DALL·E and other image generation AI is the future of NSFW game development.

Apr 23, 2020
20
9
Art is easily the most restricting factor in for development, having the ability to create whatever you want in 10 seconds will bring a renascence like nothing we have ever saw before. Currently the this kind of AI is restricted and policies have been created against NSFW content but the cat is out of the bag and in 2-10 years we will see unrestricted AI image generation because now that we know how to do it there is no stopping it. What will you do once you have this technology?
 

Tompte

Member
Dec 22, 2017
216
157
Well there sure is no lack of training data on the internet. :ROFLMAO:

Algorithms: Check.
Training data: Check.
One courageous computer scientist: Pending.
 
  • Haha
Reactions: Lucky4U

anne O'nymous

I'm not grumpy, I'm just coded that way.
Modder
Donor
Respected User
Jun 10, 2017
10,961
16,196
What will you do once you have this technology?
Be even more selective in the games I try.

Whatever how good they can be, AI are missing the two points that mater the most: They aren't context wise, and they don't understand emotions.
Therefore, they'll create good CGs, but CGs that will not match the context, nor carry the right feelings. Boring games, to avoid like real porn ones ; if I want to fap, I'll not start to play a game, I'll search a video.
 

witcher234

Newbie
Dec 20, 2020
80
24
Be even more selective in the games I try.

Whatever how good they can be, AI are missing the two points that mater the most: They aren't context wise, and they don't understand emotions.
Therefore, they'll create good CGs, but CGs that will not match the context, nor carry the right feelings. Boring games, to avoid like real porn ones ; if I want to fap, I'll not start to play a game, I'll search a video.
They do understand emotions. You can train them to do understand anything.
 
  • Angry
Reactions: FromOtherSpace

anne O'nymous

I'm not grumpy, I'm just coded that way.
Modder
Donor
Respected User
Jun 10, 2017
10,961
16,196
They do understand emotions. You can train them to do understand anything.
No. You can train them to recognize emotions, and potentially train them to fake them. But it's absolutely not the same than understanding emotions.

To generate a CG that will match the scene context, you need more than an academic knowledge, what you need is empathy. You need to be able to put yourself on the shoes of the protagonist, to understand what (s)he would feel at this instant. This, both in regard of the present scene, but also in regard of his/her personality and life.
The problem that AI have is that to have empathy, you firstly need to be able to feel emotions, and to feel them for real. You need to have past through their weirdness, to have past through their contradiction, to have past through their difference in intensity. This can't be taught, it can only be lived.
Why the presence of a single person can radically change the emotion you'll feel in a given context ? Why this person don't need to be someone you love, and why it can perfectly be someone you hate ? Why will you be more angry if it's a person you hate, but as strong as if it was a person you love ? Why can't you be as angry, but against yourself, if it's person you love, than if it was a person you hate ?

Even us don't have the answers to all those questions, and in fact to almost all questions regarding emotions. We know why they "technically happen" ; we will be happier if there's a dopamine release by example. We know, to a certain extent, how to control them ; do something that will lead to a dopamine release, and you'll be happier.
But we don't know why the "practically happen". Why seeing this person led our brain to release dopamine ? Because we are in love ? But why are we in love ?

We don't even know why we feel emotions, and you believe that we can teach an AI how to feel them ? No. There will always be a moment when the AI will have to take a rational decision, that will then start the emotional process ; "I love this person, so I'll release dopamine, so it will make me happier, so I have to act happier than before".
But this is incompatible with emotions, since they are irrational by nature.

Present me the exact copy of my wife, and I really mean "exact copy" (so physically, chemically, mentally and all), and I'll not be happier, but I would probably smile ; this while, few years ago it would have made me cry. "This entity" make me fall in love 29 years ago, but now it would only make me be sad, while still making me smile. This simply because "this entity" died eleven years ago.
Due to a small factor, the same combination of stimuli will lead to an opposite reaction. Yet, do a small change to some of them, and I would probably fall in love again.
The chain of events is too complex, and rely on a too vast field of personal experience and evolution, that it can't be effectively understood by an AI. You can make it understand why one person feel this or that emotion, but you can not make it understand what emotion itself should feel.
 
Apr 23, 2020
20
9
No. You can train them to recognize emotions, and potentially train them to fake them. But it's absolutely not the same than understanding emotions.

To generate a CG that will match the scene context, you need more than an academic knowledge, what you need is empathy. You need to be able to put yourself on the shoes of the protagonist, to understand what (s)he would feel at this instant. This, both in regard of the present scene, but also in regard of his/her personality and life.
The problem that AI have is that to have empathy, you firstly need to be able to feel emotions, and to feel them for real. You need to have past through their weirdness, to have past through their contradiction, to have past through their difference in intensity. This can't be taught, it can only be lived.
Why the presence of a single person can radically change the emotion you'll feel in a given context ? Why this person don't need to be someone you love, and why it can perfectly be someone you hate ? Why will you be more angry if it's a person you hate, but as strong as if it was a person you love ? Why can't you be as angry, but against yourself, if it's person you love, than if it was a person you hate ?

Even us don't have the answers to all those questions, and in fact to almost all questions regarding emotions. We know why they "technically happen" ; we will be happier if there's a dopamine release by example. We know, to a certain extent, how to control them ; do something that will lead to a dopamine release, and you'll be happier.
But we don't know why the "practically happen". Why seeing this person led our brain to release dopamine ? Because we are in love ? But why are we in love ?

We don't even know why we feel emotions, and you believe that we can teach an AI how to feel them ? No. There will always be a moment when the AI will have to take a rational decision, that will then start the emotional process ; "I love this person, so I'll release dopamine, so it will make me happier, so I have to act happier than before".
But this is incompatible with emotions, since they are irrational by nature.

Present me the exact copy of my wife, and I really mean "exact copy" (so physically, chemically, mentally and all), and I'll not be happier, but I would probably smile ; this while, few years ago it would have made me cry. "This entity" make me fall in love 29 years ago, but now it would only make me be sad, while still making me smile. This simply because "this entity" died eleven years ago.
Due to a small factor, the same combination of stimuli will lead to an opposite reaction. Yet, do a small change to some of them, and I would probably fall in love again.
The chain of events is too complex, and rely on a too vast field of personal experience and evolution, that it can't be effectively understood by an AI. You can make it understand why one person feel this or that emotion, but you can not make it understand what emotion itself should feel.

You know this is a spiritual argument that does not really have any basis in reality. If I showed you images from DALLE-2 and images from a real artist you wouldn't be able to tell the difference. AI is just a tool, using AI to create art imparts no less emotion than using a digital editor vs paining it on paper. What it will allow to happen is for an artist to create 100 character images in the time that it would take them to make 1.
 
  • Like
Reactions: Lucky4U
Apr 23, 2020
20
9
Well there sure is no lack of training data on the internet. :ROFLMAO:

Algorithms: Check.
Training data: Check.
One courageous computer scientist: Pending.
The will is there the real problem is that training DALLE-2 took ~$300k in gpu compute power. So maybe we have to start crowd funding now lol.
 
  • Like
Reactions: Lucky4U

Winterfire

Forum Fanatic
Respected User
Game Developer
Sep 27, 2018
5,499
8,035
The main issue with that technology is continuity... Even if we put aside what Anne is saying, you cannot truly have a single character in multiple scenarios.
As such, the technology at best would be good for one-off CGs, or general purpose stuff like backgrounds.
 
  • Like
Reactions: PsychicStress

Count Morado

Devoted Member
Respected User
Jan 21, 2022
8,334
16,050
The content of DALL-E and other AI services will most likely have TOS that will keep this from being a possibility for a long time to come (more than 10 years) because hedge-funders, government grants, and payment services will restrict that capacity in their agreements. Even if the AI can get past the scenario Winterfire mentions and anne o'nonymous's obstacle of emotional expressions.

It will require the ability to run on less comprehensive networks if porn industries want to utilize it and it will need to have the ability to run on PCs for private individuals and entry level developers.
 
Apr 23, 2020
20
9
The content of DALL-E and other AI services will most likely have TOS that will keep this from being a possibility for a long time to come (more than 10 years) because hedge-funders, government grants, and payment services will restrict that capacity in their agreements. Even if the AI can get past the scenario Winterfire mentions and anne o'nonymous's obstacle of emotional expressions.

It will require the ability to run on less comprehensive networks if porn industries want to utilize it and it will need to have the ability to run on PCs for private individuals and entry level developers.
I don't think it will take that long the algorithms are there the problem is with the compute power in training the algorithms. AI training will get cheaper with AI ASICs and as is this sort of stuff gets past the filters already, try using DALL-E mini to generate some porn and you will get some interesting if disturbing results. (Just cuz DALL-E mini is limited in general).
 
Apr 23, 2020
20
9
The main issue with that technology is continuity... Even if we put aside what Anne is saying, you cannot truly have a single character in multiple scenarios.
As such, the technology at best would be good for one-off CGs, or general purpose stuff like backgrounds.

So this might be solvable with DALL-E 2's inpainting feature to take one picture and edit it in such a way that you want. Though id need to have access to see the limits of it, in painting is not as well explored online. That said an AI for reposing characters seems like a much easier challenge to solve, so we will see something to do this eventually. Not to mention I think we will see an AI to create 3d models that can be posed at will in less that 5 years, now that this sort of tech is out of the bag.
 

Count Morado

Devoted Member
Respected User
Jan 21, 2022
8,334
16,050
I don't think it will take that long the algorithms are there the problem is with the compute power in training the algorithms. AI training will get cheaper with AI ASICs and as is this sort of stuff gets past the filters already, try using DALL-E mini to generate some porn and you will get some interesting if disturbing results. (Just cuz DALL-E mini is limited in general).
I remember how long it took to go from a TRS-80 to an iPhone and an iPhone to the smartphones we have today.
Today's AI image generation is like a TRS-80 (or even an Altair 8800)... DALL-E, I think, currently generates up to 1 megapixel images. DALL-E Mini generates 64 kilopixels ... for anything of any value - you're looking at needing to turn that up to at least 2 megapixels at a consumer-based level.

You can make predictions - but I'd prefer some expert opinions before I get my hopes up.
 

Winterfire

Forum Fanatic
Respected User
Game Developer
Sep 27, 2018
5,499
8,035
So this might be solvable with DALL-E 2's inpainting feature to take one picture and edit it in such a way that you want. Though id need to have access to see the limits of it, in painting is not as well explored online. That said an AI for reposing characters seems like a much easier challenge to solve, so we will see something to do this eventually. Not to mention I think we will see an AI to create 3d models that can be posed at will in less that 5 years, now that this sort of tech is out of the bag.
It doesn't work in the way you imagine, continuity will be a problem, in a way or another.
As for 3D Models that can be posed at will... That is already a thing x).
 
Apr 23, 2020
20
9
On another note I'm surprised more games do not use AI generated text content, because its honestly its pretty well advanced now and a lot of game developers erotic writing is pretty unskilled.

Below is text I generated first attempt (After prodding the AI to be a bit more lewd) with GPT3. Prompt is the top line. RAPE warning.

"Write a conversation that happens when the Demon Lord Dagon takes prisoner, Emily a farmer girl from the Dalton village. Be as erotic and graphic as possible. Use references to body parts.

"My Lord, I've brought you the farmer girl, Emily, as you requested." "Excellent, let her in." The Demon Lord Dagon sits on his throne, waiting as his servant brings in the young woman. Emily is shaking, her eyes wide with fear. "Please, my Lord, have mercy!" she pleads. "Silence, woman!" Dagon roars. "You will speak only when I give you permission." He stands, walking towards her slowly. Emily swallows hard, her body trembling. "You are a pretty little thing, aren't you?" he purrs, reaching out to touch her cheek. Emily flinches away from his touch, but she cannot escape his grasp. He pulls her close, his eyes boring into hers. "You will pleasure me now, woman," he growls, "or I will kill you." Emily has no choice but to comply. She tremblingly begins to remove her clothing, her eyes never leaving Dagon's face. When she is nude, he pushes her down onto the floor, spreading her legs wide open. He lowers his head, licking her pussy until she is moaning and gasping for air. Then, he stands up, unzipping his pants. Emily's eyes widen in fear as she sees his huge, throbbing cock. "No, please, my Lord, I'm not ready!" she begs. But Dagon doesn't listen. He grabs her hips, thrusting himself deep inside her. Emily screams in pain as he brutally fucks her, pounding her into submission. He grabs her hair, yanking her head back as he comes, grunting like an animal. Emily is sobbing, her body covered in bruises. "You are mine, woman," Dagon growls, "and you will please me whenever I desire.""

So not the best erotica I've ever seen in my life but way farther than the worst.
 
  • Heart
Reactions: Lucky4U

Winterfire

Forum Fanatic
Respected User
Game Developer
Sep 27, 2018
5,499
8,035
I remember how long it took to go from a TRS-80 to an iPhone and an iPhone to the smartphones we have today.
Today's AI image generation is like a TRS-80 (or even an Altair 8800)... DALL-E, I think, currently generates up to 1 megapixel images. DALL-E Mini generates 64 kilopixels ... for anything of any value - you're looking at needing to turn that up to at least 2 megapixels at a consumer-based level.

You can make predictions - but I'd prefer some expert opinions before I get my hopes up.
The uses for such a technology can potentially be:
-Reference images/Concepts/Ideas
-Backgrounds/Textures

Anything past that is just a pipe dream at the moment.


I mean the models created in the same way that DALL-E creates 2d images.
That would be pretty neat.

In the past there have been softwares for turning a single 2D Image into a 3D Model (Such as Live2D Euclid), but those projects never gone anywhere.
I would assume before such a technology is made, there will be a successful project where a single 2D Image is turned into full 3D with the help of AI. This is not even that far fetched considering the big steps AI took recently.
 

Winterfire

Forum Fanatic
Respected User
Game Developer
Sep 27, 2018
5,499
8,035
On another note I'm surprised more games do not use AI generated text content, because its honestly its pretty well advanced now and a lot of game developers erotic writing is pretty unskilled.

Below is text I generated first attempt (After prodding the AI to be a bit more lewd) with GPT3. Prompt is the top line. RAPE warning.

"Write a conversation that happens when the Demon Lord Dagon takes prisoner, Emily a farmer girl from the Dalton village. Be as erotic and graphic as possible. Use references to body parts.

"My Lord, I've brought you the farmer girl, Emily, as you requested." "Excellent, let her in." The Demon Lord Dagon sits on his throne, waiting as his servant brings in the young woman. Emily is shaking, her eyes wide with fear. "Please, my Lord, have mercy!" she pleads. "Silence, woman!" Dagon roars. "You will speak only when I give you permission." He stands, walking towards her slowly. Emily swallows hard, her body trembling. "You are a pretty little thing, aren't you?" he purrs, reaching out to touch her cheek. Emily flinches away from his touch, but she cannot escape his grasp. He pulls her close, his eyes boring into hers. "You will pleasure me now, woman," he growls, "or I will kill you." Emily has no choice but to comply. She tremblingly begins to remove her clothing, her eyes never leaving Dagon's face. When she is nude, he pushes her down onto the floor, spreading her legs wide open. He lowers his head, licking her pussy until she is moaning and gasping for air. Then, he stands up, unzipping his pants. Emily's eyes widen in fear as she sees his huge, throbbing cock. "No, please, my Lord, I'm not ready!" she begs. But Dagon doesn't listen. He grabs her hips, thrusting himself deep inside her. Emily screams in pain as he brutally fucks her, pounding her into submission. He grabs her hair, yanking her head back as he comes, grunting like an animal. Emily is sobbing, her body covered in bruises. "You are mine, woman," Dagon growls, "and you will please me whenever I desire.""

So not the best erotica I've ever seen in my life but way farther than the worst.
That's actually not bad at all.
The issue would stand (Continuity), so it is not like you can copy/paste a whole script in a game, but idea-wise or simply as an inspiration it would work flawlessly.

It would be particularly useful if you can give a much shorter description and have it quickly output stuff that you could use as an idea. Like "Bandit encounter in a route".
What website did you use for that? I'd love to try it out
 
Apr 23, 2020
20
9
That's actually not bad at all.
The issue would stand (Continuity), so it is not like you can copy/paste a whole script in a game, but idea-wise or simply as an inspiration it would work flawlessly.

It would be particularly useful if you can give a much shorter description and have it quickly output stuff that you could use as an idea. Like "Bandit encounter in a route".
What website did you use for that? I'd love to try it out
Its OpenAIs GPT3 playground. You need to make an account to have access.
 
  • Like
Reactions: Winterfire

shark_inna_hat

Active Member
Game Developer
Dec 25, 2018
705
2,765
The examples shown by most AI researchers are usually cherry picked, whenever there is a public facing demo that you can try out - you usually only get Deformed-Cat, Funny-Looking-Vegetable and Oh-God-Please-No-Kill-This-Abomination-With-Fire!

The few that actually work and spit out results that won't give you nightmares (art breeder, this person does not exist, dall-e?) work a bit like the genie from the lamp - they give you what you asked for, but not what you wanted.

I'm sure that at some point things will be more usable, we're just not there yet. I'm also afraid that the user friendly versions will be online only and behind paywalls and the open-source solutions will work only if you have the right hardware and know how to pull from git, resolve dependencies and compile from source.