Bring Back Nauseating AI Art!
I've had a fascination with AI generated things for a few years now. It started with the vinesauce streams on web toys like thispersondoesnotexist and Pix2Pix, which I found really entertaining and interesting to play around with. You'd get these odd masses of skin and hair intermingled with the occasional recognizable face, which itself was often hideously contorted or dripping all over the place. Sometimes the images were straight up nauseating to look at, in a way only AI art can be. Over the years, new services where you could play around with image generation like deepdream generator or artbreeder kept popping up up. These were still rather basic, allowing style transfers and gene-based image generation, which mostly led to vague images with barely any clearly recognizable scenery. Artbreeder (or Ganbreeder, as it was then called) was mostly useful for generating ambient, ethereal images. [1] Quite a few images on this site were actually generated in this manner, mostly backgrounds and textures. It worked well enough for someone like me, but generally, AI art was seen as little more than a bizarre curiosty by those who were aware of its existence. Until around autumn last year, when there was a sudden leap in Image Gen technology. And now we have working artists fearing for their livelihood rallying together against AI, the AI supportors joining ranks with the techbros and NFT losers, and everything's fucked.
The issue as I see it is that with the increase in the AIs capabilities for generating usable "Art", the focus has shifted away from the specificity of AI art to its potential for passing as "real" and therefore being a cheap alternative to hiring actual humans to draw your stuff for you. [2] All recent developments in the AI world point in this direction. From what I've tried out, Midjourney is particularly egregious in this regard: it will spit out generically "beautiful" images no matter what you tell it to do. [3] But all of this "art" is trash. It doesn't leave a lasting impression, there is no actual intention behind the images, just a chain of words denoting style and content, which the AI will extract and rearrange from its database to ultimately produce a derivative half coherent piece that might do well on reddit. This not only has horrible implications for the future of working arists and the media industry, it's also extremely unfortunate for AI art itself. Because I do think there's a place for AI image generation in the art world, as a tool with unique functions and capabilities, as a supplement and not a replacement.
By leaning into the quirks, peculiarities and "weaknesses" of AI image generation, one can achieve very unique effects and results that aren't really possible with human art. Particularly interesting are AI "photos", because even if they look coherent, they can't quite match the thing you're telling them to depict.[4] This results in a unique photorealistic irreality. It gets even stranger if you try making videos/animations. Check out this one I did for "Sally Valley", my crappy little "game" from last year which I should really get around to editing cause it's basically unfinished:
It's also a good idea to try pushing the AI out of its comfort zone through your prompts. instead of "beautiful impressionistic sexy goth woman trending on artstation", just freaking mash your keyboard. Go to the website of a local newspaper and copy a random sample of one of the articles. Put in the code for this website, i don't know. Go nuts. And if you do have a specific idea and want to add a bunch of adjectives and adjust parameters to ensure the best possible result, why not at first try going in raw? just enter the simplest, stupidest version of the prompt you can imagine. You might be surprised. Here's another image I made for Sally Valley, depicting the story's flayed protagonist.
It's horrifying and gross, a hulking shapeless mass of red flesh with the texture of a carrot, an unreadable expression formed by skeletal black facial features that might just be scabs... The prompt? "skinless son of a bitch." Sometimes, it just works.
But yeah even then I'm just making these for personal just-for-fun garbage projects which I don't plan on monetizing, and there's still a mountain of images being used without consent to create something similar for which I cannot be sued. Which is obviously BS. The only way around this I can think of is making the use of AI art in commercial works illegal, but that would stop my concept of AI art as a tool in the professional's toolbelt from coming to fruition. Which would be a shame, but also quite a small price to pay for the assured quality of future art and job security of future artists. I wish we could have just stayed in that sweet spot of non-threatening early 2022 AI capabilities, man.
- some things that use AI generated images in cool ways:
- this amazing rpg maker game features creepy ai-generated videos you can find around the game world
- this artist makes cool and icky semi-liquid monstrosities based on ai-generated images (thank you to lataccola for sharing this one with me!!)
- 3d comedy exploration game using textures with ai-generated words to simulate an indistinctly slavic nightmare world. Featuring ai generated images of french president macron eating peas
[1] ...And furries, for some reason. Presumably just a huuuuge part of the data set.
[2] Which is made so much more bitter by the fact that these things run on datasets unconsentingly comprised of the work of the very artists whose jobs they are prophesized to steal.
[3] Sidenote: When I was trying it out in their discord server, the channel consisted mostly of lanky men posting topless selfies as bases and setting "buff cool sexy anime character" as their prompts lmao
[4] As an example for this, simply try generating "flounder photo" with Nightcafés Stable diffusion (preferably on v.1.5 or lower) and enjoy an array of Weird Fishes. This is how I'm making enemies for the rpg I'm working on.
