discussion Viability of AI assets and interest in a High Quality AVN using them

[VNThrowaway]

Hello everyone, this is my first thread and I'm hoping to get some opinions and feedback. It's going to be rather long, but I want to provide enough context for this to be useful.

For a little background: I've been playing AVN games off and on for a very long time (think games on DVD) and I've always had a strong interest in game development, but what has always stopped me is the barrier to entry in creating 3D art assets. I have a background in art, but mostly traditional pencil to paper type drawing, and I simply don't have the time or patience to learn an entirely new skill-set and sink thousands of hours into 3D modeling.

So with the rapid improvements of AI generation, and my professional interest in learning how to use AI tools I've been tinkering with both AI image and video generation, editing and optimizing workflows. I've been pretty blown away by the results as I've gotten familiar with the tools and learned how to prompt and use reference images to get consistent results, and it's gotten me thinking that with my day job as a software developer the technology may have reached a point where I can actually create something pretty incredible without needing a team of people helping me.

So my question is this: given the current kind of "AI slop" that seems to be the only stuff in the AVN community currently, would you (personally) be interested in a high quality AVN using AI generated assets and do you think the community would be receptive to it on a wider scale if the game didn't look like generic AI?

To add a bit of additional context: I'm talking about training my own LORAs for each character, for locations, etc. so that the consistency would be good enough that you'd be hard pressed to tell the images and videos weren't made by hand. I'm also thinking about a pretty ambitious scope, a deep story and world building, something with mini games, possibly a deck building system (think slay the spire), and even potentially multiplayer down the road.

This wouldn't be a quick process by any means, and I likely wouldn't have anything even remotely ready to be played in the next year, but before I sink a bunch of time, effort and potentially money investing in additional hardware I want to get a feel for what people think. Thoughts?

 

[clevernamexxx]

In my opinion, it would be refreshing to see a game made with AI that didn't immediately look like every other AI slop game being pumped out by every dude in his moms basement with a half decent GPU right now. I think that if you can manage what you envision here, it would likely be well received.

There will always be those who push back against AI, but that's unavoidable. At the end of the day AI is a tool that is here to stay whether we like it or not so we should make the best uses of it that we can.

 

[Redyscom]

It all depends on the quality and quantity of the AI you use. Too much of it—no matter how good it is—will turn people off, and too little of it—if it’s bad—will be even worse.
A moderate amount of high-quality AI is best.
Think of it like a spice: you can’t make a dish with just spices. If you use too much, it’ll be terrible; use just the right amount. It should be an addition, not the main ingredient.

 

[VNThrowaway]

In my opinion, it would be refreshing to see a game made with AI that didn't immediately look like every other AI slop game being pumped out by every dude in his moms basement with a half decent GPU right now. I think that if you can manage what you envision here, it would likely be well received.

There will always be those who push back against AI, but that's unavoidable. At the end of the day AI is a tool that is here to stay whether we like it or not so we should make the best uses of it that we can.

That is my thought too, on one hand the part of me that has spent thousands of hours drawing wants to hate AI, and I'm really not a fan of the half-assed slop I've seen thus far... but then there is also a part of me that has always wanted to craft an experience for players the way I used to when I was leading big interactive RP storylines on the MUD I played in the late 90s and early 2000s.

It is definitely a massive undertaking and I'm not taking the decision lightly, but I'm the kind of person who doesn't do things half-way and I'd rather not start than put something out that doesn't meet my standards.

It all depends on the quality and quantity of the AI you use. Too much of it—no matter how good it is—will turn people off, and too little of it—if it’s bad—will be even worse.
A moderate amount of high-quality AI is best.
Think of it like a spice: you can’t make a dish with just spices. If you use too much, it’ll be terrible; use just the right amount. It should be an addition, not the main ingredient.

That's the thing though, if it's done properly you won't be able to tell it's AI (other than the fact I would be completely transparent about it). At the end of the day AVNs are just static images and movies, and how many AVNs have you seen that use the same generic backgrounds? I've seen at least a dozen that use the same park, tunnel and sidewalk images from different developers.

There is also a big difference between a trained and optimized model and the stuff you're probably thinking of that was generated using one of the hundred crappy online tools. I'm running my models locally, with custom workflows and LORAs that have been trained for specific things. I can generate anything from anime to CG to photorealistic images that are virtually indistinguishable from the original reference but edited.

AI is just a tool, it can't replace the creative vision, the creator's eye for style/art, and the effort it takes to create something truly great. Maybe this will help express what I'm thinking a bit more clearly: imagine if I trained a model based on a certain visual aesthetic for the houses and environment in a fantasy village, then I sketch out scenes and outlines of the locations I want, I can then plug those into my workflow and generate them based on the trained model and iterate for the result I want. I could do the same thing for characters, sketch out poses, expressions, and scenes in rough detail and then turn those into polished results without spending an obscene amount of time drawing the final product myself.

 

[exlurker]

Just have fun, and post what you find out and what your results are. I am always looking for new games and images.
My recommendation: start small, do a very little demo, and cut what you want do do into very small pieces.
Also, share the source code of what you make, and let interested people continue your work if you get tired.

 

[Paradigminant]

The short answer is yes, I'd be interested. The long answer about AI as a tool, human artistic input, ethical thoughts, etc is something I've already discussed elsewhere and everyone seems to agree on. So yes. Use the tools available to make YOUR art. Love to see it.

 

[Lemon9X]

I have a game that I like to play. Now they use AI to help create cutscenes. And it's terrible. It made me realize that they don't care enough about their game. But I think if you're serious and attentive, AI will help you a lot.

 

[tias]

Your ideas sound much better than the generic AI games.
And it will probably be helpful to consistently make progress in the development, for some games you can tell that the developer cares a lot but the wait times between updates are killing me

 

[VNThrowaway]

This definitely isn't going to be generic, I've decided to do a "test run" by working on a re-imagining of a long abandoned game I enjoyed. I've already done a lot of tinkering with different models, workflows and have now got everything set up to train my own models and generate consistent faces. I've also found that local AI agents are actually really amazing at helping with refining prompts and planning. I've sunk a fair amount of time the past week on this and I'm honestly pretty excited.

I have only just started but the results are already good and will only improve over time. I think the biggest factor is just that I'm taking the time to build things up from the ground instead of just trying to throw something together or without understanding the process. Most of the time is going to be getting the foundation laid, and once I have everything set up I'll be spending most of my time writing and coding while my second computer goes through the painfully slow process of generating assets. If people are interested perhaps in a week or two I'll put together some photos showing comparisons and my concepts.

 

[tias]

This definitely isn't going to be generic, I've decided to do a "test run" by working on a re-imagining of a long abandoned game I enjoyed. I've already done a lot of tinkering with different models, workflows and have now got everything set up to train my own models and generate consistent faces. I've also found that local AI agents are actually really amazing at helping with refining prompts and planning. I've sunk a fair amount of time the past week on this and I'm honestly pretty excited.

I have only just started but the results are already good and will only improve over time. I think the biggest factor is just that I'm taking the time to build things up from the ground instead of just trying to throw something together or without understanding the process. Most of the time is going to be getting the foundation laid, and once I have everything set up I'll be spending most of my time writing and coding while my second computer goes through the painfully slow process of generating assets. If people are interested perhaps in a week or two I'll put together some photos showing comparisons and my concepts.

Just out of curiosity, which game are you referring to?

 

[VNThrowaway]

Just out of curiosity, which game are you referring to?

Astral Lust, it was by a guy called Victorious IIRC. It was a bit of an odd premise, felt a little bit like a silent hill vibe, but it had a slay the spire deck building aspect to it that was neat and there were a lot of random events where you got to choose how to act that influenced your character.

 

[bigusdickus233]

Lot of good ideas here, seems like you have a vision which is more than can be said about a lot of AI slop games in the past year. The one thing I'll push back on is saying people won't be able to tell it's AI art. I don't know about that man, people are pretty good at telling. Knowing and caring are 2 different things though, and if the rest of the game holds up people can either be blissfully ignorant (not in a derogatory way just the truth) or not care. I for one would at least be interested in a proof of concept

 

[VNThrowaway]

Lot of good ideas here, seems like you have a vision which is more than can be said about a lot of AI slop games in the past year. The one thing I'll push back on is saying people won't be able to tell it's AI art. I don't know about that man, people are pretty good at telling. Knowing and caring are 2 different things though, and if the rest of the game holds up people can either be blissfully ignorant (not in a derogatory way just the truth) or not care. I for one would at least be interested in a proof of concept

I get the skepticism, but frankly, most people haven't seen good AI, or rather they didn't realize they were looking at AI art. Also, keep in mind that I'm not talking about someone rigorously examining every asset, I'm talking about a casual look. Someone who sees some screenshots, takes an interest and plays a bit. Like I said above, most people who have used AI are using crappy online tools that have been thrown together by unscrupulous companies hoping to cash in on the AI hype and fleece unsuspecting consumers.

That isn't what I'm doing. To give you an idea of the difference:

Most AI sites are just loading a basic model, taking a prompt and generating an output. Maybe they have some fine-tuning LORA options, but they end up looking like obvious CG art or overly shiny. By the very nature of a consumer model and how complex these workflows can be, they can't offer the same power/control as someone working locally using their own tools, but the barrier to entry essentially requiring a 5090 or pro level hardware to generate high quality assets is also a big factor. Most people don't have $5-10k laying around to get started, and that's before the absurd hikes in RAM and Storage.

My process on the other hand looks like this: load a model and pass in a detailed prompt to get a starting point with general features for the face I want. Take that face into another workflow to generate a 512x512 SDXL image and refine the face until I get what I want. Use that face in another workflow with edit models to generate different expressions with the same face, then take those expressions into another workflow that has FaceID and ControlNet to control face and pose and generate a variety of poses/angles (this is the step I'm on currently). Once I have 50~ images for a character I'm then going to use that to train a LORA for that specific character, which I will then use to generate more images in a variety of outfits/environments/poses. Then I'll take the expanded image-set and re-train a new LORA for that character.

After all that is done I can generate higher quality images of that character and use a variety of options to fine-tune the results and apply filters. I have already been able to produce images that when edited are difficult to distinguish from real photos, and I'm just getting started, but I'm also going to need to weigh the art style I want to go for. I don't necessarily want the game to look like real life, but I haven't decided yet on a visual style, and because of how I am approaching this I don't have to. The method of training LORAs for everything allows me to push that decision to later, because I can just as easily generate a pixar-style image that looks like my character as a photo-realistic one.

I will also be doing something similar for environments, by the end of this I will have spent hundreds of hours doing prep-work just to be able to *start* generating actual assets. I'm also doing all of this locally, with multiple different models (flux2, zImage, SDXL, Qwen) and I have a local AI agent that I'm using to refine my prompts and make sure I understand the quirks of each model and how it wants information for the best results.

The TLDR is that I'm fairly confident in saying there isn't anyone in the AVN community doing what I'm doing currently, and the fact I've had to go into multiple tools and debug them and write code just to get things working only reinforces that. I've spent dozens of hours just setting up environments, fixing errors and figuring out compatibility issues. These tools aren't simple or straightforward once you go beyond the bare basics of image generation, and documentation is like the wild wild west since things are changing so rapidly no one can keep things straight. Honestly, it's been a massive pain and if I wasn't so excited about the results I've gotten I would have already given up and waited a couple years for more mature tools, but I've wanted to explore creating an experience for players for a very long time and the prospect of being able to actually realize that now is a big motivator for me.

 

[bigusdickus233]

I get the skepticism, but frankly, most people haven't seen good AI, or rather they didn't realize they were looking at AI art. Also, keep in mind that I'm not talking about someone rigorously examining every asset, I'm talking about a casual look. Someone who sees some screenshots, takes an interest and plays a bit. Like I said above, most people who have used AI are using crappy online tools that have been thrown together by unscrupulous companies hoping to cash in on the AI hype and fleece unsuspecting consumers.

That isn't what I'm doing. To give you an idea of the difference:

Most AI sites are just loading a basic model, taking a prompt and generating an output. Maybe they have some fine-tuning LORA options, but they end up looking like obvious CG art or overly shiny. By the very nature of a consumer model and how complex these workflows can be, they can't offer the same power/control as someone working locally using their own tools, but the barrier to entry essentially requiring a 5090 or pro level hardware to generate high quality assets is also a big factor. Most people don't have $5-10k laying around to get started, and that's before the absurd hikes in RAM and Storage.

My process on the other hand looks like this: load a model and pass in a detailed prompt to get a starting point with general features for the face I want. Take that face into another workflow to generate a 512x512 SDXL image and refine the face until I get what I want. Use that face in another workflow with edit models to generate different expressions with the same face, then take those expressions into another workflow that has FaceID and ControlNet to control face and pose and generate a variety of poses/angles (this is the step I'm on currently). Once I have 50~ images for a character I'm then going to use that to train a LORA for that specific character, which I will then use to generate more images in a variety of outfits/environments/poses. Then I'll take the expanded image-set and re-train a new LORA for that character. After all that is done I can generate higher quality images of that character.

I will also be doing something similar for environments, by the end of this I will have spent hundreds of hours doing prep-work just to be able to *start* generating actual assets. I'm also doing all of this locally, with multiple different models (flux2, zImage, SDXL, Qwen) and I have a local AI agent that I'm using to refine my prompts and make sure I understand the quirks of each model and how it wants information for the best results.

The TLDR is that I'm fairly confident in saying there isn't anyone in the AVN community doing what I'm doing currently, and the fact I've had to go into multiple tools and debug them and write code just to get things working only reinforces that. I've spent dozens of hours just setting up environments, fixing errors and figuring out compatibility issues. These tools aren't simple or straightforward once you go beyond the bare basics of image generation, and documentation is like the wild wild west since things are changing so rapidly no one can keep things straight. Honestly, it's been a massive pain and if I wasn't so excited about the results I've gotten I would have already given up and waited a couple years for more mature tools, but I've wanted to explore creating an experience for players for a very long time and the prospect of being able to actually realize that now is a big motivator for me.

I'm always gonna be at least a little skeptical till proven wrong, just the way of the world. Sounds like I gotta be keeping tabs on you though lmao, good luck my man

 

[VNThrowaway]

That's totally fair, given what I've seen in the past year I don't blame you one bit. Honestly, until I started digging into this stuff and actually running things locally myself I would have said exactly the same thing. We will see if I can maintain this enthusiasm though, that's where projects really hinge. Right now this is all fun and exciting, but it is a massive time sink and it also means that when I'm generating things I basically can't use my PC because it's so resource intensive. I think maybe next weekend I'll post some shots of what I started with and the process, it will be a good way to hold myself accountable and if people get excited I'm hoping that will be motivation enough to keep going.

 

[bigusdickus233]

That's totally fair, given what I've seen in the past year I don't blame you one bit. Honestly, until I started digging into this stuff and actually running things locally myself I would have said exactly the same thing. We will see if I can maintain this enthusiasm though, that's where projects really hinge. Right now this is all fun and exciting, but it is a massive time sink and it also means that when I'm generating things I basically can't use my PC because it's so resource intensive. I think maybe next weekend I'll post some shots of what I started with and the process, it will be a good way to hold myself accountable and if people get excited I'm hoping that will be motivation enough to keep going.

You mind sharing your rig specs? Curious what it really takes to get this process off the ground. Sounds graphic intensive but I'd think you'd need more than 1 GPU even with a higher end to make any of this really feasible. Sounds like you have your work cut out for you, I'll look forward to seeing your updates!

 

[VNThrowaway]

You mind sharing your rig specs? Curious what it really takes to get this process off the ground. Sounds graphic intensive but I'd think you'd need more than 1 GPU even with a higher end to make any of this really feasible. Sounds like you have your work cut out for you, I'll look forward to seeing your updates!

My main rig is a 9800x3d with a 5090 and 64GB of memory, all tuned and overclocked/undervolted. I also have a second computer with a 5900x and a 3090 or 5070ti I could throw in that to do less intensive stuff if I get to that point. I'm a software developer, a pretty hardcore gamer, and I've been building PCs for over 25 years at this point so I've always had pretty high-end hardware.

I grabbed a few screenshots of some of the initial generations I did for the main girls, they are pretty early and this is without the trained LORAs, but here's a sneak peek:

You must be registered to see attachments

You must be registered to see attachments

 

[bigusdickus233]

Huh, that's actually not as bad as I was expecting. Lowkey thought you'd have a GPU farm or something with the way you talked through your workflow lol. Still crazy expensive, you gotta be makin MONEY

 

[VNThrowaway]

Huh, that's actually not as bad as I was expecting. Lowkey thought you'd have a GPU farm or something with the way you talked through your workflow lol. Still crazy expensive, you gotta be makin MONEY

I do ok, but mostly I don't spend my money on much so when I do want something nice every few years I have the freedom to buy it. It definitely wasn't always that way, I worked my ass off for a decade to pay off my debts while pinching pennies and making sacrifices knowing that it would be worth it in the long run. A lot of people struggle with delayed gratification, but thankfully it's something I learned growing up dirt poor and spending most of my 20s flat broke.

 

[VNThrowaway]

I've learned some pretty amazing things about prompt generation for LORA training this week that are a game-changer for consistency/efficiency, and I've pivoted a bit to a more stylized look rather than going for characters that look more photo-realistic. I'll probably post an update later today or tomorrow if people are interested in seeing what I'm working on.