures of young Emma Watson topless, it didn't seem ok, and reading the prompt for these images showed that the user didn't go for tags like "young" or "topless", it's just StableDiffusion randomly selecting these features, maybe because there's wierd content about Emma Watson online on shady sites, maybe something else.
I'm not sure I want all of them to close access to these tools, stop providing open source capabilities, or prevent specific keywords like Dall-E does, but it's the easy solution for providers. This is why we can't have nice things (there's so many interesting, funny and beautiful things that aren't illegal or even controversial but that are blocked as side-effects of Dall-E's restrictions).
schuke
This would be extremely controversial but I remember The Economist mentioned porn for pedophiles as part of the solution (the premise being it's a genetic condition that criminalization doesn't work very well against). AI generation would remove some the moral issues.
anigbrowl
That will definitely happen; I suspect the criminal offense will be adjusted toward making distribution illegal, but possession might be tolerated (in some societies) if an artificial origin can credibly be established.
It's very hard to predict how this will shake out. I personally doubt that looking at extreme/contraband porn is a good strategy for coping with highly illegal desires but obsessive behaviors aren't something that's easy to model reliably. Considering that weaponizing false claims of predatory behavior as a coercive social/political tactic is a big thing at the moment, you probably shouldn't hold your breath hoping for any sort of consensus to emerge in the short term.
undefined
dragonwriter
Unintentional The Exorcist Rule 34?
NSFW, obviously: https://pornpen.ai/view/27ylon4lGNpkfA4iJKJB
wiml
Made me think of this (also NSFW, mild though): https://www.oglaf.com/chauncey/
czottmann
Man, I remember when I first discovered Oglaf. What a weeeeeird comic, loved it.
bhedgeoser
> also NSFW, mild though
It either is or isn't.
omniglottal
Safety is not described by just a boolean value.
fdupress
Take a look at this one (also NSFW) https://pornpen.ai/view/CBHr5AlP8AFAZ56Jkp7i
legohead
Made me laugh. Showed my wife and she said, stone faced, "isn't this what you would want?"
yuan43
What's going to be really wild is how the availability of convincing technology like this will seep into expectations that women think men have, and how they modify their own bodies to adapt. This has already happened, even with today's limited production capacity using human models/actors.
What happens when literally any crazy fetish can be explored infinitely?
Based on the images linked here (server has melted down), this stuff is not very convincing - yet. But when it gets there and literally any image can be called up, with endless variation, on demand, that's when things will get strange.
> I explicitly removed the ability to specify custom text to avoid harmful imagery from being generated.
Eventually, this technology will be available to anyone with a few bucks to spend. The software won't be run on a server but on local machines. Those restrictions will not last. It will no doubt cause a crisis of sorts as the lawyers and politicians try to make antiquated concepts of decency and harm apply to hyper-realistic material generated algorithmically and without the direct involvement of any person.
radu_floricica
I don't know. I mean, it's been really fashionable to say stuff like this for at least a couple of decades, but in the real world I see things playing out in pretty much the complete opposite. Guys are still "yey, boobies!" no matter the shape and size, most of my male friends would like more serious and traditional relationships, and the only type of women that get rejected are grossly overweight.
The only thing porn is desensitizing us to is... milder kinds of porn, really. Relationships get to live in a completely different mental box. I mean, think about the world around you - do you know of any real life people that got split up because they had regular problem free sex... but it wasn't kinky enough? If anything, I've only ever heard women complain about bland sex. Men are still happy to have it, check that box and judge the relationship satisfaction on the rest of the metrics.
smsm42
It's almost like many adult males are capable of distinguishing between the pretty picture created for entertainment purposes, and the reality of human relationships. Who could have thought?
A lot of people also behave like pictures of naked women (or availability of actual naked women, for viewing and more, provided appropriate payment is made) is a new thing that human males only got access to with the invention of the internet. It's not that new, it's just another facet of the same old thing, and the same would happen - some small number people would develop an addiction, a bit more people would be regular consumers, a real lot of people would be occasional consumers, otherwise not a lot changes.
Kye
People who see those mental boxes as sharing the same space also struggle to understand kinky asexual people. It seems like my ace friends commission more porn than allo[0] friends.
thrown_22
>What's going to be really wild is how the availability of convincing technology like this will seep into expectations that women think men have, and how they modify their own bodies to adapt.
US waist sizes tell us women think men find walruses attractive: https://www.cdc.gov/nchs/products/databriefs/db360.htm
The idea that even a minority of women do _anything_ for attracting men is quite wrong when you look at actual behavior rather than day time television. The converse is also true. I find it astonishing that a country which is 85%+ fat has some how convinced itself that it has a problem of unrealistic body expectations. We're closer to the world represented in wall-e than that of playboy.
Melting_Harps
> The idea that even a minority of women do _anything_ for attracting men is quite wrong when you look at actual behavior rather than day time television. The converse is also true. I find it astonishing that a country which is 85%+ fat has some how convinced itself that it has a problem of unrealistic body expectations. We're closer to the world represented in wall-e than that of playboy.
Best post thus far, well done.
CTDOCodebases
Is it fair to include people in relationships in these statistics then? It’s not uncommon for people to “let themselves go” once they are in a relationship.
Shugarl
I always thought that if something's bothering you, it's much easier to change yourself in order not to be bothered anymore than it is to change the world so that the thing bothering you doesn't exist.
It's easier to be careful with where and when I go somewhere than it is to create a world where I won't get mugged/assaulted. It's easier to be content with the life I live than it is to become rich. And so on...
lofatdairy
To be clear I'm not making any statements about the creator, I imagine this is a reflection of the dataset: but the fact that there are so many categories for east Asian (Chinese, Japanese, Korean, and I hazard that the "Asian" tag would probably mostly generate east Asians) says volumes about the sexualizing gaze cast on Asian women. While certainly there's an ethical argument over such an algorithm (both for and against) with regards to sex workers and exploitation, it seems to me that the granularity of choice and what choices are available to the consumer also may require additional consideration, lest it reinforce negative stereotypes, introduces new stereotypes, or allows for harmful attitudes to be stimulated.
I'm not trying to be sanctimonious or overly moralizing, but half the tags I would associate with ethnicity are east Asian. And while I lean towards treating it as a mere curiosity, I think it's worth discussing as having unintended consequences and interesting implications at the very least.
Jensson
Blonde, brunette, ginger and white all refers to Europeans, which represents less people than the Chinese tag alone, so it doesn't seem like it fetishizes Asians more than whites.
The strange part is why we don't have equally rich categories for Africans. Native Americans not having their categories makes sense since there are so few of them left today, so Africans are the only large group left out.
lofatdairy
I thought about that while writing my comment, but while hair color correlates pretty well with ethnicity, the level of granularity given to Asians still seems somewhat unique. There are, for instance, no distinction between French and Italian, nor British and Scottish categories. Blonde and ginger do have fairly ethnic implications, but not many people would answer with their hair color if asked where their family originates. That, and the fact that these hair colors are not restricted to white people, due to multi-ethnicity being a thing, as well as commercial hair dye, made me drop them in my "half the ethnic labels" comment (though I will concede that the image in most people's heads when hearing any of those terms would be a white person).
This is also coming from a perspective where I'm assuming that the audience for this algorithm (outside of academic consideration) would be westerners where white people tend to be the majority. In that case fetishization generally would only apply to racial minorities, due to the fact that the term seems to imply a sexually gratifying quality to the "uncommon". I add that, speaking as an Asian American, that it's not necessarily all that easy for most people to tell Chinese from Japanese from Koreans, whereas it's basically tautologically given that a blonde can be differentiated from a brunette since the terms are more strictly aesthetic rather than geographic.
On your point about Africans, I would extend it to both Latina and the fact that "Indian" seems to be a catch-all for any South Asians. Middle Easterners are also not represented (though this doesn't strike me as a thing where racial equality is necessarily desirable). However, this is where I think the data set is mostly responsible.
numpad0
I think they took the tags from some uploader website. CJK ethnicities can be inherently identified by types of sources(smuggled amateur films, commercially made, etc) as well as by styles, when appearing in those sites.
lodi
> but the fact that there are so many categories for east Asian... says volumes about the sexualizing gaze cast on Asian women
I think you're reading too much intent into 4 labels. From an American perspective, Japanese was the original--commercial, censored--"Asian" porn, "Chinese" is typically amateur uncensored, etc. I... won't elaborate further here, but there are stylistic differences that have nothing to do with race.
Kye
The tags should reflect that if it's really about the niche and not the ethnicity.
gricardo99
You raise some very interesting points. For me this also demonstrates how deeply rooted, historical bias and prejudice bleeds through to this technology and could become a perpetuating factor. These models will reflect societal values, for better or worse.
But also what will the next model, and successive models, that are trained on data from AI generate images, be like? It seems you could get an amplified or skewed model that is pretty far from reality. At least currently they are all trained on “real” images (granted some images are heavily edited or photoshopped). I can imagine that In the (not too distant) future most images will be generated and not “real”.
lofatdairy
Yeah, I wanted to stress that the categorizations seem more to me to be symptomatic than a reflection of OP's decision making. I loathe the label "problematic", but it seemed worth noting a potentially unintended consequence that could introduce some harms.
That said, it might be too early to speculate on models trained on the outputs of other models. This algorithm seems to benefit greatly from the post-DALLE 2 art-gen renaissance, but still has a number of oddities and uncanny outputs (noted in the rest of this thread) that I imagine are nearly unavoidable given how familiar the human form is. But certainly, if such decisions end up amplified in later work, or through widespread usage, then the consequences are more significant than, say, simply reflecting uncritically a skewed data set.
tsol
This is one of the most interesting comments here. I notice a lot of biases in image generation AI, it's so subtle it's almost at an ideological level. The style of art it creates, the color and dress of the people it generates-- there are little things that that make you wonder "why this instead of something else?"
garmanarnar
Keep in mind this AI is biased towards white men's interests which means it heavily fetishizes and degrades Asian women.
Even searching for other races shows predominantly Asian women in the results, its quite disturbing.
vimy
That didn’t take long. :p Stable Diffusion released just yesterday.
How did you design the backend? What kind of server are you using?
tomalaci
People doing porn can now truly claim they are doing "science" by providing valuable, insightful data samples :)
It shouldn't be surprising to see quick future advances in this domain as most of the porn material is usually available free, easily scrapable and extremely low risk of copyright litigation.
IMHO, the real game-changer as with normal art will be some sort of reliable assistant tool to quickly generate various components. Further finishing touches can be then easily done in Photoshop or similar software. Sometimes along this we will probably see results in 3D art or animations.
yieldcrv
> and extremely low risk of copyright litigation
but for a different reason than now!
sometimes porn fails a copyright infringement claim because it can be argued that the particular piece of work doesn't satisfy Article 1 Section 8 of the US constitution backing the copyright concept "To promote the progress of science and useful arts, by securing for limited times to authors and inventors the exclusive right to their respective writings and discoveries;"
It's still unsettled case law because it's only troll law firms that are challenging on behalf of porn copyright holders, and they never appeal.
rootw0rm
hmmm...
https://pornpen.ai/view/g56NHhdoqjvO90wF7Wyt
https://pornpen.ai/view/0dzppk9MhxNJvZiYw3Dc
https://pornpen.ai/view/fXBYznNQuJx8gQfV0nKi
https://pornpen.ai/view/CDgIuqd7KhWLxZzrrSa1 (bonus copyright logo)
https://pornpen.ai/view/3k0tTyXmjFQDiUPFJebo
okay, that's enough nightmare fuel for now
ticklemyelmo
You have something against the ladies of Eroticon 6?
tartoran
Nude bodies and fused body parts… These things are very disturbing to my mind as an adult. Scares me to think children would run into this kind of content online and what it would do to their psyche. Now I have to admit that some of them made me laugh but still, I fail to understand why in the 21st century we’d need to waste energy on these types of AI applications. It stops being interesting after the initial wow reaction. Hope this won’t open a subculture and will be relegated to curiosities in technology
tsol
I can see strange subreddits dedicated to girls with their heads backwards or with three arms. If people can get turned on by the cartoons they grew up watching, I think that shows the brain is incredibly malleable when it comes to sexual desires
nichaku
https://neuralblender.com uses the same algo but with open prompt. no login required
NSFW Example: https://s3.amazonaws.com/ai.protogenes/art/d917ae82-2236-11e...
nichaku
They could make use of a face restoration model post generation like Tencent gfpgan https://replicate.com/tencentarc/gfpgan
nichaku
also seems to be very fast (I waited like 5 seconds)
wikitopian
One positive outcome of this sort of thing is that revenge porn will become less effective as women will be able to plausibly assert that this tech was used to generate the images.
nichaku
yes, I think with all those existing deep fake tech we are already there.
Get the top HN stories in your inbox every day.
Hey HN, I've been working on https://pornpen.ai, a site for generating adult images. Please only visit the site if you are 18+ and willing to look at NSFW images.
This site is an experiment using newer text-to-image models. I explicitly removed the ability to specify custom text to avoid harmful imagery from being generated. New tags will be added once the prompt-engineering algorithm is fine-tuned further. If the servers are overloaded, take a look at the feed and search pages to look through past results.
For comments/suggestions/feedback please visit https://reddit.com/r/pornpen
Enjoy!