A fascinating genre of AI-generated boomer art is crystalizing on Facebook before our very eyes. Ben Garrison is going to be put out of work. Another casualty of the AI revolution
this post was submitted on 12 May 2024
155 points (100.0% liked)
the_dunk_tank
15840 readers
698 users here now
It's the dunk tank.
This is where you come to post big-brained hot takes by chuds, libs, or even fellow leftists, and tear them to itty-bitty pieces with precision dunkstrikes.
Rule 1: All posts must include links to the subject matter, and no identifying information should be redacted.
Rule 2: If your source is a reactionary website, please use archive.is instead of linking directly.
Rule 3: No sectarianism.
Rule 4: TERF/SWERFs Not Welcome
Rule 5: No ableism of any kind (that includes stuff like libt*rd)
Rule 6: Do not post fellow hexbears.
Rule 7: Do not individually target other instances' admins or moderators.
Rule 8: The subject of a post cannot be low hanging fruit, that is comments/posts made by a private person that have low amount of upvotes/likes/views. Comments/Posts made on other instances that are accessible from hexbear are an exception to this. Posts that do not meet this requirement can be posted to [email protected]
Rule 9: if you post ironic rage bait im going to make a personal visit to your house to make sure you never make this mistake again
founded 4 years ago
MODERATORS
you are viewing a single comment's thread
view the rest of the comments
view the rest of the comments
Is it just me or is AI art getting worse with time
There's too much AI art on the internet now and it's getting fed into the models so now AI art is literally getting inbred
it doesnt help that this is being used by uncreative boomers to paint tasteless crap.
This is actually the biggest problem. It's boomers coming up with the worst ideas of all time. It's gotten a lot better at making things that could reasonably happen, but these boomers are just coming up with the worst possible images to generate, and because there's nothing even remotely close to that in the data set, it has to hallucinate harder. You can make a picture of guys playing baseball pretty easily because there are lots of pictures of guys playing baseball. Harder to make giant bibles being moved by cops because cops never move anything on their cars, and there's no such thing as a giant bible.
Oh yeah?
The bibles waved about grow bigger each time a fascist coup happens, maybe the next lurch at Bolivia's lithium deposits or Venezuela's crude oil will be as big as the AI image ones.
That is like 5% the weight of the Codex Gigas
The Codex Gigas is the largest extant medieval illuminated manuscript in the world, at a length of 92cm.
Cool.
It unironically has simulated mad cow disease
Huh, suddenly Digimon makes more sense.
It got significantly better, year ago there was no chance for anything even remotely readable and now you can sometimes get few words ok, like in the second pic. Also people on those pics have nearly correct number of limbs and fingers unlike before.
This looks like some SDXL checkpoint, maybe even the base version, so that'd be last year's big hobbyist local model. A midrange graphics card from a few years ago can churn out several of those images a minute, and a high end current gen card could bring that down to around a second (even less for SDXL Lightning).
This is probably the result of a completely uncurated spambot that's just trying to dump SEO shit into a prompt and post whatever comes out the other end. Like literally one single machine could be providing these images for dozens or hundreds of different spam bots because the local models make a local server that can be communicated with via an API by a GUI frontend or other applications.
Edit: checking popular SDXL checkpoints, I think it may be an older version of juggernaut XL, like the bot farm operator just grabbed the top ranked checkpoint from civitai a few months ago then just set up the bots for it and hasn't bothered to grab a newer one or fix any issues - why bother, when the whole thing is some baffling scheme to grift ad views from what seem to be entirely other bots.
It looks like Dall-E 3/Bing image creator to me. It's probably the most accessible and therefore the most used for this sort of thing, especially for weird chudposting
why do you think it's juggernaut XL specifically?
I just went and looked at the top rated SDXL checkpoints on civitai, and that's at the very top of the list and has a similar sort of plastic sheen to it. But after posting that I thought about Stable Cascade, and that both has an even more pronounced sheen like that and similar text adherence to some of the images, but that would be a weird one because it's more complicated to run and not particularly popular.
Someone else suggested it could be dalle, but I don't know anything about the quality or prompt adherence of remote proprietary models so I can't comment on that. The combination of them being strictly SFW and clearly entirely uncurated could point to an aggressively censored corporate model, especially if it was something with a free tier that bots could be querying themselves.
If part of the feedback it gets is positive interactions on social media, facebook boomers might be actively contributing to this lmao
It looks better if the person writing the prompt has several goes at it, refines the idea, and picks the best one. But Shrimp Jesus is spreading so fast that the people(?) writing prompts are just giving it one go and posting whatever it throws out.
The lack of taste doesn't help either.