AI and Development of CRUFT

Hot Topic. TL;DR - I decided not to use AI.

CRUFT has a lot of sprites. I am not a game artist. I paint little army men in my spare time but I really don't have an art background at all. It seemed to me that employing some amount of AI to generate a bunch of sprites was a sound idea so I did some experimenting. I mainly tried out Adobe's AI features and spent some time over a week or so trying to get a basic style together so that I could just type in "oak log" and shazam free art. What I ended up with was a bunch of different sprites that were superior to what I could create myself, but had no thread between them. They looked like they were made by a few thousand different people because they were.

Deciding not to use AI is a little weird to me. I like AI, in many respects, in both theory and practice. But when it comes to Art, I ran into a little hitch in my thoughts and feelings. Nothing looked quite right in concert with everything else. Nothing at all looked like I actually had a hand in creating it.

I am VERY experienced with wanting something to look a certain way and having someone else do the lifting. Beyond that, I'm most used to what I say being fractally interpreted through many people who each do different parts of the process to bring art to life. The process I am most used to looks a little like this:

1) I write down a description and hand it off to an Art Director.

2) Art Director gives it a look.

A) They always at least put a twist on it.

B) Sometimes they have serious concerns and punt it back to Step 1.

3) Art Director works with their multi-disciplinary team and shops out the idea for feedback.

A) They always at lest put a twist on it.

B) Sometimes a specialist will have serious concerns and they punt it back to Step 2.

4) Some artist or artist works on a draft version that can be critiqued and massaged a bit.

A) The AD and I always put a twist on it.

B) Sometimes you prove out that the current concept needs an overhaul and punt it back to Step 1 or 2.

5) Some artist or artists make real stuff.

A) Everyone and their uncle comments on it and wants to change it fundamentally, except for the artist, the AD, and me. We filter the feedback but hopefully stay in Step 5. Maybe we go back to an earlier step as appropriate.

B) Integrate the art into the game.

6) Ship the game.

7) Profit... maybe.

I'm pretty solid with prompts, but handing them off to AI rather than an art team just didn't ever get to anywhere near Step 7. There is a lot of tweaking that not just does happen, but needs to happen. Even with all the different people involved, I've always felt like that game art that I thought of in Step 1 was reflected in Step 7 when working with real people. There's civil arguments, there's a lot of horse trading, there's a ton of technical back an forth... there's a team of artists watching me struggle to explain what I mean when I describe a boat as being too somber. All that back and forth is what gets you where you want to go.

One might say that I just need to improve the prompts I give the AI. I've messed with Midjourney and I've prompted some pretty rad pictures of some really cool stuff. They are all one offs though. Even when you drill into a single piece and tweak the prompt like crazy or genetically pair the ones you like to try to get good offspring each one looks like you "hired" a million artists to paint you a picture and you eventually just picked one you liked the most. By "hired", I mean paid nothing.

Therein lies the problem, and the same is true for game design and film-making right now. You can definitely ask for a million options and choose the ones you like. You can definitely put them together as best you are able and string together a coherent narrative. What you can't do right now (at least I can't) is give anyone a reason why this game, or film, or even just sprite actually matters in comparison to its peers. It just ends up feeling like a pile of stuff.

I'm pretty sure that the chemistry of human idiosyncrasies was far more important to the best art I've been a part of creating than the skill of anyone involved. I'm not downplaying the importance of skill in making great art. I've worked slews of amazing artists that are deeeeeeply skilled. What got us to Step 7 though, was how we worked together. Same is definitely true of game designers. Painfully true there.

Sooo... After I decided not to use AI, I sucked it up and had to make the sprites myself. Art is hard. That said, once my wife saw some of the garbage i was churning out she said she'd be interested in giving sprite-making a shot as well. She's better at it and I am definitely faster... we each have a role. So we have a team of two spare-time-no-art-background folks making our sprites. We go through the same process I listed out above. There is far less skill cooked into the participants (though we are getting better every day) but I gotta tell you the results are way faster and more in line with expectations. It feels like our game.

It's a pretty cool pig.

I'm stuck with what we can come up with. Learning as you go means the first stuff you made is leagues behind where you are today. Is it consistent? No. Is it top quality? No. You know what though? It feels pretty good to make your own stuff. Maybe some day I'll have the budget to pay another human being to make sprites. That would be awesome and much better still than using AI, even if less cost effective. The act of making and critiquing and collaborating is a joy unto itself.

I absolutely respect the work in AI that leads to amazing results that are just not possible to scale with human effort. AI is a tool though, not a creator. Maybe some day soon I'll eat those words, but when that day comes I don't think I'll even want to use it for my own purposes. I want what I make to be mine and not crafted on the backs of hard-working strangers. I'm not doing crazy compositing work to make it look like I'm on mars. I don't personally hate AI but the absolutely crazed push to use it for everything just seems a little wild to me.

The simple fact is that I can, and do, suck it up and just push pixels when I need more pixels. I don't need to pause and ask for help from an AI, so I don't. I can just do the thing. I make pigs poop and I can do that with these hands. You can quote me on that.

poop

PS - While writing this up, an AI is constantly hounding me to let it write for me. I can missspell things myself, thank you very much.