Concepting the Fortune Teller & Generative Tool Thoughts

Wednesday, May 31, 2023 at 09:44 AM

The Ultima VI Fortune Teller sitting in a mystical chamber with Tarot cards.

Concept of The Ultima VI Fortune Teller. (Stable Diffusion)

I generated this in Stable Diffusion using the analogMadness_v40 checkpoint, with a "color plate" (basically a splotchy unfinished SD image I saved, I have a folder full of them)" used in image-to-image with a .9 denoising strength.

The Ultima VI Fortune Teller. The Ultima VI Fortune Teller. The Ultima VI Fortune Teller. The Ultima VI Fortune Teller. The Ultima VI Fortune Teller. The Ultima VI Fortune Teller.

More concepts of the Ultima VI Fortune Teller. (Stable Diffusion)

Using a color plate and this checkpoint allows us to approach Midjourney quality in Stable Diffusion. That color plate makes a HUGE difference. These are the 7 best of about 150 images, though most are passable at this point, and I do see the ratio of garbage to useable images, improving over time. With Midjourney it's already almost 1 to 1, as in, every single image is awesome, but SD is getting there.

Update in September 2024: now Flux.1 is an amazing model similar in quality to Midjourney V6.
On FB UDIC someone said:

"Or...look up an artist. You can get quality stuff even from 30 bucks."

And here were my thoughts:

So, 200 characters x $30 = $6000 just to get ONE concept image each for each of the characters for the free fan game. Then I also have location concepts, stone, wood, metal, and foliage textures, food, fabrics, sand, mud, rocks, animals, weapons and tools, armor, monsters, particle effects, sky, cloud, and weather textures, etc. And that's assuming I use each concept as the final game texture as well.

Even if we assume I could get artists to do exactly what I want at exactly the skill and quality level I want in a coherent way across the game's assets - on the first try no less--, even for as cheap as $10 per piece, we're still talking about several, several thousands of dollars, right? (it might be cool to work up an exact dollar amount once the project is complete later on.)

If I could afford it, of course I'd do it, especially if the entire project was to create a single piece of art, then it would make sense.. But I also don't want to spend a decade saving up money for a single fan project when I have a dozen other large games i want to make before I'm dead, too. So when you say this, you're essentially saying "be rich or don't make the games you want, or you are immoral." Surely you can understand why that's not an appealing - or even viable - option to creative worldbuilders on a budget?

I have a more thought-out 7-page text document about my stance on generative tools like Stable Diffusion, but who knows when that'll be edited in a coherent way. And I'll present all that for feedback when the time is right for it. For now the basic idea is, I don't see any harm in using it for a free fan game, especially just for concepting, when I plan to hand-craft the 3d models myself, espeically when I could not afford to hire people anyway.

And as far as corporations firing their real artists and replacing their work with goobly ai stuff - the reduction in quality will show, I think. Then the real issue becomes "will general audiences notice or care, and if they don't notice or care, should we be mad at those audiences?" I'm not so sure that's the answer either. But the corporations aren't being evil either. They're just trying to maximize profits for shareholder while reducing costs - machine. So at the end of the day....

At the end of the day the real problem in that regard is the efficiency-seeking aspect of capitalism (which in many other cases is actually beneficial) and the passive consumer aspect of (what seems to be) general human nature when it comes to entertainment consumption. Neither of those things are evil, but of course we should try to ensure generative models are trained in ethical ways, and in the meantime I don't think we should place a huge onus onto random broke creators for using the legal tools they have available to them. At least that's my basic take. (Of course we'll find out how legal this stuff is as the current court cases are resolved, and hopefully we can get some clear legislation addressing the new field.

Discover, Not Creation

I also have this other strange idea that, like, images deserve to exist. The images in this post, for example, are really pretty! And I think a big aspect of the generative tools is shifting our thinking from "Look at the cool art I made because I am technically skilled!" more toward "Look at the cool image I found in the latent space of this model that we can experience together!" The same way we thrill at discovering stars, but astronomers never claim they "invented" those stars. And yet once they are discovered, we can all enjoy them together.

That attitude, that it's about "discovering" generated beautiful things, is something I explore a bit with generative music as well. Below I've added a playlist of music I generated using Udio. I never claim to have "made" the songs. In essence they are an extracted essance of all human musical creativity, something pulled from that latent space cheeseball, the soup of the collective human cultural subsconscious in binary form.

I didn't make these images or songs; I merely found them, and now we can all enjoy them the same way we might enjoy looking at a pretty orchid we discovered in a swamp. Some say they are just "soulless" but they sound nice to me! While imperfect (and generated in NON-English just to hide how terrible AI lyrics are) I see the potential in the songs and the themes to become something more in a project worthy of the "found songs."

Here's the playlist. Check them out and judge for yourself!

Here's the original fb UDIC post.

You've reached the end of this devlog entry.