What AI-generated images actually cost (and what they replace)
I know what the title and subtitle suggest but this is not actually going to be talking about any of those things.
Im going to rant/spiel/waffle about AI art as someone who is not a creative or artist, nor am I someone who truly understands how ai art really functions in the first place. I know how LLMs work but for some reason, and without a care, Im going to talk about this without looking into that.
I will preface this with the fact that I use ai. This is not something I am going to be ashamed of, it helps me with my work, thought processing, and works as a better google function (but as per 2007 wikipedia rules, gotta click some links).
I don’t believe that individuals should be responsible about their ai use. This is the same concept as oil companies pointing blame on individuals for their consumption, without recognising the fact that a handful of firms around the world are responsible for a majority of the greenhouse gas emissions, no thanks. When a hedge fund or an data analytics company comes out talking about ai implementation, just know that they will have prompts, processes and workflows running constantly at a 24/7 cycle using tokens (examples).
I believe the role of government is to balance out the negative externalities of the “free” market and society as a whole, and therefore solving this issue needs to fall within the realm of the public. (Be it Mario’s brother in action, Greenpeace or political action (lol))
anywho lets get this shit show on the show road.
Entry points
I find myself thinking about the times I was in school, sitting around farting about, drawing these stick figures wars (Source). And id look over at the desks of my classmates and see someone absolutely rocking a photo realistic eye. And Id be in absolute awe. Like how can someone use a grey pencil to draw something that in some ways, looks more real than their own eyes.
and I took that as a personal challenge. An insult to my own capabilities and functionalities. A fucking reminder of everything I wasnt (creative) and everything I wanted to be (again, creative).
I’ll use a quote in the style of Michael Scott: “ “Humans create because it’s inate” - Rachel Barr “ - Jonny Thomson, Miniphilosophy (Source).
and I think that what this really highlights is stated further down in the quote but I learned that over quoting is bad juju from Mr Todd, my 9th grade English teacher. I think at least.
The inability to articulate the human experience is what makes art such an inate urge for people to follow. Art allows for a metaphorical understanding of the world (I know im talking about a photo realistic drawing of an eye here, but bear with me) to explain feelings that are not easy to put together. The guy who suffered hell on earth hitting you with a “shit sucks yo” may be able to draw an interpretive explanation of his feelings with a sharpie.
anywho back to the eye, so I went home and I hopped on google by typing google.com into the browser and searched for a realistic eye tutorial. And I made that tutorial my bitch, (I dont know which one I used but I found this one where I source the image and looks pretty good).
I spent hours and days even drawing those eyes. And for a short while, I went to class, and I drew eyes on my paper. Not stick figures. Not random scratches or filling in the gaps of letters such as P, Q and of course O. But eyes.
and I realized for a while that nobody cared. I cared about someone else’s eye but now others had seen this. And the person who I got “inspired” from, had moved on. They were doing a whole face now, like fucking moving goalposts am I right?
and I realized I didnt have what it took to be an artist in that moment. To draw things out is more than emotion. It’s more than inate. It also takes time and focus. Something I had none of as a (at the time) undiagnosed adhd child.
but looking back at it, was I inspired? Did this art come with any emotions or feeling? Did I 1000 monkeys on a typewriter indirectly role-play AI/LLMs? Like a 2050 kid going as a text input box for halloween.
Comparing outputs
Im not gonna go grab a bunch of AI art or generate a new one. But the cover photo of this article is from a picture I found on Pinterest.
Idk why I am sourcing things without validating true origin but give me a break please.
Anywho, I did some things. I took this photo, and I threw it into claude (it’s better, fight me on it) to describe the photo in an incredible level of detail. I then took this description of a photo and provided it to chatgpt (free plan, fight me again) to generate a photo based on that example.
now fuck, going back to what I said earlier let me do a bit of research of how AI image generation actually works.
A little searching, ai chatting, and christianity backed science articles later (saint Benedict: AI patron saint) we can understand that its really just working the same as an LLM does treating text and finding patterns but using the pixel representation for each point of a picture to recognize patterns. If we use the 1000 monkeys scenario, it’s 1000 monkeys with splatters of paint and a guy managing them / giving them direction. Also, there is something about diffusion where you add static to a visual until an image is destroyed. Then train a model to reverse static. which idk, anyway
you can see the other photo.
and I want to remind you that chatgpt never saw the original photo (although its on Pinterest so might be a part of its training materials but this isnt a statistical analysis, we dont need to focus on that too much).
the photos are different. Obviously, something something diffusion, something something monkeys.
but I want to talk about how they each make me feel. I picked the first one cause I was feeling like typing this out. I wanted a relaxing vibe, I wanted something that said, Im waiting, im chilling all good.
we can see the guy is standing now, we can see the car in the back is faced the wrong way. And I immediately feel a bit uncomfortable. People call this the uncanny valley effect. The road, the car, the spacing, it doesnt make sense to me. And it ends up throwing me off.
the piss yellow sky isn’t helping either. It does look like a filter or color grading on the original but on the latter, it’s the rick and Morty evil piss guy in action.
The reflection is missing from the car’s rear view lights. Im not gonna go into the prompt and see if claude didn’t mention it or anything (I did later, it didn’t)
but this may be a scenario of less is more. For some reason, the back of the car looks more real in the AI photo. Those headlights look more realistic on an older car variant like that. And that is where uncanny transfers to scary.
I dont think it’s impossible for Ai to create photo so realistic it ends up being difficult to detect. I know we have meta data and details of that nature but am I going to do that with every photo in existence? What if I take a screenshot of the photo and then upload it?
problems
Ai images have one core issue. I think that with text based AI, you end up with slop and slop of text endlessly long. And there isn’t much for us to do about that, twitter has been around for a lot longer than ai and we already had that problem. But images are worth a 1000 words (or in water usage for AI images, 10x to 300x more thirsty). And we end up with scenarios that could convince people the truth is different than you expect.
does this mean that behavior needs to change amongst people? Regulation needs to change amongst the permissions? Cats out of the bag so to say, in fact I’d say the cat is eating the bag. (Stay tuned for my rant on government structures).
I don’t have an answer, I don’t think anyone does. We could ban it, we could leave it. We could require watermarks to be implemented on all AI photos but think the real question is whats the point? I used AI imagery here. Ive seen articles that go against AI imagery, use them to prove a point (don’t do crack, here ill show you how to do crack so you dont mistakenly do it) and even ive seen articles talking about how insane the amount of water image Ai generation uses, use ai generated images.
Asher Perlman talks about this related to a cartoon he made. He uses different examples to show how AI can recreate his one scene cartoon. I think it’s a funny commentary on the situation but it misses a key point which is, does the reader of the sunday cartoons care? I have learned since growing up with Dilbert comics in the bathroom as reading material, it should be used as toilet paper. But ai generated comics making fun of Scott Adams are now, in hindsight, better than the original comics. But if was still on the toilet at 10 years old (Jesus Christ what a cursed scenario) would I care whether he drew it? Did I know who he was? Is the point of the cartoon to be consumed or to be created?
as an art connoisseur myself (stick figure war artist), I do think that it fails to match any of the value real art provides. Be it the consumption, the creation, the interpretation or the legacy of the artist, ai art fails to provide me with emotions. It falls short every time.
and perhaps this line alone will be my will smith eating spaghetti moment. One day I’ll find an ai photo that makes me cry instantly. Or puke who knows. But until next time, don’t give me your thoughts, don’t respond to this, please please please please please discourse is dead.