I first dabbled with AI image generation back in 2022 and sprinkled a few such images throughout my worldbuilding project. It was easy to look past all of the flaws with the idea that it was nothing more than a novelty. And I never cared nearly enough about my worldbuilding to pay anyone for artwork of it.
Now that I look back at it, those images are obvious slop, which I’ve grown to dislike as much as the next person. But recent comments I’ve seen here and on other sites have made me wonder if my brain has rotted in the same manner that makes some boomers fall for AI slop. There will be videos where the use of AI is not very noticeable to me, but not with deceptive intent. Maybe an illustration to get the point across or a subtle two-second animation. Commenters will very passionately point it out. To be honest, I don’t see the creator either paying for the equivalent human work or drawing anything better themselves.
Does it really just look that bad? Is it an issue with what AI and the companies that sponsor it stand for? Theft of real artists’ work? Does it change at all if the images were generated locally with the creator’s own hardware and resources? What about upscaling images, like I do with old wallpapers so that they look better on new monitors?
I assume what I’ve just said will attract downvotes, but that was my thought process and I do want to understand where other people draw the line and for what reasons. Should we limit it to quick-and-dirty illustrations, pure novelty, upscaling existing images, a model that only incorporates work if the artist consents, or something else?
I think it’s always fine to use AI stuff as intermediary material, but almost never as finished product unless you specifically need something that feels samey and average.
For world building it’s pretty cool cause you can finally visualize that city that you’ve spent time imagining in detail, and use that to sharpen your description of it or spot incongruities. And for writing in general there’s nothing wrong with story boarding scenes and characters to ground your writing if you have more of a visual mind. Anything that helps you on your specific craft without bogging down the quality of the finished piece.
When teaching people how to spot AI images.
Never
blurry background art (eg. in a game like hollow knight, where there are thousands of images, i wouldn’t mind if like 100 of them which you barely notice wouldn’t have been made from scratch, saving peoples time)
removing ur ex from that one photo
I see no issue with locally run models like Z-Image Turbo. They take very little resources to run, so there’s no real issue with energy use here. They produce good results when prompted properly.
The whole theft of artist work applies to commercial models where companies are profiting of other people’s work. However, I simply don’t see the argument when it comes to open models that anybody can use freely, and especially in cases where you’re not producing images for profit.
There are different reasons against AI, it depends which ones you consider more important.
A) taking jobs away from artists - you can argue that if you would never have hired an artist for this anyway (throw away character portrait for a D&D game) then it’s justifiable.
B) That AI was trained on stolen content. - depends on your definition of stolen, this is the area that makes me think twice.
C) Environmental reasons - I try to generate stuff locally if I use it, but you still have the impact from the model being trained originally.
Basically NEVER
Apple has “GenMoji” and I find that a good use for it. It’s using emojis that they designed for the model, generating other emojis that you can only use on their devices (or sending an image). I use it to make new funny reaction emojis for me and my friends. I cannot use any art for other types of reactions the same way that emojis work (like reacting with an emoji for an image).
Learn using image generation properly rather than accepting slop and use it when you need it’s. Just like with LLMs generating text.
These tools are going to stay and they can be useful. Refusing to use them when they are useful for a good cause is not going to help anybody.
Also, problematic aspects of these tools are always related to capitalism. Artists whose work was used against their interests suffer because of conditions of our current socio-economic system. In a system that doesn’t make relationship with art alienated, it wouldn’t exist.
No.
Even if you manage to find a niche valid reason, even if you train on exclusively your own content, even if you run it on your own hardware powered by your own solar panels, it’s still normalising the technology for everybody else who won’t use it like that.
Honestly my main issue is that this gives more power to the corporations who can steal any content and then sell it as a subscription.
Even local generative AI is ethically sus, but as an occasional pirate I don’t have any moral high ground. I don’t really mind as long as its user in a way that doesn’t harm the jobs of the artists (which it often does).
Quality wise, realistic images are often sloppy but I too am starting to fail to notice sometimes, which doesn’t feel good if it’s something that can misinform me. Recently I played for some hours with an anime based model, and I have no chance telling what is real there. But it still really drew home the fact that it lacks some of the artistic expression. No matter how much you write in the prompt you can’t control all the details with intent like the artist does, and you’re at the mercy of the model on anything that you leave to interpretation. And if the model doesn’t have enough data on what you want it will just break down. So it only works as long as you have a relatively vague idea and don’t care about the specifics. Which is why I also think it can’t replace anyone as long as the audience cares about intent being there and being consistent (which again often they don’t care).
Anyway, I consider these models to basically be creating interpolations of existing art now.
so, my opinion is, when i want to showcase something, i do the following:
- first, think about what i actually want to show
- second, do some sketches on paper to get a feeling for the thing (often helps with thinking about the thing)
- third, to a cheap crappy drawing digitally, mostly libreoffice impress or krita
- then, ask the AI to basically make it more alive by adding details etc.
Wat. Details are what AI is worst at.
yeah it’s visible that it’s AI but the stuff is still recognizable
meanwhile my own sketches look like this:
also recognizable but less graphic --> less catchy.
also in case anyone is wondering: the greenhouses are well-rendered by AI. it’s not a mistake, i specifically asked it to draw it like this. here are some more:
greenhouses for algae
the reason why it looks like this is because it’s essentially a flat sheet of plastic filled with water, to grow algae in. algae can be surprisingly nutritious actually. (comparable calories to bread)
Goodness, those solar panels have been battered. And somebody spilled their spaghetti all over the ground.
So what purpose is this supposed to serve?
well i originally did the images for the !mars@discuss.tchncs.de community to have some nice visuals to go with. just like pepper makes the food tastier, even though it serves no role in diet, so do visuals make discussion more lively, even though they’re not technically necessary.
anyways, after a significant backlash against AI (that btw i still don’t get but whatever) i stopped posting these images, now they’re sleeping on my disk.
When the model isn’t built on stolen licensed works and built/run in a way that isn’t in conflict with the community where the data center resides seems like some basic requirements to meet.
Hypothetically, any time you’d not be able to have a person do it.
Thing is, all the current models out there are built on stolen talent. So you have to decide if that matters to you.
Me? I don’t hate generative software per se. I think it does fine for the exact cases you mention.
