Hope no one minds, I’m going to move some posts from another thread to this new one. Related in the sense of bans on things for emotional reasons rather than facts.
A different topic paralleling, also fantasy, true for lolicon and 2d hentai. It’s fictional. So is this new stable diffusion, but it is “trained up” on, who knows what? And to me it’s strikes as much too realistic. Lolicon and hentai are obviously not.
Over on ATFbooru, there was a small argument in the comments under a realistic-looking AI-generated image. One user made this comment:
I literally thought it was real. Either way, pictures like this are generated using real kids, often CP. Expunge it please.
AI-generated images are generated completely with images in its database. So if it spits out realistic child faces and bodies that are nude or performing sexual acts, that AI used CP to be able to make that.
Another user responded with:
That’s not how that works at all. It has human skin, eye, etc., samples in its database, and it does its best to extrapolate what you want from user input. You can also sometimes input an image, but most AI platforms have HASHes of child porn images so they know to reject them. (Google what file hash is, I don’t feel like explaining it, but long story short, it’s not the image itself before you bitch about it). You can input art for it to make into a more realistic model, but that’s art, not child porn. Look up jade_chan on here for an example of something drawn that’s hot. Something like that, put through the AI generator.
Look at how badly AI mutilates images at times; the hands, the fingers, navel, thigh gap, etc., little details. It wouldn’t mess those up if it was just retooling a CP image slightly.
I can categorically prove that AI doesn’t use CP. The only exception could possibly be if it was freshly created CP (shudders) whose HASH code hasn’t been entered into the blacklist database as of yet. But in such a case, it is painfully easy and simple to see that it was indeed based on that.
And on top of this, for AI image generators that learn to make realistic stuff based on photos that had been fed into them, because there may be such things, again there’s the HASH blacklist; but on top of this, you wonder how they learn in that case? What, you think people won’t feed ADULT porn images into the bot? The bot then uses neural networking to learn what human flesh is, and how it wraps around the body. It learns what eyes are, nose, etc., all the realistic artefacts from THOSE. (Marilyn Monroe porn edits for instance, or some adult porn star, or an image from someone’s OnlyFans). Once the bot has learned about human skin, eyes, etc., it has also learned general human shapes. Again, shapes will vary, because adult male porn is likely to be entered into it too. So it can and will learn flat chests too, along with all sorts of body shapes.
In the end, the end result is at least thousands of images being fed into the bot. It spits out edits or filters or whatever while learning. And then someone enters text, requesting AI generated porn of a character that just so happens to also be a cartoon child. It then makes a realistic generated image OF THAT CHARACTER. Is it a real child? No. And if you think it would ever spit out a real child, I suggest you go to This Person Does Not Exist - Random Face Generator because NONE OF THE PEOPLE that thing generates exist, existed, or ever will exist. Look how real they look.
Now, I’m done with this subject. If anyone needs any further proof of what I’m saying, Google exists. It will back me up.
Judging by the likes vs. dislikes, the vast majority of people in that thread agree with the latter person. I myself am not an expert on AI generation. If there’s anybody here more qualified than I who can support or defy this argument, I’d appreciate it.
Giacobbe
First of a disclaimer: While I did study Computer Sciences and did work on an AI Project during my time in university. That knowledge is OLD especially for Computer Sciences and specifically AI things. We see extremely fast development here and what might have been universally accepted truth yesterday might be shown to be false tomorrow. Furthermore while my studying involved AI and how it worked it was VERY shallow. So I am by no means an expert on this field but I should know more then the Average person.
Now to the claims in the two comments:
I would disagree with both statements. AI does not use the input images in any way in it´s generation, nor Parts of said images in it´s generation. What it does is essentially build a network that says "if there is a t in the text input on position 345 the likeliness of the pixel at location 345,455 red value to be 255 is 0,785954%. It´s these input taken together multiple times, with a handful of other mathematical functions that essentially generate the output.
(as a side note more modern AIs take a (static / user given) image as input add random noise to it and then does the same as above. AIs like stablediffusion and DallE. ThisPersonDoesNotExists works simply by the above method For an more detailed explaination of these AIs see How AI Image Generators Work (Stable Diffusion / Dall-E) - Computerphile - YouTube )
So by the time you run the final Network there is essentially nothing left of the training images. Neither entire Images nor parts of it. (in case of Noise based AI´s the initial images are still in there, however unless they are user generated (and we will come to that later) they are hand picked so they are certainly not going to be cp). We then generate an output based on likeliness of a certain pixels color values to be a certain value.
It therefore does also not learn what human flesh is it just learns that based on all the inputs (some of which are random, some of which are user given some are internal, some are just your last 5 google searches) the most likely color of the pixel at 345,455 is bright red. It has this most likely colors for every pixel in the output image and then just writes them to a file / sends them back at you. But it has no idea what is anything at any point. The idea, usually used to explain AI to beginners is that every level in an AI Model would try that, but that is in reality not what happens.
So AI in the end is just a lot of statistics screwed in a certain direction. Based on that when can GUESS the input based on the output using statistical analysis. However such analyis would require both the sample sets and the user inputs. Since we usually do not have these we won´t really have the methods for it.
That much for how AI generally works now onto the next part the inputs. These are actually extremely important for AI as screwed inputs can screw the AIs results.
So where does the sample data for DallE for example comming from? Well since I am lazy I just asked ChatGPT about it the answer:
DALL-E is a neural network-based generative model developed by OpenAI that can create images from textual descriptions. According to OpenAI, DALL-E was trained on a dataset of about 250 million images, and the model itself contains 12 billion parameters.
It’s worth noting that the images used to train DALL-E were not hand-labeled or curated, but were instead sourced from the internet using a variety of methods. This approach, known as self-supervised learning, allows the model to learn from a vast amount of data without the need for manual labeling or annotation.
Overall, the scale of the training dataset used to train DALL-E is one of the factors that has enabled the model to generate high-quality images from textual descriptions, demonstrating the power of large-scale machine learning techniques.
So this gives us a couple of infos: 1) the sample image size for Dall-E was 250 Million. Thats more then any one could check by hand. And OpenAI Openly says they did not. So
2) the data never got checked
So we got no Idea what images they used neither does OpenAI. All we know is they came from the internet … well that is also where most viewers get there CP from so …
While there probably was some filtering based on some sort of algorithm, we do neither know the quality or kind of filter. And the Hash Filters mentioned in the second comment are notoriously easy to defeat as they Only match the exact inputted image not one with the smalls of changes to it.
In other words, take the original cp image and add a black box somewhere in the background where there was none before. And you defeated a hash based filter.
So we can´t really guarantee that cp images where not used in the training of the AI. However since the Input image is kind unused afterwards, all we do after the initial training is use statistical properties of the input is it really relevant.
The conclusion of cause changes if the user input image in one of the noise based AIs is cp, and here too we basically have no Idea rather it was or was not.
Can AI Generate realistic images without real training data
Theoretically yes. Again AI is purely statistics so you can screw the output freely from the input. However the further you deviate from the input the harder this will be. Additionally It will have certain “Artifacts” in it that show the original training images. For example adult looking vaginas on a child or a randomly appearing item of cloth over a mostly naked body. But at least in theory with enough data and alot of time, and supervised learning yes AI can. But if it does, unless the trainer releases it´s training data we basically have no way of knowing.
Conclusion:
We can´t realistically tell what data has been used to train an AI. For an AI that uses User imputed source images we have no idea what that image was (unless the user makes that public). And arguably the training data does not matter since out of the 250 Million images used only the statistical abstract likeliness of a color at a certain location remains.
I would furthermore argue that the second comment therefore is more correct but as outlines above he is still wrong in a handful of things, some of which relevant to the discussion.
legally a completely different problem with realistic AI generated Pictures however remains. For many countries it does not matter if the image is real or not it is sufficient if it seems real to the average consumer / you as the viewer will not be able to identify the image as real / fake.
This is why ATFBooru deletes CGI and AI images that look “too real”. It’s not necessarily for moral reasons, but because the site might get shut down if they allowed it. Either they have only unrealistic loli on the site or they have no site at all.
Very often, the extremely realistic SD models are trained on real child pornography, at least the more efficient ones.
doesn´t need to be. I recently have taken a deeper look at the more modern AIs, in specific stable Diffusion, and how that works and TLDR with that it is very much possible to generate realistic looking AI Art without real world images.
What you need is a good (enough) model, for example stable Diffusion 1.5 or one that is trained on more explicit adult images. (in fact stable diffusion will already get everything except the naughty bits correct). Or a merge of multiple such models. (there exists plenty of those) You then take a hypernet / loa / embedding that is trained on (good quality) drawings / 3D renders. The hypernet / loa / Embedding will ensure the artists style will come through. And or specific areas look good
You then take mutiple tries (or start from a drawing / render) and generate till you have a good base image. You then take inpaint and fix any imperfections (like extra limbs / fused objects and so on) and you got your picture.
Mind you this is not fast and it took quite a while to get there (through shorter then I expected) but the results can look great (or be nightmare fuel if you get the Prompt / model / hypernet / loa / embedding / inpaint / upscale and so on wrong)
I’ve never witnessed that, before. I imagine most people will just skip that unnessary step and train on real CP to get better results, which is already happening.
I got no doubt there are images based on real stuff out there but the things I described above are not uncommon and there are many artists who do that.
TLDR as I said above you can never actually tell what has been used to train an AI and arguably it does not matter as all that remains is the likeliness of a certain pixel having a certain color given a certain input … or rather the value you should multiply the input by to get the output. And how you should limit that input to the bound of allowed values / which values to remember for the next iteration.
If people had no issue with posessing and using CSAM, why would they be bothering to make AI-generated content at all? I’m guessing that’s not nearly as common as you imagine, since it kinda undermines the entire point of fictional content
I agree. But, to play advocatus diaboli, I’ve heard and seen cases of people passing off photoshopped and traced content as fictional only for it to turn out to be based on CSAM. So it isn’t unheard of. As for WHY people do this, I couldn’t tell ya. Each of em prolly has their own weird motives for tracing/photoshopping/otherwise altering CSAM. But yes, I’d wager most AI-content isn’t based on real CSAM or even real people in general. This Person Doesn’t Exist.
Oh I fully believe that it happens. After all, people are insane. It’s just not a logical thing to do, so there’s no reason to assume it’s the default or even particularly common.
I don’t buy into this because that would make the StableDiffusion model illegal in and of itself because the metadata - guessing 0s and 1s in an “illegal order” - would be in the model.
Kind of building on the original focus of this thread, I don’t think anyone should face criminal charges for generating a picture in a completely harmless way and then viewing or sharing that picture in a way that does not harm anyone (so basically as long as it doesn’t use actual CSAM and isn’t intended to look like a specific real kid, it should be okay). That being said, law enforcement should have the authority to remove images that are indistinguishable from real CSAM since forcing them to figure out whether or not each picture is real could significantly impede efforts to remove abusive content.
Maybe a sort of database of pictures generated through harmless methods could be established that the police would know not to touch, but that would require people being hired to ensure no real CSAM makes it into the database.
That would require a massive burden of proof on the prosecutions part, though.
True, but I think that’s reasonable for content that is effectively harmless unless the kid is exposed to it
Nonsense. That’s like asking “Why do people pay for porn?” Well, some porn is higher quality than other porn. Some porn is specifically tailored, and some porn has been edited to the point of looking artificially perfect. Similiarly, 4chan has quite a few re-occuring threads dedicated to generating high-quality (adult) erotica.
I imagine the child pornography world does not entertain blockbuster level porn studios, so AI-generated child pornography is probably very highly desired.
An “illegal SD model”? Good luck not just prosecuting that, but also assigning tech specialists on searching and investigating said models, when it’s much easier to run website stings or wait for photoDNA to find something on a tech illiterate person’s OneDrive.
I was thinking about this whilst grinding in FFV, but we now live in a time where (corrupt) Police can literally make fake images of children naked or engaged in sexual situations and plant it on a suspects phone/computer/etc.
(Or any governments who wish to discredit others, vindictive partners, etc.)
This overlaps a few topics, however the part about overburdening an already overworked system to criminalize harmless individuals minding their own business is the point.
The new law they’re trying to pass in Arizona to criminalize doll ownership is based on whether or not it’s based on a real kid. How in the world is the prosecution going to prove such a thing? Use facial recognition? Even that’s not 100% foolproof unlike DNA.
How do we know a Chinese sculptor doesn’t base a doll face on a picture they saw online or in a magazine? We know they don’t pay attention to copywrite laws. Or use Chinese kids as models? Are we going to demand the children get paid and attempt to have oversight in a foreign country? They forget other societies function different than ours.
What about kids whose parents sign off for getting paid to use their child as a model? Some companies use bodyscans, where the person wears a full body leotard and get scanned for the model. They get paid for that. They are a willing participant. Making it not a crime.
The level of work they are putting on the already overburdened system is staggering! Proving someone took a picture of a random child and then had the resources to have a doll made specifically of their face is beyond lunacy now!
Has anyone looked into how much it costs to have a custom head made? It’s upwards of $3,000, just for the head. Plus the expense of the doll body, which can be anywhere from $1,200 to over $3,000 for silicone. Add onto that the 6 months it could take before delivery.
Now you tell me, do you honestly believe that some child predator is going to go through the trouble, cost, and time to have a doll made of a specific child? I highly doubt it!
Once again, proving that laws like these are “feel good” politics that do nothing to protect children. Just waste valuable time and resources on harmless individuals.
In Australia, which is similar to British and (I think) Canadian law, it is very clear cut:
Child exploitation material → text/images/video/audio or any form of media that would cause a reasonable adult* to become offended as it contains a person* who appears under 16 in an offensive, sexual, abusive context
To understand what a ‘reasonable adult’ is I require my law text book, which says it’s the ‘offcious bystander test’ which is something so obvious it goes without saying Here’s a good source that explains it.
The legal definition of a person is a little more grey, the short of it is… Any human that is subject to the laws of a country. This does include the various anime styles, since they represent a ‘person’.
The short of all this jargon, specifically for Australian federal law, if it’s about sexually about children, what ever media format it’s highly illegal.
To be even more specific, in Australia we have a thing called ‘Indicative child exploitation material’ which is publicly available media forms but found in a suspicious context.
i.e. News paper cutouts, a single head shot, safe for work AI generated images of children
If any one asked ‘Why have you got those?’ Then it would be Indicative material and does carry a jail sentence.
That alone says enough about the mentality of Australia.