• Sign up or login, and you'll have full access to opportunities of forum.

Discussion about A.I.

Go to CruxDreams.com
There's another good news for the prospect of open-source models. Mark Zuckerberg expressed his belief in the open-source tradition in this surprising interview.


I already mentioned how the free and open-source software(FOSS) movement achieved remarkable victory against corporations about 20 years ago and has transformed the whole IT industry ever since. But I think it wouldn't be too out-of-place to repeat it here since the real threat of AI could be its potential to revert such achievements, but non-programmers rarely recognise it.

In short, pretty much every online service or software you know is based on open-source technology to some degree. The advancement of IT technology in the past couple of decades was, in large part, a direct result of the victory that the FOSS community had achieved, as it allowed everyone free access to frameworks, libraries, or tools to build their own software - even such a program like Daz Studio is built using various FOSS libraries like Qt, for example.

But, the advancement of AI technology has opened up new possibilities for corporations to turn the tide and monopolise the IT industry once again. Training large AI models typically requires enormous time and effort, which only a few large companies can afford. And it doesn't take one to be an IT expert to predict how AI will start affecting and regulating ever more aspects of our lives in the foreseeable future. It could be the start of a dystopian reality if such power falls exclusively to a few select big corporations.

As such, I'm glad at least we have a few players in the AI field who profess their commitment to the FOSS tradition. I don't know if such companies will become a norm, but it's good to see that the hope is still alive.


P.S.: On a side note, a new version of SillyTavern was just released with built-in support for OpenRouter, which seems to have improved the output quality. Now, I get pretty amazing RP experiences using relatively inexpensive models. If you have a good PC or don't mind a subscription, I encourage you to try. Even free models can be quite interesting at times.
The problem I have with AI , is it takes away from the creative artist's creativity.
 
There are no eyelashes on the eyes, although this is possible. The length of the thumb of the right hand is wrong, the little finger is not visible. The left hand is deformed.
Epic realism.
It’s a demo image from the model, which means there’s no inpainting or post editing involved. It’d be nice if we can get a perfect photo like result without all the hassle needed to fix such defects, for sure.

Still, it’s not that difficult to fix, and you won’t get anywhere close to such an output in terms of realism without AI anyway, unless you’re a top-level professional 3D artist with a tons of free time to burn.

If you somehow can’t stand anything created with AI, you can simply ignore it and happily indulge in whatever traditional art that you see as more “real”, “human”, or whatever. I fortunately don’t have such a problem, so I can enjoy creating AI images that look so much more realistic than I had ever made with Blender or Daz3D.

The problem I have with AI , is it takes away from the creative artist's creativity.

Honestly, I don’t even understand what you mean by “creativity”. Say, for example, do you think Daz3D is a legitimate tool for expressing creativity? Then how much percentage of Daz3D users have created their own 3D props? Maybe 0.5%?

Vast majority of Daz3D users just purchase 3D models off the shelf and lay them out in space, then use some HDRI background they downloaded from the internet and hit the render button. Want to depict torn clothes? Buy it from the store. Need a whip prop? Again, buy a model someone else made. If you can’t find what you want from the store? Then sorry, you’re out of luck.

How come this can be “freer” in terms of creativity than AI art, which doesn’t have such limitations?

To be clear, I do think Daz3D is a perfectly legitimate tool for artistic expression because you don’t have to make everything from scratch to make art. But if you are going to argue AI isn’t real art because there’s some aspect of it you don’t control, you have to accept that Daz3D is even more “inferior” medium for art, because you don’t have half the freedom and control with Daz3D that I have with AI tools.

Can you imagine an android with porcelain skin? If you can easily depict such an idea into an image, that’s what we call “creative freedom.”

1709312091906.jpg
(Image source)

But if you can’t do it because you can’t find anything like that on the prop store?

And that’s what I would call “restricting creativity.”
 
It’s a demo image from the model, which means there’s no inpainting or post editing involved. It’d be nice if we can get a perfect photo like result without all the hassle needed to fix such defects, for sure.

Still, it’s not that difficult to fix, and you won’t get anywhere close to such an output in terms of realism without AI anyway, unless you’re a top-level professional 3D artist with a tons of free time to burn.

If you somehow can’t stand anything created with AI, you can simply ignore it and happily indulge in whatever traditional art that you see as more “real”, “human”, or whatever. I fortunately don’t have such a problem, so I can enjoy creating AI images that look so much more realistic than I had ever made with Blender or Daz3D.
I'm afraid that with this position of her fingers you won't have time to edit anything; what's in her hands will fall to the ground in 1 second. The image is absurd from the very beginning.
I work with Dalle-E3/ChatGPT, and he copes with this much more successfully.
Stable Diffusion is good at drawing only faces or abstract worlds
 
I'm afraid that with this position of her fingers you won't have time to edit anything; what's in her hands will fall to the ground in 1 second. The image is absurd from the very beginning.
I work with Dalle-E3/ChatGPT, and he copes with this much more successfully.
Stable Diffusion is good at drawing only faces or abstract worlds
I don’t have any objection to admit that those closed source models are better than SD, which I mentioned myself in this thread.

But to say that SD is only good for drawing faces? Just because it often struggles to depict fingers in one shot? When there are countless SD renders that depict hands which are virtually impossible to tell from real photos?

Well, if your standards are so high that you can’t stand any kind of imperfection then so be it. You’re free to have your preferences, although I find it rather odd to see it in an erotic art community.

I’m not sure if we have any medium of art that would satisfy your high standards, considering how Dall-e doesn’t allow generating any sort of NSFW images, and all the other tools - Daz3D, photo-manipulation, or whatever tools people on this site use - fall far short of SD in their capability of producing photorealistic results.
 
I don’t have any objection to admit that those closed source models are better than SD, which I mentioned myself in this thread.

But to say that SD is only good for drawing faces? Just because it often struggles to depict fingers in one shot? When there are countless SD renders that depict hands which are virtually impossible to tell from real photos?

Well, if your standards are so high that you can’t stand any kind of imperfection then so be it. You’re free to have your preferences, although I find it rather odd to see it in an erotic art community.

I’m not sure if we have any medium of art that would satisfy your high standards, considering how Dall-e doesn’t allow generating any sort of NSFW images, and all the other tools - Daz3D, photo-manipulation, or whatever tools people on this site use - fall far short of SD in their capability of producing photorealistic results.
If you like to restore missing fingers or deformed eyes on naked models, this is your choice. And yes, Dalle can generate NSFW.
 
And yes, Dalle can generate NSFW.
Really? That’s great. Try generating something like this then:


It took me two days to make it look exactly as I like. But I’m sure you can do it much faster since it apparently can generate a perfect image at one shot.

You will do everyone posting AI images here a great service if you can share your superior work, and maybe also how you managed to trick Dall-e to generate NSFW images without getting you banned.
 
It took me two days to make it look exactly as I like. But I’m sure you can do it much faster since it apparently can generate a perfect image at one shot.
I’m not a fan of looking at such images in detail, and I don’t want to say that it’s bad, but there are also mistakes there, the nails on the left hand are bad. Perhaps there are more of them.
When SD can able to generate a picture for you, for example, that someone is eating at the table and is holding a fork and knife correctly in their hands,or handcuffs,for example, we can talk.
maybe also how you managed to trick Dall-e to generate NSFW images without getting you banned.
This can be found on Google, and Microsoft accounts are free to register.
 
Last edited:
When SD can able to generate a picture for you, for example, that someone is eating at the table and is holding a fork and knife correctly in their hands,or handcuffs,for example, we can talk.

dinner.jpg handcuffs.jpg

Fine. Let's talk then. Before we start, please save your trouble of telling me how this or that part of those images is imperfect because I know. I just didn't want to spend a good part of a Saturday morning just to prove my point.

I believe now it's your turn to actually prove what you claimed instead of just nitpicking my works. I'm very much interested in seeing what you'll produce as proof that Dall-e can generate NSFW images because you are pretty much the only person I know who claimed that. Give me a simple image of a naked girl with her nipples and private part, then we can talk.

I'll be waiting.
 
Last edited:
Strangely to me, it looks like there's only minimal interest in AI RP so far, although we have a growing number of people trying out Stable Diffusion now. If we happen to more people showing interest in the subject, I may try to write a simple guide to set up Silly Tavern for kinky RP.

FYI, I'm interested :)
 
FYI, I'm interested :)

I was almost finished writing a pretty long guide on how to get started when my browser froze so I had to restart it. :doh:

So, I'm afraid I'll have to be a bit more brief this time. Feel free to ask me if you need more details or clarification.

To enjoy AI-based role-playing, you'll need a client and a model. For the former, I recommend SillyTavern, which is feature-rich but also quite lightweight, so it can run on a low-end PC, or even on a mobile device.

Running an LLM model, however, is a different matter. You'll need top-end hardware to be able to run a decent model locally. So, unless you have a $5,000 PC at home, you'd probably want to go for one of the options below:
  • Kobold AI Horde - It's a free cloud run by those who donate their GPU time, so it can be hit or miss. If you consider this route, don't forget to register and get an API key, which will get your requests higher priority so they can get served faster.
  • Paid LLM service provider - You can use one of the commercial LLM services like OpenRouter or Mancer to run your model. Personally, I use OpenRouter as I can get a decent model for a very low price, like $0.000462 per request. They also provide a few free 7B models, which are great for testing things.
  • General cloud provider - You can also rent a GPU host from a cloud provider like RunPod, instead. In case of RunPod, it's not as complicated as it may sound since they provide a "template" which you can choose to get the server ready to run an LLM model. Compared with the previous option, it involves a bit more hassle in return for complete control over the server and more privacy.
I'd recommend creating an account on OpenRouter and using one of their free models to test your SillyTavern settings (i.e. character cards, world info entries, etc.) before considering a more dedicated option. In case your graphics card has 10Gb or more VRAM, you can run a good 7B model like OpenHermes 2.5 Mistral 7B locally with a decent speed, without having to use an online service like OR.

To get the best experience out of it, you'll need good character cards and optional world info entries on SillyTavern. As such, I'd recommend to spend some time to read the relevant section of their documentation and learn how to create them.

When you're done with the settings, try a larger model like 13B, or 70B and see which one suits your taste the best. Avoid those from OpenAI or Google since they won't allow any NSFW content. I found Mixtral 8x7B variants have a good balance between quality & price. Goliath 120B, on the other hand, is relatively expensive but it really excels with the right settings. So you may want to try it if you're willing to spend some money.

Hope this helps! :)
 
By the way, below is a dialogue history from a text-based slavery game I've been working on lately. It's not generated with SillyTavern, but I used the same Nous Hermes 2 Mixtral 8x7B DPO model I use on OpenRouter. So, you can see it as an example of NSFW content you can expect from a model of similar size (8x7B).

Zakhar's grip on your hair tightens, causing you to yelp in pain. He leans down, his lips brushing against your ear, making your skin crawl. "Don't worry, I'll go easy on you this time," he whispers, the malice in his voice unmistakable. He releases your hair and pushes you onto your back, forcing your knees apart. You feel a surge of humiliation and terror as you realize what he's about to do.

"Please, sir, I'll do anything you want, just don't hurt me," you plead, your voice barely more than a whisper. "I'll be your obedient slave, I promise. Please, don't do this." You can feel tears streaming down your cheeks as you struggle to hold back your sobs. Despite your fear, you know that submitting to Zakhar's twisted desires might be the only way to save yourself from further pain. You lay there, vulnerable and exposed, offering yourself to him in the hopes that you'll escape this brutal punishment.

Zakhar's hand slides between your trembling thighs, his touch cold and impersonal. You let out a strangled whimper as he reaches your most intimate area, his fingers roughly exploring your vulnerability. "You're nothing but a filthy whore," he sneers, his voice laced with contempt. He pushes two fingers inside you, causing you to gasp in pain. "This is the only lesson you need to learn."

"Please, sir, I'm begging you. I'll be your perfect slave, I promise. I'll do anything you want, just don't hurt me,"
you plead, your voice quivering with fear. You can feel your heart pounding in your chest as Zakhar continues to violate you. "I'll be the best slave you've ever had, just give me a chance, please." You close your eyes, tears streaming down your face as you desperately cling to the hope that your submission might save you from his cruelty.

Zakhar's fingers twist and thrust inside you, causing you to cry out in pain. "You'll never be anything more than a whore," he says, his voice cold and emotionless. "But I'll enjoy breaking you." He withdraws his fingers from your abused flesh, leaving you shaking and sobbing on the cold stone floor. You watch in horror as he wipes his fingers on your face, smearing your tears and snot across your cheeks. "Clean up this mess," he commands, gesturing to the puddle of your tears. "Then, you'll learn to serve me properly."
 
Last edited:
I was almost finished writing a pretty long guide on how to get started when my browser froze so I had to restart it. :doh:



Hope this helps! :)

Wow, thank you! I'm a bit reluctant to rely on cloud services for what I have in mind, though. My machine used to be a $4k one, several years ago, now it isn't even a decent one for 3D rendering, and this is one of the reasons why I have practically stopped doing art for good.
As for AI imaging, I'm afraid that AI RP is out of reach for me, too.
 
Wow, thank you! I'm a bit reluctant to rely on cloud services for what I have in mind, though. My machine used to be a $4k one, several years ago, now it isn't even a decent one for 3D rendering, and this is one of the reasons why I have practically stopped doing art for good.
As for AI imaging, I'm afraid that AI RP is out of reach for me, too.
Ah, I see. The Kobold option is completely anonymous, even if you register, though. The RunPod option is also practically anonymous but theoretically it’s possible for the company to spy on your traffic (which is highly unlikely).

I’ve been using the less private option (#2) to do extreme racial slavery RP myself and I even used my actual email to register. Probably it wasn’t a good idea although I have a feeling that it won’t get me banned since the primary reason why people use OpenRouter or Mercer is that they don’t want censorship. Still, I understand why you feel hesitant to use a cloud for NSFW RP.
 
Honestly, it did a long time ago for me. AI makes images look more fake than 3d renders. Yeah i know that makes no sense, but eh' what do I know. lol
Some AI stuff looks really good but yeah a lot of it looks very fake with artificially smooth, almost plasticky skin textures and maybe it's just me but a hell of a lot of the AI stuff that gets posted here looks very generic, with all the characters looking more or less the same :(

AI is getting really good but it's still a long way away from understanding the nuances that make women desirable. Real bodies are imperfect, and that's what makes them unique. AI tends to distill this down to an idealised formula which banishes any hint of physical individuality and results in everything looking the same :(
 
I mean, I’ll be using AI more substantially soon. But this endless spam from unskilled and unartistic hacks will really dull the wonder for a lot of folks here.
I get it, you don't really care for AI, you seem to feel it devalues all your previous SUPERB efforts with drawings and 3D. But just as your efforts yielded top notch results, alot of others produced IMHO "amateurish spam" that was freely published on these forums.
I think in the same manner people who master AI will rise to the top, just as you did in your genre. I see AI as complementing your work , but DIFFERENT. The same way the "comic book art" of Steve Cordero still titillates the hell out of me! I think there's room for all types of expression, and that these efforts deserve PATIENCE as the producers develop their skills!
Cheers! :icon12:
 
Honestly, it did a long time ago for me. AI makes images look more fake than 3d renders. Yeah i know that makes no sense, but eh' what do I know. lol
Some AI stuff looks really good but yeah a lot of it looks very fake with artificially smooth, almost plasticky skin textures and maybe it's just me but a hell of a lot of the AI stuff that gets posted here looks very generic, with all the characters looking more or less the same :(

AI is getting really good but it's still a long way away from understanding the nuances that make women desirable. Real bodies are imperfect, and that's what makes them unique. AI tends to distill this down to an idealised formula which banishes any hint of physical individuality and results in everything looking the same :(

I feel most of it is because AI still involves a lot of unnecessary hassles to make a good image as exactly one intended, and that there are many more people using it as an image slot machine than a serious tool for creative expression. It can certainly lead to the feeling that they look the same due to many factors - for example, the faces generated in a single prompt will usually look very similar. Still, few care to generate each face to give it individuality. Also, AI models tend to be biased to generate idealistic females because their training sets usually contain more images of good-looking women than otherwise.

Combined with the fact that it tends to introduce anatomical deformities and other similar errors, especially if you don’t care to upscale fix them by inpainting, they give the impression of being uncanny or fake.

But I feel the other parts of what you said could be due to bias. Art is a subjective matter, of course, but if the subject is limited to (photo)realism, we can objectively compare different artistic expression methods. Contrary to what was mentioned, photorealism is one area in which AI is definitely superior to traditional methods like 3D art.

PBR, or Physically Based Rendering, is the current standard for a 3D renderer to depict materials. As the name implies, it tries to mimic how light particles interact with a surface according to its physical traits. When it became popular, it revolutionised the entire 3D field by allowing much more realistic renders previously impossible to achieve. By the way, that’s why Daz3D renders from its early days look much more cartoonish in comparison because it didn’t have a PBR workflow back then.

The problem is, however, that it’s impossible to perfectly simulate every little physical interaction between a surface like human skin and light particles that way. Even if you can render using a supercomputer, it won’t look perfectly realistic because the few - typically 3 - textures we use to depict different aspects of a physical surface are utterly insufficient to capture every little detail of it.

Of course, it is possible to achieve photorealism with 3D art, and that’s why we have all those films with perfect CGI effects. But to approach that level of realism with the PBR approach, you’ll need top-level talent in the industry as well as an enormous amount of time and effort to photo scan and/or sculpt required textures for each character you create, which is well beyond the reach of any Daz3D artist.

In comparison, AI doesn’t try to mimic physical traits. Instead, it simply predicts the most likely pixel values given a condition. So, if it was trained over photos, it would produce photographic results.

As such, it’s just a fact that AI is far better than 3D—not to mention Daz3D, which doesn't have a top-quality renderer—at producing photorealistic images. One can easily tell a Daz3D render from a photograph, but it's not always easy for an AI render.

Compare these two images: the left is a promotional image from Daz3D's website, which I'm sure was made by someone proficient at using the tool, and the right is an AI render I just made with a single prompt without any editing. Which do you think feels more "fake" with "plasticky" skin?

Screenshot_20240307_093502.jpeg ComfyUI_temp_lfgxo_00018_.png

I think you know the answer.

P.S.: Regarding photorealism in 3D art, it's not all about materials and lighting that count. Sometimes, small details like that slightly depressed skin under the underwear on her back or those tiny body hairs on her arm that you see from the right image can give you an impression that it's a real human being.

Details like those are extremely difficult to depict with the traditional 3D modelling method and nearly impossible if you just use ready-made assets and a simplified tool, as is the case with most Daz3D creators.
 
Last edited:
Back
Top Bottom