Meet 'Unstable Diffusion', the Group Trying To Monetize AI Porn Generators - Slashdot

2022-11-23 17:18:56 By : Mr. jack jia

Want to read Slashdot from your mobile device? Point it at m.slashdot.org and keep reading!

The Fine Print: The following comments are owned by whoever posted them. We are not responsible for them in any way. Sexy Dolls

Meet

Back in the late 80s and 90s they were postulating the existence of AI powerful enough to make you whatever sex doll with real people's faces on it that you would ask for.

Strangely enough, while both Roddenberry and his successors acknowledged the ethical ickiness of fucking virtual versions of your friends and co workers without their consent, no one in the writer's room thought the computer would have had lock outs that prevented users from making sex toys with real people's faces.

no one in the writer's room thought the computer would have had lock outs that prevented users from making sex toys with real people's faces.

That's because that would make stunningly dull sci-fi.

Hey futuristic speculative technology raises shines new light on the human condition and raises new ethical questions except it doesn't because a different piece of technology prevents it and everything is actually exactly the same as it is now.

And also everyone knows that technological lockouts are 100% effective e

also everyone knows that technological lockouts are 100% effective especially to those with lots of resources.

also everyone knows that technological lockouts are 100% effective especially to those with lots of resources.

Sure, if they had lockouts then they could have episodes about the people who bypass them and why, though.

They didn't have lockouts for the very simple and obvious reason that it isn't the real person. If you can generate fantasy images of people in your head then why should anyone stop you from doing the same on a computer?

If you can generate fantasy images of people in your head then why should anyone stop you from doing the same on a computer?

If you can generate fantasy images of people in your head then why should anyone stop you from doing the same on a computer?

The only person I can recall doing this was Reg Barclay, and everyone shakes their head every time they catch him at it like it's a shame, so clearly there is still a stigma in the 24th century. There's even a reference to it in Star Trek Online. Straight to horny jail, Reg.

But to answer the actual question asked, the answer is entrainment [frontiersin.org] .

Yeah but I take that as an allusion to the stigma attached to masturbation in general. Even in the modern day with full awareness everyone masturbates your friends will still poke fun at you a bit if they catch you at it. The jokes are too good to pass. ;)

The question was rhetorical. Maybe it's just that I haven't had my coffee yet but I feel I might be missing a few dots to see what you are getting at when you suggest a connection to entrainment?

There are lots of claims that people are just using certain things as an escape valve, but the truth is that doing things leads to a habit of doing things. I know that's vague, but it's a general principle.

So you are speaking either of habit forming or perhaps even behavior patterns in fantasyland starting to color your interactions in the physical world? There is a lot of debate that could be had on that but fantasy in your head vs modeled on a computer have virtually all the debatable stuff in common.

Will some psychos go obsessive and act out their fantasies on unconsenting people? Yes. Will yet more make some sort of pass or engage in a behavior because they aren't mentally on the same page as the other pe

Tom Paris set Tuvok up with a holographic version of his wife to alleviate his pon far sex rage.

It's alluded to that Seven of Nine had intimate relations with holo-Chakotay without his consent as well.

The was an episode of DS9 where Quark was trying to get a scan of Kira for a customer. You would think that in the 24th century it wouldn't be hard to get a high quality image of someone and have AI fill in the blanks, especially in a bar with gambling.

Anyway, from memory it was presented as something that you can't really do much about. Same a celebrities having their faces photoshopped into porn today. Kira managed to sabotage the program in the end, replacing her face with Quark's.

I assumed that the holo-imager was doing more than just capturing photons, it was a sensor array that captured other detail that the holodeck would need to faithfully reproduce something. Things like surface texture, temperature, perhaps some kind of internal imaging like a passive x-ray or MRI.

I figured it was that level of detail that required special "imaging", rather than just having the computer fill in the missing information with an educated guess. It's the same with these AI systems. If you ask for

Or by Wren from Corridor Digital to go full Kaiju [youtu.be] (among other things)...

I for one had always wondered how that was supposed to work as the holodeck's computer would have to work on a lot of assumptions and likely high context based on the person giving the commands in order to not always have the user refining their prompts.

I for one had always wondered how that was supposed to work as the holodeck's computer would have to work on a lot of assumptions and likely high context based on the person giving the commands in order to not always have the user refining their prompts.

Presumably it could learn what each person is expecting when they say certain things, so it would get better over time. Also, if you're asking for what other people have already fine-tuned, it could borrow from them.

the ethical ickiness of fucking virtual versions of your friends and co workers without their consent

the ethical ickiness of fucking virtual versions of your friends and co workers without their consent

Well, better this than do it to their real versions. Of course content like this must not be published. But if that is assured, it is really no different from doing it in your head. Got any "ethical ickiness" with that too?

As long as it isn't passed off as being real or used for illegal purposes like harassment then I see no reason it shouldn't be published as well.

Preventing publishing is merely punishing people who have less skill and time to have less wank material.

Well, as long as it is done for people that are on the public eye anyways, maybe. Doing it for your neighbor is crossing the line IMO.

Honestly I could see it either way. I really think it comes down to how it is used and not so much who. Ideally the materials are just kept anonymous or in the case of someone like a public figure should clearly be denoted as fake.

We've all got doppelgangers out there anyway who might well be putting out all sorts of content and in the vast ocean of porn that is the internet it's unlikely you or anyone you know will EVER see it. You can even look at two pictures of the same person with different lighting an

Stable Diffusion (not talking about Unstable Diffusion), which I believe has now become the most widely used, does not contain built-in lockouts. Some of the web services have them, but you can run SD quite well at home on a $350 RTX 3060. Which has unfortunately led to the StableDiffusion communities being flooded with porn. It's actually rather annoying. It's not even creative porn either.

You can run SD just fine for typical generation sizes (up to about 512x1024) on a 1070, using some of the forks (I use automatic1111)

Not so good for training, but you can do that on colab

All I can say is that I run out of memory aplenty even on a 3060, so it must be a pain to run on even less for anything more than basic generations.

These are just pictures. Closer to what you describe is virt-a-mate. Though these tools might progress to being able to build appearances for virt-a-mate. Build or buy a tempestmax SR6, slap on your VR headset and you've got something far closer to that reality than I ever really expected to see.

As for ethical ickiness. People have been doing exactly that in their heads when masturbating for as long as there have been people. Besides it isn't really their body or face, just one that is similar and everyone

no one in the writer's room thought the computer would have had lock outs that prevented users from making sex toys with real people's faces.

Why? Because such lockouts have to be specific to the scenario being envisioned, and in general, people seem to respect the privacy rights of everyone else in not digging around other people's fantasies.

And by not having it, it allows the writers to explore areas of humanity that are difficult in other ways. Look up Lt. Barclay episodes on TNG for examples on such. Sur

Let's postulate some scenarios to clarify the direction of the discussion:

Is it okay for a star trek character to go into the holodeck and conjure up a version of their co worker to have sex with?

Is it okay to conjure up a more physically attractive version of the co worker?

How about a version to violently rape, or otherwise abuse, torture, maim, kill, and then fuck the corpse?

Substitute co worker for spouse. What would it say if Lieutenant Redshirt walked into the holodeck and demanded a version of his wi

Porn has inspired or moved technologies along from photography to VCRs to broadband Internet (what, you think there was any other reason for needing 50Mb/s download speeds?)

They missed the mark on HD DVD vs BluRay though...

It does. It is not the only thing, but it is a very effective accelerator. Denying that would be stupid.

50 Mb/s isn't really needed for video. Even Netflix [netflix.com] says you only need about 15 mbps for 4k video, and for 1080p you only need 5 mbps. The only real use I can think of for higher speed downloads for the average person would be things like video game downloads and updates, which seem to have reached ridiculous sizes in recent years.

what, you think there was any other reason for needing 50Mb/s download speeds?

what, you think there was any other reason for needing 50Mb/s download speeds?

Yes, but then my imagination extends beyond my penis.

They're going to make a fortune.

Mostly tasteful nudes on that discord server.

Anyone who didn't see this coming has not noticed the world.

OTOH, I can quite understand that folks wouldn't want their own images used that way without their permission. (This isn't going to stop people, but may keep it from being acceptable.) But perhaps artificial images could be an acceptable substitute. Unfortunately, just about every reasonable face looks like somebody, usually a whole bunch of somebodies. (People don't really vary THAT much.) So they're going to need to be able to prove that they didn't copy someone's face. That should be an interesting challenge.

It's the other way around. People's faces are built out of a pretty limited set of traits, yes there are outliers but most people fall into a relatively narrow range of options. I saw a black guy whose facial features were almost identical to my father's once, and my dad was "only" Mexican and Norwegian (yeah, almost nobody but the chinless hapsburg wonders are as "pure" as they think, but anyway.) That, or he faked his death and was living in makeup, but that seems relatively unlikely. The reason there are

Acid trip amalgamations of existing publicly accessible porn content? Yeah, that's gonna be really hard to fap to.

The folks investing in this might want to read up a bit about how drum machines and computerized synthesizers were supposed to put traditional musicians out of work back in the 80s. Turns out people like to be entertained by actual humans. Imagine that.

Don't need AI for that, just some plastic surgery [etonline.com] that cost $20k back in 2014 (about $25.2k adjusted for inflation). That's a seriously expensive extra boob.

And yes, there are also guys who have duplicated "equipment", and that can actually happen naturally (although it is incredibly rare). The condition also has a silly sounding name, diphallia (do not look that one up on Wikipedia, you've been warned).

Chicks with 3 tits Don't need AI for that, just some plastic surgery [etonline.com] that cost $20k back in 2014 (about $25.2k adjusted for inflation).

Don't need AI for that, just some plastic surgery [etonline.com] that cost $20k back in 2014 (about $25.2k adjusted for inflation).

Snopes says false [snopes.com]

Snopes says false [snopes.com]

Snopes says false [snopes.com]

Actually, Snopes says they just find her story highly suspicious. This is actually a great example of why social media fact checking algorithms can sometimes be subjective, because even if it is just a prosthesis and some makeup, it'd argue that the hoax itself is still newsworthy. It's not every day someone gets the idea to try to convince the world they have 3 tits.

https://www.snopes.com/fact-ch... [snopes.com]

That has already been done. Check out the Arnold Schwarzenegger pic titled "Total Recall". Look for the boobie flash in a bar scene.

That's not how diffusion models work.

These programs cannot reproduce any specific image, unless that image recurred hundreds or thousands of times in the dataset. This is a myth. Something like, say, the Mona Lisa or famous photos from the Apollo moon landings or not, yes, you'll be able to reproduce that fairly well. But if you're Jimbob the Artist from Artstation, even IF your (publicly posted) art was used in the training - which is a big IF - it's not going to be able to reproduce any of your works. If you draw mainly cats, it'll tend to include a cat. If you draw neoclassical synthwave pointillism, it'll produce mainly neoclassical synthwave pointillism. It will capture styles and motifs, but it's not going to be able to reproduce any of your actual works.

To be more specific: neural nets are statistics solvers. Think something like Newton's Method, but on massive scales with huge numbers of variables with complex interrelationships. They're solvers of correlations between inputs and outputs.

Diffusion comes into the picture by analogy of diffusion in liquids. Imagine you have a drop of dye in a glass of water that's been diffusing, and you want to roll back time to the original droplet, you could train a neural net to try to figure that out - though the more diffused the drop has gotten, the more that its guessed endstate - while a perfectly plausible starting drop - won't match the original.

Diffusion models do the same with images. Add gaussian noise as a bunch of "diffusion" cycles, then backpropagate based on how well it removed the noise and restored the original image (reverse diffusion).

The problem you get is that images are really big and contain tons of data; doing reverse diffusion on images as a whole is thus impractical. So you first train an encoder and decoder for a latent space. That is, you connect the input and output images, but pinch down the network in the middle. So what passes through that bottleneck must be a conceptual representation of the image, rather than actual image data. So then you can reverse diffuse the latent image, and then decode that.

The concept of latent spaces applies to text as well. And early experiments with textual latent spaces showed that they have interesting properties - they respond to mathematical operations. For example, the latent space for "woman", plus the latent space for "king", minus the latent space for "man", will resemble the latent space for "queen". And it's this sort of property that becomes useful for pairing reverse diffusion of images with text inputs. During training, you encode both the image and text to latents, and then you dot product the two together to create a unified latent that represents both the image and the text. So reverse diffusion has to keep the paired combination of both the image and the text coherent.

So when a user runs such an application, their text is being encoded into a textual latent, but then it's just being dotproducted with random noise. Repeated cycles of reverse diffusion then make the latent coherent again. And then the decoder from latent space to image space is run, and that's your output.

So again, with these tools, you need to think in terms of latents. What concepts is it learning about its training dataset. Despite popular perception, they're not some big database of images that they photoshop together in random manners; they're "applied image statistcs". Indeed, the common perception would be literally impossible; the size of the checkpoints, as measured in bytes, is smaller than the number of images used to train them! All it knows about any artist as a whole (let alone an individual work) could in general fit ito a single tweet.

Lastly, your music analogy is terrible. Most music today is created electronically.

There's less than 1 byte stored per training image, so it'd be pretty difficult IMHO to argue that it's derivative to any particular work, unless said work is itself so derivative that it could be described in less than 1 byte.

You can reproduce pretty well specific works that are frequently repicated in the training dataset, like the Mona Lisa, certain famous photographs from the Apollo landings, etc. But most works are one-offs, and it really can't reproduce them on its own (through anything but sheer luc

That was the best and most accurate description of the process I've seen, well done.

What is the problem with AI generated porn? When we criticize porn, we very often talk about the "victims" (the actors) and argue that the immorality is the exploitation of vulnerable people to perform in sex acts for money. Much like our criticism of prostitution.

When people are upset about AI generated porn or other porn that don't involve actual human beings, I suspect they may be morally corrupt; trying to apply their own moral standards to others. If you choose to not use porn, fine. If I do, fine. That's a respectable moral choice. If I get upset because you use porn, and my arguments about the "victims" fall apart because your porn is AI generated - well, then I should re-visit my own morality and motives, rather than go on a crusade against your sexuality.

If a company make money, selling AI generated porn or the technology to make such things, is that a problem? Isn't making money from new ideas at the very core of the American Dream?

If the AI generated porn use real persons as models, well, then they cross a line, in my humble opinion.

My point is not that AI generated porn is great. I'm already conflicted about AI generated art, novels and movies. We need a thorough discussion about AI in our culture(s), but I don't think porn is the appropriate starting point for that discussions. Simply because too many people are outraged about other peoples sexuality, and will make up all kind of bs arguments to attack it.

Criticizing porn and prostitution generally is already morally corrupt. Both are known to significantly reduce rape and sexual assault when they are made legal. Both are known to be significantly safer than doing it with amateurs. The whole "victim" narrative is bogus and stems from the fantasies of the religiously (and otherwise) deranged and essentially just shows what they fap to or would like to fap to. There is a reason so many priests rape children and it is not because they have no alternatives. It i

Agree with you but although forced prostitution is rare relative to the entire prostitution reality, it is much more frequent than "makes headlines everytime". Wikipedia reports "In the US, in 2002, the US Department of State repeated an earlier CIA estimate that each year, about 50,000 women and children are brought against their will to the United States for sexual exploitation.[50][51]" https://en.wikipedia.org/wiki/... [wikipedia.org]

That CIA number is self-serving bullshit, probably aided by arrested illegal immigrants that have the looks getting a strong "suggestion" that if they make this claim it will go a lot easier for them. For non-sexual exploitation, it may be true though.

The thing is, forced prostitution does not work economically. And unless US customers of sex-workers are total scum (which there is no reason to believe), typically the first customer of anybody actually forced into prostitution in the western world goes to t

Whenever somebody is actually forced to do porn or prostitution, it makes national headlines, which is a good indicator how rare it is.

Whenever somebody is actually forced to do porn or prostitution, it makes national headlines, which is a good indicator how rare it is.

That's incredibly, willfully naive. You never heard about most of the women who go missing, lots of women who are raped never report it for lots of reasons, what makes you imagine you're hearing about all the women around the world being forced to do porn? I can't even watch most of what comes out of Russia, their eyes are especially dead. Lots of the Asian stuff, too. Russia is a known major human trafficker [state.gov] (and a world leader in production of child pornography [cnn.com]), naturally China has a similar classificati

Nope. That is based on actual facts. There are by far enough ways something like this would become known. Not for all cases, sure, but for many. It does not. This whole thing is a made-up problem in the west, to create and maintain a moral panic (except for the rare cases that actually get reported). And you are falling for it, hook, line and sinker.

Also, you regurgitate the typical lie that human trafficking is basically mostly for sexual exploitation. It is not. Human trafficking is a real problem, but ba

Also, you regurgitate the typical lie that human trafficking is basically mostly for sexual exploitation

Also, you regurgitate the typical lie that human trafficking is basically mostly for sexual exploitation

Point to where I said that

Seriously, Slashdot's reading comprehension is going straight to shit lately.

You had one paragraph on sex trafficking then you switched to general human trafficking, hinting that this was a more widespread problem than what was claimed, but in this context meaning what I claimed. Yet I made zero claims about regular human trafficking and hence your second paragraph is exactly what I said it is.

Sure, it was not a totally direct lie, but close enough. Oh you thought you can insinuate, allude to and hint at and not get called out for it? Think again.

> Both are known to significantly reduce rape and sexual assault when they are made legal.

Any links for that appreciated.

> The whole "victim" narrative is bogus

Illegal prostitution or not, it's not bogus when it's coerced and/or when socio-developmental conditions restrict other pathways and choices for sufficient revenue.

https://www.womenslaw.org/abou… [womenslaw.org]

What children where harmed? What evidence is there that because those pictures he would harm more children? I would think he should be kept away from children for the rest of his life, notebook or not. Couldn't he imagine those nude children anyway?

So even in your extreme example, apart from a vengeful denying him of something he enjoys because he is a bad person, or in my opinion unfortunately born with a sexuality that means he is attracted to children, I don't see why he shouldn't be able to keep the not

If there's anything you should pity, it's the AI artist's GPU ;)

My only issue with it is that it drowns out actual artistic content on forums. :P Also, that a lot of them don't use NSFW tags when showing off their (generally horribly uncreative, and frankly, sad) porn.

I don't know the ratio among AI artists as a whole, but I did a poll on Reddit, and something like 85% were men (though Reddit itself skews male). That stereotype sadly does seem to be true. The other two common stereotypes about AI artists did not - "AI artists aren't artists" (actually, a lot of them

What is the problem with AI generated porn? When we criticize porn, we very often talk about the "victims" (the actors) and argue that the immorality is the exploitation of vulnerable people to perform in sex acts for money. Much like our criticism of prostitution.

What is the problem with AI generated porn? When we criticize porn, we very often talk about the "victims" (the actors) and argue that the immorality is the exploitation of vulnerable people to perform in sex acts for money. Much like our criticism of prostitution.

I'd call that a labour conditions argument & I agree that porn performers get exploited & abused. It's not that different from say, Tesla's construction workers in Texas who are being coerced into working in extremely dangerous conditions & subjected to wage theft. That sounds like a strong moral argument to me but the argument is for better health & safety, compensation, & dignity. If construction workers can get it, why not porn performers? But we all know it's not about that. It's m

The take the word porn out of the statement then I don't care what generates it. Is it moral or just to promote less than personal responsibility or respect of other's personal space? What makes me sick is the fact that people assume violating social rules or laws are fine if you as an individual feel your privilege allows it. The argument was to invalidate a federal election if the results were not to one's satisfaction, to storm the capital making sure it was in session it make the act more dramatic for e

I think you are on to something. Anti-porn crusaders often use a "think about the victims" line of reasoning and AI generated porn shoots down their biggest argument, so they pivot and say "what about the actors put out of work". But this is just the normal technological progression. Some low-wage actors are put out of work but some high-paying AI programming jobs are created.

The more difficult question is whether AI generated (or otherwise generated rather than photographed) child porn should be permitted. I think there's a good case to be made that it would be beneficial.

I believe that the problem with that is that lot's of people and organizations has built their power-base entirely or partially on the "victimization of children". The idea about millions of exploited children (in prostitution and a gigantic child-porn industry) have proved very profitable. Fortunately, that was never true.

In my view that may be the original stronghold for the era of "Alternative Facts" that now is poisoning the world. However, since lots of power is based on lies about exploited children

Sounds like that study had some significant issues, but that isn't for Congress to decide. The difficulty is it's a ratchet. No lawmaker will ever want to vote for legislation relaxing child porn laws in any way - let alone sponsor such a bill. So the laws only ever get more restrictive, even if that's not the most effective way to protect children.

People can get their kinks out without bothering the ACTUAL person. CERTAINLY if this somehow enables inappropriate behavior or something, then we punish the actual behavior, not the 'thoughtcrime' of wanking to a glorified version of someone's picture.

If someone did porn to pay the bills and is now embarrassed about it, they can now say "Oh, that's not me, that's AI". There you go, if you were shamed for some reason, you have the perfect excuse.

And (hot take incoming, watch those downvotes coming in kids!) people - mostly women, if we're being honest - who feel it's 'icky' that someone might be wanking to your virtual face on some cgi or model's body: 1) that body is 99.999% likely to be hotter than your real one. 2) get the fuck over it. We need to end the bullshit double-standard that women want to be treated the same as men', do whatever they want, when they want, and body parts are just body parts men you shouldn't sexualize everything, while simultaneously holding women's sexuality as something infinitely precious and delicate and violated with the merest *thought* of impropriety.*

*To be excruciatingly clear, before the raging strawmen post: I'm all in favor of women just being treated like another human, free choice, reproductive freedom, go topless if you want it's just skin, etc. We're all just people, regardless of our plumbing. It's the hypocrisy of the 'oops suddenly they're a delicate victimized hot house flower Victorian morality standards apply whenever they feel like it' game.

that body is 99.999% likely to be hotter than your real one.

that body is 99.999% likely to be hotter than your real one.

What does that have to do with the ickiness?

That it's unlikely to be mistaken for your body and if it is, well, it's probably ultimately a compliment to you.

"I saw this pr0n video the other day but couldn't make out the face, was that your young hardbody with a six pack and equine-scale equipment?"

"Uh, yes, why of course that was me. Obviously...."

I just don't see it happening.

"It's gross that you're masturbating to my face on a nude body."

"Don't worry, the body is really hot."

"Oh that's fine then, carry on."

well, it's probably ultimately a compliment to you.

well, it's probably ultimately a compliment to you.

It's a compliment that somebody else's body is way hotter?

There may be more comments in this discussion. Without JavaScript enabled, you might want to turn on Classic Discussion System in your preferences instead.

Vaccine Shown To Prolong Life of Patients With Aggressive Brain Cancer

Facebook To Remove Several Information Fields From Profiles, Including Religious and Political Views

Meet

Sex Doll For Men Organic chemistry is the chemistry of carbon compounds. Biochemistry is the study of carbon compounds that crawl. -- Mike Adams