Why so few games with AI art?

[Bretg68]

Stable diffusion produces awesome imagmes that would great for these type of games. Why are there no major games with AI art?

 

[Youpm]

Probably are worried about IP/copyright issues which are yet to be resolved.

 

[fyl3toys]

Is it possible to create a character with stable diffusion and then use that same character through many different iterations on that platform to create images with continuity?

 

[anne O'nymous]

Is it possible to create a character with stable diffusion and then use that same character through many different iterations on that platform to create images with continuity?

Accordingly to the promotion screenshots from some AI game threads I end on by error, yes and no.
You can have characters that will be lookalike in all images, but there's still enough inconstancy for the characters to constantly change their hair cut and clothing. And, while globally the body shape stay the same, it's still fluctuating enough to looks weird.
This being said, perhaps that the difference are less noticeable when you play the game, and therefore can't directly compare an image to another.

But in the end the issue is still the same, I'm not sure that the time spent worth the pain. Between the time needed to find the prompt that will lead to an image effectively corresponding to the situation, the time needed to redo the bogus image, and the time needed to correct the image you'll finally use, is using an AI a help or a burden ?


And it's the kind of topic where I would have loved to have my fellow grumpy compatriot input...

 

[fyl3toys]

Interesting.

Sounds like it would be ok for say gotcha/gambling games that serve up single images, but anything trying to tell a story wouldn't be worth the hassle.

 

[Tompte]

Generating any one image is easy. Generating two images of the same subject is incredibly difficult.
SD is inherently random. It can be done, but it's harder than you think.

 

[MitFORKL]

Accordingly to the promotion screenshots from some AI game threads I end on by error, yes and no.
You can have characters that will be lookalike in all images, but there's still enough inconstancy for the characters to constantly change their hair cut and clothing. And, while globally the body shape stay the same, it's still fluctuating enough to looks weird.
This being said, perhaps that the difference are less noticeable when you play the game, and therefore can't directly compare an image to another.

But in the end the issue is still the same, I'm not sure that the time spent worth the pain. Between the time needed to find the prompt that will lead to an image effectively corresponding to the situation, the time needed to redo the bogus image, and the time needed to correct the image you'll finally use, is using an AI a help or a burden ?


And it's the kind of topic where I would have loved to have my fellow grumpy compatriot input...

I've been messing with Stable Diffusion for over a while now, and that's mostly my experience. It's the small differences that will break immersion. It's pretty easy to have a combination of checkpoints and Lora's/Embeddings that make the character consistent (even clothing, color, etc.). The problem is that for that to be "artistically viable" you will be spending a LOT of time fine-tuning it. For example, a character will have a jacket, in all pictures the jacket will be very similar, but the zipper will be in a different place. Also, you can force posing with Control Net, so now there are a lot more tools to make something consistent. But it's a LOT easier to just draw what you want, or set up a scene on DAZ or whatever.

I think AI is great for "extra" content. For example, artwork inside the game could use AI trained on the characters of the game. Or unlockables could be the characters with anime/cartoon/ultrarealistic styling, etc.

 

[Deleted member 1040613]

Accordingly to the promotion screenshots from some AI game threads I end on by error, yes and no.
You can have characters that will be lookalike in all images, but there's still enough inconstancy for the characters to constantly change their hair cut and clothing. And, while globally the body shape stay the same, it's still fluctuating enough to looks weird.
This being said, perhaps that the difference are less noticeable when you play the game, and therefore can't directly compare an image to another.

But in the end the issue is still the same, I'm not sure that the time spent worth the pain. Between the time needed to find the prompt that will lead to an image effectively corresponding to the situation, the time needed to redo the bogus image, and the time needed to correct the image you'll finally use, is using an AI a help or a burden ?


And it's the kind of topic where I would have loved to have my fellow grumpy compatriot input...

if you can get a character's face -- then yes you can customize hair with prompting - though it isn't as fluid as using a 3D program...if you don't mind minor changes then AI could be used.

I am skilled at creating AI art to the point that I can customize age, sex, hair color, eyes, poses, backgrounds, clothing, color clothing, etc etc. But I have no experience in creating a customized character...something I should probably figure out how to do...Just gotta get 15 or so images that look similar to get started...

 

[SuccubusSituation]

All sorts of reasons

>generating the same character in different poses is doable but very hard
>Having it drawn on layers for animation purposes is impossible
>No Ip rights
>Steam won't allow it.
Etc.

 

[FoolsBeGone]

Stable diffusion produces awesome imagmes that would great for these type of games. Why are there no major games with AI art?

I am still a relative rookie when it comes to AI but I would think part of it is getting the AI to actually give you what you are asking for. Most of the time its a fight with the AI to make it do what you want it too do which would become very frustrating if you are on a schedule to complete an update or to complete a game.

 

[Winterfire]

There are many reasons:
1- Hating people for using AI is the current trend, most people do not want to deal with crybabies that copy/paste the tears of artists from twitter without even understanding how it works, despite being very well document and accessible to everybody. Although this is not as bad as before anymore, since Adobe adapted AI, so now it's cool.

2. Consistency and the previously mentioned "fighting with AI", to create a coherent and useable image with AI you might spend hours on it, and as much time editing it if you have the skills to do so. It is not a fast process.
So if the choice is between this, or taking screenshots off KK/HS2, the choice is pretty simple, especially...

3. Copyright issues, if you create an image purely with AI, it will not be allowed on Steam, same as with KK/HS2 but with much more effort/time put into it.

4. AI art is actually not good. It may look pretty and colorful at first glance, but look at it for more than a few seconds to spot many mistakes. Nonsense stuff, things bleeding into each other, and very bad outlining.

In short, it seems worth it until you actually use it. You'll either spend way too long to make it look good and find it is not really convinient to use, or you'll use whatever you end up with and make a mess of a game that no one wants to play.
-Edit-
Also, KK/HS2 can be used by people with a potato. AI (local one) requires at least a mid-range GPU and online services are either limited or costly.