AI in the NSFW world ? \ Test : Using NovelAi to generate NSFW image

[cornxxx]

Do you think that AI has a future in the NSFW world?
Either in a hentai that would have been partially generated by an AI or in a game with an AI that would have generated
graphics part such a sprites/drawings ?

A few months ago, this seemed obviously impossible to me as I guess everyone else, it seems that today with the technological progress this thought could change.

Today I'd like to talk about this.
It's been several years now that I'm passionate about AI, especially in a technological way, but apart from a few projects that stood out like the creation of autonomous text with GPT-3, we didn't really think that AI could more or less revolutionize the world of NSFW (at least in this context of NSFW, in reality it has helped a lot in other areas such as SEO for example).

In short, as you probably noticed especially this year 2022 has been rich in the development of this niche, with the arrival of several impressive project this year, I speak obviously of the generation of image that can be realized with only one sentence.

We finally had the arrival of "DALL E2" which was quickly copied by several alternative such as the most known "Midjourney" or more economically "DALL mini".

But obviously always in the context of NSFW, even if Midjourney & Dall E2 proposed beautiful and impressive images, the images of anime and manga characters remained sometimes to be desired and obviously the NSFW tags were blacklisted not allowing to create all the things what we wanted.

But still thinking maybe this business will develop in the years to come, I discovered a new project, the NovelAI one (

You must be registered to see the links

)

It's been 1 day now that I know this project, so what I can say about it is that it's also an AI that generates images like Dall E2 / MidJourney but only this Ai is directly inspired by Manga and Anime characters, allowing more and more successful images in this context, and of course what I noticed is that NSFW terms are not blacklisted!

So I'd like to know your opinion on it, in this context I'll show you some pictures of my curent work in progress game that have been really drawn by me (so a human)
and what the AI of NovelAi has reformed these images


HERE IS THE ORIGINAL IMAGE MADE BY ME FOR MY GAME ( SO DRAWN BY A REAL HUMAN )

-

-

HERE ARE THE ALTERNATIVE PICTURE THE AI GENERATED ( SO DRAWN BY THE AI )
-




-

-




-
HERE IS THE ORIGINAL IMAGE MADE BY ME FOR MY GAME ( SO DRAWN BY A REAL HUMAN )
-

-
HERE ARE THE ALTERNATIVE PICTURE THE AI GENERATED ( SO DRAWN BY THE AI )
-





-

-





-
HERE IS THE ORIGINAL IMAGE MADE BY ME FOR MY GAME ( SO DRAWN BY A REAL HUMAN )

-

HERE IS THE ORIGINAL IMAGE MADE BY ME FOR MY GAME ( SO DRAWN BY A REAL HUMAN )
-



-

-




-
And of course this is just only the visible part of the iceberg, for this test, I just asked the AI to replicate my drawing, but it is also possible to ask to be inspired by the image and to add things to the sentence

So what do you think of it ? I think in my opinion this is a great technological evolution, which evokes both the immense evolutuon of AI, maybe make me feel a little bit of scariness, I feel like this can go wrong in many ways

 

[Sphere42]

Short answer: slaves are cheaper. Or starving artists in this case.

Neural networks require insane amounts of data and effort to get going so ultimately they are only "remix generators" in this context not something capable of "creativity" or engaging with abstract human understanding. This means that variety and capabilities of new AI tools will be rather limited while the sheer quantity of generated outputs will be massive.

In other words exactly what we have now with a few skilled human trendsetters and droves of rehashes following in their wake. Only those behind the cashgrab Paytreons will be wrangling keyword prompts and sifting through several random outputs instead of shifting models or sliders in DAZ.

The big question is whether we can make the output consistent enough to run the image synthesis live during gameplay. As your curated examples demonstrate that could begin a new era of customisability without ballooning the costs of each individual project. Just have the player go through a simple character creator and they can play as a black woman protagonist or an older dude or a green-skinned bimbo with elf ears and three breasts and two penises, at no additional work for sprites or renders or even text substitution if you can get a second AI for that. We can sort of do this already (Skyrim, Honey Select) but not quickly nor easily enough for the average consumer.

One way in which I can see near-future AI be incredibly powerful though is in private use for personal customisation. With how many tech and IT jobs there are nowadays I'd wager there are plenty of people who can grasp the basics of operating a neural net but can't draw or render to save their life. Copyright doesn't prohibit personal copies or modifications either so an "AI image filter" which can make hair blonde or boobs bigger might even be mainstream marketable for less pornographic uses while distributing modded artwork without permission from the original artists almost certainly isn't.

The scary part is that some asshole will inevitably try to use such AI to create images of real-life children and then all the pearl-clutching moralising idiots will unite against it and fight to have it destroyed entirely. While covering for their friends and priests molesting actual children of course.

 

[Catapo]

The scary part is that some asshole will inevitably try to use such AI to create images of real-life children and then all the pearl-clutching moralising idiots will unite against it and fight to have it destroyed entirely.

Maybe I am too cynical but ... 100 % this.

Add to that all the offensive content that this can be used for: racism, sexism, religion and political slander. Even in just meme forms.

Add all the awful ways that some old farts in management positions come up with to use it: like product advertisments, marketing bullshit or whatever. (Nothing will kill this tech faster for me)

Add all the scandals that would come with when some "artists" are exposed to have been paid for comissons, and turns out they were using something like this behind the scenes.

From a technical point of view, this is mindblowing even with all the obvious mistakes and limitations and I would love to play with it and learn more about how the neural net was trained for this stuff but I believe it is shitty human behaviour that will prevent this from evolving too far.

 

[Sphere42]

Add to that all the offensive content that this can be used for: racism, sexism, religion and political slander. Even in just meme forms.

Add all the awful ways that some old farts in management positions come up with to use it: like product advertisments, marketing bullshit or whatever. (Nothing will kill this tech faster for me)

See those aspects to me only make it more likely to succeed not less. A broad-purpose AI or a DIY toolkit which can be used for bigotry will be seen more favourably by the right-wing nutjobs so they might tolerate it also being able to create porn. Political (ab)use is a bit more nuanced because most unsavoury factions do want such a tool but they also want to deny their opponents access to it.

And needless to say no matter what our individual opinions are (to me personally it doesn't make a difference whether the ads are created by bots with silicon or flesh hardware, praise uBlock Origin!) if this kind of AI becomes economically viable for commercial purposes that will lead to a plethora of innovations for us to trickle down/pirate(/occasionally obtain legally?) so ultimately a net benefit so long as the market lead doesn't go to Disney, Nintendo or the like.

Add all the scandals that would come with when some "artists" are exposed to have been paid for comissons, and turns out they were using something like this behind the scenes.

If anything that would discredit the trade of human artists and bolster the popularity of AI even more. Obviously those guys would be scammers but that's humans being scum and we don't need them, because the AI was good enough to enable the scam in the first place!

With what's currently available I doubt Western society could even wipe it out entirely if it wanted to. Dark net gangs making deepfakes for blackmail or CP could just circulate backups endlessly or even train their own dedicated niche systems and that's with a hard coordinated crackdown. After all drugs and non-AI-generated CP still get traded and both require more telling infrastructure. Without a coordinated ban there will definitely be community efforts to create porn, might just take a bit longer if they can't use ongoing advances from big tech companies. Things will just suck more for the citizens of whichever countries decide blowing up the bathroom with people inside it is a good way of changing the bathwater.

 

[NRFB]

Hi, I can weigh in on this with personal experience. I'm not an artist at all, but I do have a decent amount of image manipulation experience with GIMP. The art in my game is not ENTIRELY AI generated, but I do make a lot of use of the tech.

The current workflow for the art in my game is like this:

1. Create a model mostly matching my vision of the character in a 3D program (I'm choosing not to saw which one, but it's not like there's that many of them)
2. Pose and screenshot the character as needed for the concept of whatever art I want
3. Do some basic manipulation of the screenshot (color palette stuff mostly. Technically at this point the art could already be considered acceptable without any AI work, but I don't like it as much in the 3D program style)
4. Use AI image to image to lightly change up the style. This could take quite a while depending on how cooperative the AI ends up being in terms of actually delivering what I want and also not changing the character so much that they don't look like themselves.
5. Take the AI generated image and do more manipulation (touch up AI weirdness, fix colors, match style with the rest of the game art, etc)

The thing is, this takes a decent chunk of time, and it would probably be much easier to just work with an artist. It would mean getting an end product that is ACTUALLY what I want rather than what I can put together, and it would save time that I could use for other things. I'm not sure if I consider the art in my game placeholder or not because I don't think it's terrible and I have some pride in what I've pulled together, but I know it could be better and I'm open to the idea of working with an artist, just not sure how to set that kind of thing up.

Can't deny that it sure is less expensive this way though.

 

[Jaike]

HERE ARE THE ALTERNATIVE PICTURE THE AI GENERATED ( SO DRAWN BY THE AI )

So what do you think of it ?

I'd say the vaginas separated from the vulvas, or the double anuses, sometimes hard to tell which is which, are a little concerning.

 

[Count Morado]

Today's AI is like Jeff Goldblum's teleporter from the movie "The Fly" ...

"The computer is giving us its interpretation... It's, uh translating it for us; it's rethinking it... and something is getting lost in the translation... The flesh. It should make the computer, uh crazy. Like those old ladies pinching babies. But it doesn't; not yet because I haven't taught the computer to be made crazy by the flesh. The poetry of it..."​

This is Stable Diffusion attempting to create an original image (no image inputted to guide) simply from a text prompt for a blowjob scene:

Spoiler: WARNING: Not for the faint of heart.

You don't have permission to view the spoiler content. Log in or register now.

And here is a sample when it attempts to create one in an anime style

Spoiler: Abstract Anime

You don't have permission to view the spoiler content. Log in or register now.

 

[mfcg1754]

I've had some free time the last few days, and have been messing around with NovelAI's image generation using a mixture of text prompts and tags. I've been pleasantly surprised with the results, especially going in with extremely low expectations. Here are some sample images I've saved to give you an idea of what I've managed.

Spoiler: 2B

You don't have permission to view the spoiler content. Log in or register now.

Spoiler: Tattooed Girl

You don't have permission to view the spoiler content. Log in or register now.

Spoiler: Lady with glasses

You don't have permission to view the spoiler content. Log in or register now.

Spoiler: tfw no buff big tiddy orc gf

You don't have permission to view the spoiler content. Log in or register now.

Spoiler: Ginger Tomboy

You don't have permission to view the spoiler content. Log in or register now.

Spoiler: Y'Shtola

You don't have permission to view the spoiler content. Log in or register now.

I would have to spend more time experimenting with tags, prompts, and styles in the future, but at the moment it seems to me that this is best used as a desktop wallpaper or waifu pinup/selfie generator. You'll also have to be ready for a bunch of weird shit like extra/missing limbs or whetever. That being said, if you know what you want beforehand, feed it decent info, and have a fair amount of patience (and some extra money) you can get some ok results.
I could see this being used to generate images for a game using portraits like Strive4Power/Strive4Conquest or maybe to round out girl packs for famous characters (Princess Peach, Tifa, Zelda etc.) in something like Brothel King. In my experience it takes way too much effort and hand-holding of the AI to use it for a quick fap or anything, but I still have some fun with it.

It's been a nightmare generating anything more than naked pictures, and even that takes some work. Here's the only stuff that's remotely fappable imo.

Spoiler: NSFW Stuff

You don't have permission to view the spoiler content. Log in or register now.

Getting these characters to do anything sexual has been a huge pain in the ass. I guess even AI art imitates life.

I've uploaded my entire collection of stuff I've considered worth keeping around. The filenames have some of the tags and a seed if you want to try to fuck around with them. Maybe you can get something hot out of it.

You must be registered to see the links

 

[CardinalRed]

Things will change for real when we have one professionally trained for NSFW

also

And here is a sample when it attempts to create one in an anime style

Spoiler: Abstract Anime

You don't have permission to view the spoiler content. Log in or register now.

feet fetish

 

[aura-dev]

Thanks for the nightmare fuel, just what I needed before going to bed.

 

[anne O'nymous]

So what do you think of it ?

I think that you are too enthusiastic to impartially judge the quality of your examples.

In the first image, the girl have a really small eye on a weird oblong face.
In the second image, the hairs are an integrated part of her face.
The third is not bad, but would be better with an actual nose.
The fourth have a double anus.
The fifth have something that shouldn't exist on top of her boobs, and what with the weird combination nose/lips ?
The sixth also have a double anus.
The seven is the only one that is right, except the strange deformation of her anus.

All the BJ images are pure nightmares. Well, the fourth one and its double dick can possibly be some man dreams.

The last batch of images is the only part effectively not so bad. The images still have problems, but they are more subtle, and therefore less visible, than for the previous ones.

I think in my opinion this is a great technological evolution, which evokes both the immense evolutuon of AI, maybe make me feel a little bit of scariness, I feel like this can go wrong in many ways

The concept of Machine learning appeared in 1959, when

You must be registered to see the links

applied it to his checkers software. It's the first time a software was learning by himself from his previous experiences, and the first time an AI scientist was probing this approach. In the same time, if we keep apart

You must be registered to see the links

and co, and

You must be registered to see the links

none programmable machines, the first machine that can be named "computer" was created by

You must be registered to see the links

in 1937.
In the end, AI needed 63 years, over the 85 years since the first computer was created, to reach the point where they can sometimes draw not too bad images. I wouldn't qualify this of "immense evolution". Especially since so far all this rely almost exclusively on

You must be registered to see the links

; therefore it works, but no one, not even those who made it, effectively know why.
And this is the main problem that AIs have. They need to be fixed, not only in order to give consistent results, but also in order to remove their weirdness ; this whatever the AI, it don't limits to drawing ones. But they can't be fixed. Since no one know how the AI achieve to reach this result, no one know what have to be modified. Actually, AI scientists working on machine learning are mostly working in a dark room, groping the walls in order to find a switch, that they'll turn on or off before searching for the next one, while hoping that one day they'll found the right combination.
Not that it isn't impressive by itself, but doing it that way, they can find the solution in hour, or in one century ; both being equally plausible. And nothing guaranty that they'll be able to replicate this solution.

This is Stable Diffusion attempting to create an original image (no image inputted to guide) simply from a text prompt for a blowjob scene:

Stable Diffusion can do better, at least for posing girls. There's some amazing examples that can be found here (don't remember in what thread). But your own examples are pointing the second problem, after the constancy during the whole game, that actually can't be solved ; the more complex is the scene to generate, the faster it will turn into a nightmare.

 

[sliveart777]

While I think the technology is somewhat impressive, I think we're still far from having AI replacing human artists.

The biggest issue with the technology and what at the end of the day makes it less impressive than it otherwise may seem is the black box problem as mentioned. No one knows how it works so we can't reliably tweak it. When a program is made by human programmers, knowledge of the code allows for tweaks to be made easily and iteration follows naturally.

Black box AI is a shot in the dark and when it doesn't work as it should we have no idea why. Even the iterative process is beyond our understanding.

What the AI can do reasonably well is individual characters in various poses. But any additional dimension of complexity and it completely falls apart. Adding additional subjects, 3D sculpting or animation and its worthless.

We probably will get there in the future, but we aren't anywhere close.

 

[lolart]

you can use it to cut corners, but there's a consistency issue with ai that only a human artist right now can solve. ai images looks good from a distance, but frequently have issues with anatomy, perspective, lighting etc. when you look up close. of course, you could always paint over the image in post, but that requires artistic skill.

they're good at creating one-off images, but creating a sequence of images to tell a story and keep it in the same style and proportions is still basically impossible at the moment, and nowhere near the quality of a good human artist.

 

[Deleted member 440241]

you can use it to cut corners, but there's a consistency issue with ai that only a human artist right now can solve. ai images looks good from a distance, but frequently have issues with anatomy, perspective, lighting etc. when you look up close. of course, you could always paint over the image in post, but that requires artistic skill.

they're good at creating one-off images, but creating a sequence of images to tell a story and keep it in the same style and proportions is still basically impossible at the moment, and nowhere near the quality of a good human artist.

On the plus side, someone could generate ai images for their game and hire an artist to fix the flaws. It'd be less work for the artist (in theory) and if they back out the dev still has all the art.

 

[rk-47]

Intriguing

 

[cornxxx]

I have once again on my side tried to experiment with this AI

And honestly, even if some defaults are still present, there is absolutely no way to distinguish sometimes if it's done by an AI or by a human, it's terrifying

Once the tagging is well mastered

It allows to generate really beautiful images even better than a human sometime ....

A human will always be better, but I think the help of an AI can really help artists, especially with poses or backgrounds that can save a lot of time

 

[Kallel69]

I think its another good tool to use in a digital artist's toolbelt.

 

[desmosome]

It's certainly looking like it could be a tool for some people to generate art for their games. I'm not sure if it's quite there yet, but it should get there in the near future.

The first step that needs to be solved is character and style consistency. The second step would be to deliver what the input is looking for with high consistency. Maybe when the AI is perfected, it would be possible to literally just write keyword prompts and get a scene you want easily, but that's looking too far ahead.

I think the most feasible application for this would be in the img to img generation. If you could just draw an outline or sketch and have the AI complete the work in the same pose, that would save a huge amount of time or enable non-artists to make a 2D game.

It would be best used by actual artists, of course. The workflow could be something like this. Make a sketch. Have the AI complete the pic. Go to Photoshop and iron out the mistakes or details.

 

[Zlaito]

I have once again on my side tried to experiment with this AI

And honestly, even if some defaults are still present, there is absolutely no way to distinguish sometimes if it's done by an AI or by a human, it's terrifying

Once the tagging is well mastered

It allows to generate really beautiful images even better than a human sometime ....

A human will always be better, but I think the help of an AI can really help artists, especially with poses or backgrounds that can save a lot of time

What tag did you use for these images?

 

[Deleted member 5189987]

I think that AI can be useful for artists if they want to have a quick reference for specific scenes. AI will not replace good artists.