AI-Generated CP

[TooLittleTime]

I'm not an attorney and this is not legal advice, but I am pretty sure some jurisdictions in the U.S. have 'Representation of' type laws that can get you in trouble. My usual blah blah about not wanting to go through that even if you prevail is still true.

And an AI generated image that would create probable cause for a full blown search warrant would be disastrous for most people who collected sexual AI images of children. I know someone will respond 'that's all I collect', but for the rest, the warrants premise would be correct. where there is AI images there are likely real images.

[Shatterhand]

I'm not an attorney and this is not legal advice, but I am pretty sure some jurisdictions in the U.S. have 'Representation of' type laws that can get you in trouble. My usual blah blah about not wanting to go through that even if you prevail is still true.

And an AI generated image that would create probable cause for a full blown search warrant would be disastrous for most people who collected sexual AI images of children. I know someone will respond 'that's all I collect', but for the rest, the warrants premise would be correct. where there is AI images there are likely real images.

I agree with you in some ways.  I think the way local law enforcement gets around the vagueness of these laws especially the whole constitutional right to FREEDOM OF SPEECH AND EXPRESSION, is to look at the local and state obscenity laws.  At least that's what I have heard in a few stories about this issue.

[OneLove]

Images of underage kids does not fall into any freedom of speech or expression category, legally speaking. The same goes for nudism/naturism. It is technically legal to have images of naked kids in non-sexualized poses, ie making a sand castle at the beach, but if LEA finds your stash of images, you're in big trouble. Even the NN sites have been raided and shut down.

[olderman]

If you are unlucky enough to live in a country that has "objectionable material" laws then you are pretty much screwed with anything even remotely related to pedophilia, doesn't matter how it was created. However if you are the one doing the creating, you are doubly screwed, or maybe entirely screwed is a better way of putting it.

Comes down to the judge and how they interpret what it is you are in possession of.

If it were a bunch of non nude AI images I would say you would possibly get away with it, but you would forever be on the radar of LEA. So I say treat anything child related as potentially damaging to your freedom.

[OneLove]

If it were a bunch of non nude AI images I would say you would possibly get away with it, but you would forever be on the radar of LEA. So I say treat anything child related as potentially damaging to your freedom.

Even NN photos are illegal if the child is posed in a sexually alluring manner, ie, legs spread open looking at the camera with "those eyes." Why do you think all those NN sites got taken over and shut down by LE?

[Shady]

I once got access to an AI image generator that generate lolis and shotas. But it was a while back and I don't exactly remember the page's name anymore. It's not a pity though because I honestly don't fancy AI images much. I'm much more into things that are authentic, with soul and variety, than a predictable picture of a slippery balloon-ish child.

[hö]

Okay, so this is a bit of a story for me. As you might know I quit CP entirely a long long time ago. And when generative AI became available to the wider public it was only natural for AI CP to spawn - so I thought to myself; "well, it ain't real girls, so it'd be fine by my morals". I gave it a go and soon found that site OTR mentioned (at least I think it's that same site lol).

Turns out most people who create CP with AI don't put much effort into making it look real. Wax skin and balloons everywhere. You know, there are ways to edit AI images of people into looking like actual photography but somehow nobody does it. Probably because it's a shitload of work. So it did look like a better version of the old-days 3D CP, which never did it for me.

Other than the looks of it the pure knowledge that this wasn't "the real stuff" was a major one. So no, AI-generated CP won't do it for me. I do know but it's that "if you do it at least get a hold of the real thing" idea that also applies to many other things in my life. It's like the crappy Chinese rip-off version vs. the genuine original product - what are you picking? I'm going with the original product because I don't like fake stuff.

Buuuut, since I quit CP for moral reasons and don't like the Chinaman product I still don't wank in front of a screen

[Lillab]

I agree that kawaii is interesting, but there is very little you can do with it. I find it gets old pretty fast, but it is awesome they have done so well. CP AI is always going to be like 6 months behind other AI because it needs to be done on open source tech. And although, as OTR pointed, it triggers uncanny valley worse than CG, I like looking at both. Once I get my own stuff set up, I was planning on getting the AI to instead mimic my favorite CG artists. I assume the site people keep mentioning is xDArt, and yes that does have some great resources that I am excited to get to.

I haven't started yet because I am trying to work through some more important projects first, mostly around improving my OpSec. I have been going through and making sure everything I own is encrypted. I just finished getting my Windows encrypted (that was stressing me out because I hadn't been very careful). I am also almost finished setting up Tails and WHONIX. Once I finish setting up that environment, I plan on getting all sorts of tools and libraries installed. I am pretty excited about what is possible. I assume within a year from now I will be able to feed my favorite stories into a model and get a movie. I am pretty close to that with the current best models, but those you can't do porn or kids, let alone kid porn. I just need to try to apply it to the CP models and see what I can do. Within 2 years I should be able to make photo-realistic Hollywood-quality CP, which I am going to enjoy so much more than real CP. I hope the demand for authentic CP dies off.

[outlier]

Yeah, I quickly grew bored with kawaii. But the comments about slippery balloon-ish images made me laugh. You just haven't had an opportunity to play with the correct AI model. I've spent the past couple months scouring xDArt, Hugging Face, Civitai and elsewhere learning how to use ComfyUI with various models, checkpoints and LoRAs. I even toyed with training my own LoRA with some of the images I have saved. The custom LoRA thing didn't pan out. And even though the stuff I was able to generate was better than kawaii, it quickly started all looking the same.

Then I came across a clearnet site called perchance, and again I was blown away. At first I thought maybe it was somehow cheating and returning real photos from an extensively tagged database. But then I noticed there is an occasional extra finger or leg in some of the images. Not only does perchance generate realistic photos, but it generates such a variety of poses, ethnicities, faces, body types and environments, all from a single prompt, that it can hold my attention for hours. The perchance developer is pretty tight-lipped about the model used for his casual photo plugin, simply saying it's just one of the top-rated models on Civitai and Hugging Face. Someone on a forum suggested it is a Chroma model. I looked it up. Chroma is an open source model with a stated goal of NO censorship of the images or captions used to train it.

Long story short; I'm using some chroma-unchained-v50-flash-heun-Q8-blah-blah model to generate my own images that are almost on par with perchance. But I'm struggling to match the variety produced by it. My attempts to tweak the workflow to weaken prompt adherence and get more variety often result in increased artifacts or other unwanted effects.

Look up perchance and try out the "casual photo" style. I'm curious to know if anyone here still hits that uncanny valley with it. It's clearnet and requires javascript, so I wouldn't recommend prompting anything too explicit and young. That said, I found that it's easy to work around the prompt filter by just using less common synonyms for blocked words. Be careful. If you evade the prompt filter, the model will oblige just about anything. And be patient with the queue. It will often error out, but it automatically retries and eventually rewards you with an image.

[Lillab]

I've spent the past couple months scouring xDArt, Hugging Face, Civitai and elsewhere learning how to use ComfyUI with various models, checkpoints and LoRAs. I even toyed with training my own LoRA with some of the images I have saved. The custom LoRA thing didn't pan out. And even though the stuff I was able to generate was better than kawaii, it quickly started all looking the same.
...
Long story short; I'm using some chroma-unchained-v50-flash-heun-Q8-blah-blah model to generate my own images that are almost on par with perchance. But I'm struggling to match the variety produced by it. My attempts to tweak the workflow to weaken prompt adherence and get more variety often result in increased artifacts or other unwanted effects.

Thanks for sharing your journey! This is quite encouraging. It does sound daunting to have to learn several new technologies to make this work. I just need to set the time aside, and start playing with ComfyUI. Eventually I want to get into training LoRAs, but one step at a time. I don't think I could get the chroma-unchained-v50-flash-heun-Q8-blah-blah model to run on my graphics card. Maybe the Q4 will work. But this is very useful information. Thanks! I think this is exactly the direction I want to go first. Now to see if I can set some time aside this week for this adventure.

Also, have you played at all with any voice models? I think I am just as excited about getting audio as I am anything visual.

Look up perchance and try out the "casual photo" style. I'm curious to know if anyone here still hits that uncanny valley with it. It's clearnet and requires javascript, so I wouldn't recommend prompting anything too explicit and young. That said, I found that it's easy to work around the prompt filter by just using less common synonyms for blocked words. Be careful. If you evade the prompt filter, the model will oblige just about anything. And be patient with the queue. It will often error out, but it automatically retries and eventually rewards you with an image.

I am not interested in using this, other than to try to break their filters. That sounds fun. Just need to boot into Whonix first. I've got way too many projects though, I might never get around to this. Does it make the images you generate public? I don't want to create any public panic from my experiments. I'd hate to get them shut down accidentally. I think the thing I enjoyed most on Kawaii was trying to get it to generate the most tame images I could. Trying to get that pornographic image generator to generate non-pornographic images. I'm so devious.

[OneLove]

As you might know I quit CP entirely a long long time ago.

I also disposed of my stash a few years ago, buy staying off cp sites is still an ongoing struggle. I feel like a drug addict who's generally successful at staying clean but occasionally falls off the wagon in times of weakness.

There's a word for people who have a disdain for artificial and technological things which I can't remember, but I'm one of those. Maybe I should have been Amish.

[outlier]

Thanks for sharing your journey! This is quite encouraging. It does sound daunting to have to learn several new technologies to make this work. I just need to set the time aside, and start playing with ComfyUI. Eventually I want to get into training LoRAs, but one step at a time. I don't think I could get the chroma-unchained-v50-flash-heun-Q8-blah-blah model to run on my graphics card. Maybe the Q4 will work. But this is very useful information. Thanks! I think this is exactly the direction I want to go first. Now to see if I can set some time aside this week for this adventure.

You're welcome. Q4 would probably be fine. My system had enough capacity that it can do Q8 with no memory swapping.

Also, have you played at all with any voice models? I think I am just as excited about getting audio as I am anything visual.

Nope. But I can see the pull of having a LGF to talk to.

I am not interested in using this, other than to try to break their filters. That sounds fun. Just need to boot into Whonix first. I've got way too many projects though, I might never get around to this. Does it make the images you generate public? I don't want to create any public panic from my experiments. I'd hate to get them shut down accidentally. I think the thing I enjoyed most on Kawaii was trying to get it to generate the most tame images I could. Trying to get that pornographic image generator to generate non-pornographic images. I'm so devious.

Perchance has a share button on every image generated, along with a warning about getting banned for sharing inappropriate content, so I don't think they are public by default.

It's kind of the wild west out there with censorship efforts. Some image gen sites simply block certain keywords (really easy to work around), while others have some sort of adverserial model that analyzes the final image before deciding if it's appropriate. I expect that in a year or so, there will be laws regulating how to apply censorship if you want to host an ai service. It's one of my motivations for figuring out how to do this on my own system.

[on the rocks]

I have the utmost confidence in the broader on-topic community's ability to forge ahead with uncensored, generative AI regardless of whatever laws get made.  Any law is no match for Rule 34.

[OneLove]

I have the utmost confidence in the broader on-topic community's ability to forge ahead with uncensored, generative AI regardless of whatever laws get made.  Any law is no match for Rule 34.

And All the Fallen if you're a gamer

[Arthur]

I have little interest myself. Interested to learn more about it maybe but I feel like its 'less than' but this may be my ignorance and I am up for being corrected.