Machine learning and AI generated image for adult games: a roadmap

[chuangshiren]

Let me apologize for my English skill in advance. Not my first language but should be understandable.

Recently, there are a few new projects using AI generated images as visual and the feedback is generally positive. This makes me wonder how far we are from applying the latest technologies for our vicious entertainment. Below are some brief ideas of what I consider a roadmap toward my dream game:

Level 1: VN using offline AI generated images

We are already there. Only the beginning of course. Not everyone is fond of these arts but they are quite good IMO.

Level 2: VN using real time AI generated images

Some HTML game and maybe others let the players choose their preference (hair color, body type, skin tone, etc.) before the game starts and use models/images accordingly. These attempts are well received by the player base. But the choices are still very limited since they are using premade models/images. With current AI Image Generators, it is very possible to generate AI characters with the players' own preference. So the players will see different images even if they are playing the same VN. For characters, people can share their creations (same way like sharing HS or KK characters) so others could use in their playthrough (well, some may consider this borderline NTR ).

Level 3: Highly customizable VN using real time AI generated images and progressively generated characters/scenes

Why stop there? In addition to images, AI can generate lewd stories and scenes. The developer can set up a story outline and major branches. Then the AI and player will fill the flesh of the story together. Infinite replay value, customized experience. The AI could even learn from your past choices to create characters that suit your personal flavor and challenge your underdeveloped fetishes. The VN will be difficult to keep convergence though. So the developer needs to be a good storyteller.

Level 4: Personalized sandbox with real time AI generated characters, randomly generated events, AI written scenarios, agent-based simulated NPCs. etc.

This. The ultimate form of adult gaming. The developer provides a framework, a world with basic lore and rules, some background, and leaves everything else to AI and the player. Seems impossible at first but I'd say it's totally doable with today's technology. It's just not mature in all aspects. A few games on this site, namingly Loser, Lab Rats 2, and maybe Superpowered, are pioneers for some aspects of this personalized sandbox. Sorry if I missed any ongoing projects. If this fully personalized game is made someday in the future, it will be my last adult game for life (until the next game with better visual engine of course LOL).

Now, I'm curious of your opinions on the use of AI/ML for adult games. How far are we from Level 2-4?

 

[Müptezel Emo]

Alternate Title: How to create Nightmare fuel for poor bastards?

Just stop this AI topics, really.
This AI topics just like some talentless perverts are hope AI will eliminate all game devs.

AI is just another tool for artists, so you don't need to worry about it.

 

[chuangshiren]

Alternate Title: How to create Nightmare fuel for poor bastards?

Just stop this AI topics, really.
This AI topics just like some talentless perverts are hope AI will eliminate all game devs.

AI is just another tool for artists, so you don't need to worry about it.

"This AI topics just like some talentless perverts are hope AI will eliminate all game devs." I wonder based on which part of the OP you jump to this conclusion. Are you one of these people who'd assault strangers on the street just because you don't like what they wear?

I don't think you read the OP. I never thought AI could eliminate developers. On the contrary, I pointed out good developers with strong storytelling skill are needed. Also, developers are still needed for framework design, background setting, world lore creation... If you don't read, then your opinion is just prejudice.

Also I don't follow your logic of stopping AI discussion all together. AI is something happening right now even if you don't like it. Yes they are tools, but discussions can help people to use the tools better. "Stop all AI topics" sounds like burning down 5G towers because "they spread virus".

 

[Müptezel Emo]

"This AI topics just like some talentless perverts are hope AI will eliminate all game devs." I wonder based on which part of the OP you jump to this conclusion. Are you one of these people who'd assault strangers on the street just because you don't like what they wear?

Not for what they wear, but for what their dumbness levels are.

Also I don't follow your logic of stopping AI discussion all together. AI is something happening right now even if you don't like it. Yes they are tools, but discussions can help people to use the tools better. "Stop all AI topics" sounds like burning down 5G towers because "they spread virus".

Like a said, AI just another tool for artists, not for you.
So, if you are not artist, just go away and don't steal their work.

And "AI is something happening right now" sentence shows your lack of knowledge about AI.

 

[chuangshiren]

Not for what they wear, but for what their dumbness levels are.

So you just hit yourself when you walk down the street?

Like a said, AI just another tool for artists, not for you.
So, if you are not artist, just go away and don't steal their work.

Probably the dumbest thing I ever heard from this site. Let's say I'm not an artist, then by your "logic" I cannot talk about AI because apparently talking about AI equals stealing artists' work. How? Oh and AI can be tools for much more people than just artist, in case you don't know.

Well, guess your dumbness level prevented your from understand my sentence was just a lazy expression of "using AI generated art in games is something happening now." But your comeback really shows your knowledge about AI.

People will still talk about AI even if they are not artist, or not using AI as a tool. What can you do, huh?

 

[CardinalRed]

Until people start training their own models to have consistent characters in the scenes (which is not that hard), I don't think we'll see adult games being made with it.

It's just a matter of time.

 

[Deleted member 440241]

Am I stupid for expecting this to be a guide for which AIs produce consistent images (and how to make them do it) or can create adult art, like a road map to help devs navigate the sudden wealth of options? I think everybody knows it'd be awesome to have a roguelike engine generate a collection of AI art characters with a character.AI chatbot simulating distinct personalities for each. The potential of these tools isn't really in question. It's the implementation that's difficult.

 

[Müptezel Emo]

AI equals stealing artists' work. How?

You must be registered to see the links

 

[Müptezel Emo]

So you just hit yourself when you walk down the street?

No, for your sake, just stay away from streets for a long time

 

[chuangshiren]

No, for your sake, just stay away from streets for a long time

LOL sounds like a middle school bully would say. That really makes you look smart . Just because you'd randomly run into garbage bins when you walk down the street doesn't mean others should stay away from streets.

 

[chuangshiren]

You must be registered to see the links

Watched the video out of curiosity. So the take-away you got from the video is "don't even talk about AI unless you are an artist"? Seriously... Dude, you should sue your elementary teacher who taught you reading and comprehensive

 

[Müptezel Emo]

LOL sounds like a middle school bully would say. That really makes you look smart . Just because you'd randomly run into garbage bins when you walk down the street doesn't mean others should stay away from streets.

Yeah, whatever. You are just empty trash.
Just ignoring you.

 

[TechnoParagon]

In case you are aware, DLsite has a game listed made entirely with AI art.

You must be registered to see the links

The thing I notice about Ai art is the almost always do the hands like crap, so unless you want to redraw the hands there really isn't a way around it.

 

[unocrus]

Right now AI art is a copyright nightmare and I don't see it reaching broad commercial adoption until two things are met:

• People generating AI art need a way to certify their work was produced using original material. This would be akin to an artist training an AI with their own drawings.
• Artists need a way to lisence their work for AI training. Current rights to commercial use are too vague when it comes to this sort of derivative work.

With the first point met, I can see stock image companies moving to adopt the second point on their own as a way to flip their biggest threat into a new revenue stream. With both points met, smaller artists needn't worry about their work being ripped off any more than they did back when it was actual people tracing their art.

 

[chuangshiren]

Right now AI art is a copyright nightmare and I don't see it reaching broad commercial adoption until two things are met:

• People generating AI art need a way to certify their work was produced using original material. This would be akin to an artist training an AI with their own drawings.
• Artists need a way to lisence their work for AI training. Current rights to commercial use are too vague when it comes to this sort of derivative work.

With the first point met, I can see stock image companies moving to adopt the second point on their own as a way to flip their biggest threat into a new revenue stream. With both points met, smaller artists needn't worry about their work being ripped off any more than they did back when it was actual people tracing their art.

Good points. For the first point, I was in MICCAI a couple of years back and some algorithm developers talked about embedding digital signatures/fingerprints in CGIs and commercialized algorithm should be enforced to check these identifications in the training data. It was in one of those regulation topic panels. Back then it was mostly within academia. Dunno if any discussion has been moved to regulatory in recent years. Considering the cross-border nature of these stuffs, I'd say still long way to go.

Decentralized solutions like block chain might be easier to implement. But the problem with decentralization is the penalty to violators is hard to be prohibitive.

 

[water00]

Until people start training their own models to have consistent characters in the scenes (which is not that hard), I don't think we'll see adult games being made with it.

It's just a matter of time.

GPT3 cost over $100 million to train. Not going to happen. It's also wasteful to train a new model, and on what? Just take your favorite model, whether it's a huge CNN or Transformer or whatever, and do some transfer learning at the end.

Right now AI art is a copyright nightmare and I don't see it reaching broad commercial adoption until two things are met:

• People generating AI art need a way to certify their work was produced using original material. This would be akin to an artist training an AI with their own drawings.

Impossible. Not enough data. It's easier to change copyright law around AI, to fight a legal battle to say that this is fair use.

 

[water00]

Regarding OP. You really haven't written anything rigorous. There's no lewd lit review, not even a taxonomy of ideas. All you have is a rough list from Level 1 (a trivial application of readily available models) to Level 4 (fantasy artificial general intelligence, AGI).

If we could do Level 4, we'd have solved every problem currently known to humanity. It would be a watershed moment in human evolution.

There's really nothing in your OP that offers a springboard for discussion. I'll give you one. 99% of the time AI will be used to shorten dev cycles and make it cheaper, it'll simply be a bigger hammer, no more different than Photoshop or Blender or a faster CPU. Most people who buy into the AI buzzwords are people like you, who just believe the hype without thinking actually where in the pipeline it can be applied.

1% of the time, AI will be used in a truly novel way. Maybe the user can tell their waifu that their childhood fantasies, like wetting the bed? My point is, 99% of the time AI will have 0 direct impact on the user, because it was simply a bigger hammer.

 

[CardinalRed]

GPT3 cost over $100 million to train. Not going to happen. It's also wasteful to train a new model, and on what? Just take your favorite model, whether it's a huge CNN or Transformer or whatever, and do some transfer learning at the end.

Not sure if it is what you meant, but you can train a model with dreambooth, it's not very hard

 

[chuangshiren]

Regarding OP. You really haven't written anything rigorous. There's no lewd lit review, not even a taxonomy of ideas. All you have is a rough list from Level 1 (a trivial application of readily available models) to Level 4 (fantasy artificial general intelligence, AGI).

If we could do Level 4, we'd have solved every problem currently known to humanity. It would be a watershed moment in human evolution.

There's really nothing in your OP that offers a springboard for discussion. I'll give you one. 99% of the time AI will be used to shorten dev cycles and make it cheaper, it'll simply be a bigger hammer, no more different than Photoshop or Blender or a faster CPU. Most people who buy into the AI buzzwords are people like you, who just believe the hype without thinking actually where in the pipeline it can be applied.

1% of the time, AI will be used in a truly novel way. Maybe the user can tell their waifu that their childhood fantasies, like wetting the bed? My point is, 99% of the time AI will have 0 direct impact on the user, because it was simply a bigger hammer.

I won't deny that I didn't provide enough fuel for in-depth discussion. But my intent was simply sharing a perspective and asking for timeline estimate. You comment provided a timeline you estimated, which is "If we could do Level 4, we'd have solved every problem currently known to humanity". About that, I respectively disagree with you.

Technically speaking, nothing is unachievable for Level 4. If I want build a Level 4 game for just 1 people, and if I don't care about the quality of the game at all, and if the gameplay lasts 5 minutes, I can do that with existing tools. But those are three important "IF"s. They represent the challenges in computational cost, data availability, and quality consistency. Those gaps, adding to all the regulation issues, are the main reasons Level 4 is a long term goal.

Yes, AI is a bigger hammer. But that doesn't mean people cannot use it to accomplish great things. Nuclear power generation is essentially just boiling water more efficiently. You cannot say nuclear powered and steam engine powered are the same thing. I well understand what AI can do and what AI cannot do. If you accuse me for "buy into the AI buzzwords" and "just believe the hype", please point out which sentence in OP overstated AI's capability. I don't think AI is the answer to everything. I just did a deduction based on AI's known potentials.

For what I'm doing in real life, AI (along with ML, ANN/SNN, NLP, etc.) were "buzzwords" like ten years ago. And actually they have went through ups and downs several times now if you look at a longer time scale. So I won't overestimate what AI technologies can do. However, I also won't avoid talking about AI (or consider others think positively about AI "gullibly fool who knows nothing about the real thing") just to prove I don't "buy into buzzwords".

Underestimate is just as stupid as overestimate.