Why don't indie devs use AI-generated images as art?

[Humanophage]

Many indie games have hideous illustrations or deliberately avoid any images because they can't afford the artists, despite the fact that images would make it more enjoyable.

Why don't they use AI-generated images instead? Both portraits and backgrounds are already good enough. Give the specifications so that they are done in a similar style if you want uniformity, or even just slap some filters on - it's still miles better than most of the stuff we see. You can get extremely particular images within a minute.

This also pertains to things like interface elements such as icons or maps. E.g., generate a map and then move around the things like pictures of cities, boats, forests, etc.

It can be a little

, but often not, and it is still better than the awful stuff that you often see. For example, some generated backgrounds:

Figures:

Portraits are a bit tougher but can be fine if you cut off the odd bits or if it's the style (e.g., NWN has many odd portraits):

You can have consistent variations of the same character:

Or a present it as a race:

It also works with things like textures or interface elements - say, an illustration to an alchemy screen:

You can also make different units or figures in a similar style:

 

[MuffinBun]

Well... are they public domain?

 

[CyberWhale]

First and last batch of pictures - what apps and keywords were used to generate them?

 

[rusty_shackleford]

They sometimes do, it's just too new to notice it yet.
Definitely seen at least one game use portraits from that recent AI portrait website.

Well... are they public domain?

In USA, there's nothing limiting them from being released as such. There's been a few landmark cases that have found machine learning to be a transformative work .

 

[Bester]

Well... are they public domain?

Midjourney has always allowed commercial use.
Dall-e has allowed all commercial use a few days ago.

Why don't they use AI-generated images instead?

I offered Vault Dweller to use AI-generated portraits 2-3 years ago, when there was a project that generated really nice and diverse faces. He made nothing of it.

I'm currently generating images for my game, but it's far from ideal yet.

A few problems:
1) They're small. Upscaling algorithms suck and make the image total shit most of the time.
2) It takes 10-20 attempts to generate something that's not trash. Each attempt takes half a minute, so that's 10 minutes of just patiently trying different prompts.
3) The two I mentioned are bad at generating faces, hands and some other intricate details when they're part of a bigger picture.
However, there are other models that specialize on faces or anime faces - those are GOOD.

I can't say why nobody else is using them. I'm using. I'm probably one of the first to jump on that thing. More people will eventually, obviously.

I think artists are probably already using them extensively. If they can't generate a perfect image yet, they're very good at generating ideas, angles, color palettes, composition, etc. They excel at it already.

 

[bionicman]

Good point, OP. Now we just need machine learning to generate entire games for us, that way we won't have to make games ourselves anymore!

 

[MuffinBun]

In USA, there's nothing limiting them from being released as such.

Midjourney has always allowed commercial use.
Dall-e has allowed all commercial use a few days ago.

Huh... that's surprising. If this is the case, then I'll look into using them myself.

 

[Humanophage]

First and last batch of pictures - what apps and keywords were used to generate them?

They're all Midjourney. Last batch I generated myself with prompts like "mysterious Slavic wizard, in the style of Edmund Dulac" and "African goblin with a sword, in the style of Edmund Dulac".

For the first batch of pictures, it's just random nice images I took within 10 minutes from Midjourney chat. Prompts are things like "dark elf village", "fantasy world outside a cave", and other fairly basic things.

Good point, OP. Now we just need machine-learning AI to generate entire games for us, that way we won't have to make games ourselves anymore!

Well, that's a little futuristic, although I can imagine how it can help return to old-school RPG features where you type in your answers and the NPCs generate replies in accordance with that.

But art is already perfectly possible, right now. Wouldn't it be nice to have decent portraits in Vogel games, or attractive illustrations to each monster in KotC2? Or illustrations to units in an indie TBS?

 

[bionicman]

I don't think there's anything wrong with it as long as you can figure out how to make it all consistent and not be a jumbled mess of a bunch of completely different art styles. You could perhaps apply some sort of filters to all the portraits or backgrounds, etc, so it looks like the same art style.

The reason I personally didn't go for it (and I could be wrong about this), but I fear that it will be obvious who is using AI-generated portraits/art, kind of how it is obvious which game is made in rpgmaker today. But we'll see...

 

[CyberWhale]

But art is already perfectly possible, right now. Wouldn't it be nice to have decent portraits in Vogel games, or attractive illustrations to each monster in KotC2? Or illustrations to units in an indie TBS?

TBH, with a paid/unlimited version of this, anyone could make comic books now.

 

[Humanophage]

I don't think there's anything wrong with it as long as you can figure out how to make it all consistent and not be a jumbled mess of a bunch of completely different art styles. You could perhaps apply some sort of filters to all the portraits or backgrounds, etc, so it looks like the same art style.

The reason I personally didn't go for it (and I could be wrong about this), but I fear that it will be obvious who is using AI-generated portraits/art, kind of how it is obvious which game is made in rpgmaker today. But we'll see...

They all look fairly consistent so long as it's not wildly different things like cartoons, HD photo, and fantasy painting. It's also fairly easy to do so by adding "in the style of X". Some variation isn't really a bad thing - take M:tG illustrations, for instance. Or NWN:

At the moment, we're facing things like this - you could say it is a charming "f u" to high-budget games, but it would still look nicer with better and more illustrations (no offence to the artist, it's far from the worst).

 

[Zenithsan]

I'm definitely interested in making, at least, a short game using AI art.
What are the best generators? Are they free or do you need a license? What's the status on censorship, are there any banned prompts?

And another (perhaps the most important question) that I have.
Is there any reassurance that the generator isn't going to go offline while I'm in the middle of the design process? That would be a dealbreaker. Or if the devs changed the algorithms and now the style is totally different, making the art of my game inconsistent.

The standard game creation process can last years. I would need an "offline", static, version of the generator.

 

[Bester]

I'm definitely interested in making, at least, a short game using AI art.
What are the best generators? Are they free or do you need a license? What's the status on censorship, are there any banned prompts?

Adult content and gore.

For someone who's "definitely interested", you don't look like it. When I was definitely interested, I went on their sites and read up on it.

 

[MF]

I've started using it on occasion. Not to generate final images, but as a base or just inspiration. It's awesome.

 

[Bigfass]

What would make this awesome is if the image generator was part of the game engine, and it could reliably generate usable art 100% of the time. For those less boring replays.

Then eventually you could have the AI generate the levels, or even the story. Or, hear me out, even the rules.

I guess that's a few decades off still.

 

[Humanophage]

I'm definitely interested in making, at least, a short game using AI art.
What are the best generators? Are they free or do you need a license? What's the status on censorship, are there any banned prompts?

DALL·E 2 is quite good, but you need to be on the waitlist forever.

Among the regular ones available to common people without any fuss, Midjourney is the best. You're limited to 25 free images, but you generate them via discord, so you can just get more accounts if you want to test it. Otherwise it's $10 for 200 images or $30 per month for unlimited images.

 

[Üstad]

It takes time for a culture to be entrenched. And it's maybe to do with it's on experimental phase at the moment plus maybe copyrights stuff. Don't know how the latter works honestly, what's the source by the way?

 

[Zenithsan]

$30 per month for unlimited images.

Oh that's quite good. Right now I'm spending x20 times that on pixel art.

 

[Spectacle]

It's really just very recently that these generators started being able to make images that aren't

I expect we'll see a lot of AI generated art in games going forward, but since the tech is so new it will be some time before the games that use it are ready for release.

I've been using Midjourney to make quick illustrations for use in my D&D tabletop game, it's really cool how easy it is to get professional looking art, even if it can take some tweaking to get them just right.

 

[Chimera]

https://store.steampowered.com/app/1315610/

A moderately well-received rogue-like that uses AI generated art extensively to deliberately evoke an inhuman, Lovecraftian atmosphere.

 

[MadMaxHellfire]

links. links.

Spoiler

zwo, drei, vier, links, links

 

[The_Sloth_Sleeps]

I've not actually seen anything remotely capable, with non curated results. I've only seen YT videos/academic papers which supposed outputs which are to be treated as lies until you try the program yourself.

 

[Naveen]

Some probably are using it already, but hiding the source or editing the images to make them look more natural. Still, I guess it's mostly because it's a new thing, but also because unless you are making a game set inside the Warp or some weird reality, the AI-look will make it look, at best, uncanny, at worst, like everthing is melting. There's probably some fear of backlash, too.

 

[Mark.L.Joy]

Why not skip the middleman and just generate entire games?

 

[Humanophage]

I've not actually seen anything remotely capable, with non curated results. I've only seen YT videos/academic papers which supposed outputs which are to be treated as lies until you try the program yourself

The whole OP are images generated within a few minutes. What is wrong with this image, for example, that makes it worse than typical indie game art? Why are the two images on the right capable, while the image on the left is not? To my mind, it looks just fine, even very good for an indie game. It's not like they are selected at random - rather, the writer generates appropriate images and picks which ones fit the narrative, the monster, the character, the interface screen, etc. They don't have to be picked at random as some procedural exercise.

Existing art already looks even more like everything is melting, even in more high budget games.

Why not skip the middleman and just generate entire games?

Because it's impossible, while creating decent AI-generated images is perfectly possible. Surely it's more fitting than stuff like that or terrible 3D heads.