>>161523
>AI are just souped up versions of Cleverbot recycling the same things in its database
Eh, it's a lot more than Cleverbot, even if you say it's "souped up".
Cleverbot just straight-up recorded people's replies to it, and replies to replies, so if someone says "hey, how are you?" it'd pull up someone's reply from when it itself said "hey, how are you?", and it just kept on recording replies forever, and the devs could probably open a file and read all the replies ever made (I hope nobody posted anything important to it.)
AI works a lot more like a human brain, where concepts and "ideas" aren't stored in some easily-readable list in a folder, it's just a blurry mess of who-knows-what that just kinda... figures shit out on the fly.
There's hundreds of billions of images, low res, high res, mirrored, flipped, scaled, shrunk, labeled, unlabled, reposted, re-reposted, tagged badly, tagged correctly, all crammed into a 4-8GB file. You'd need some serious alien tech to compress art that small, there's more pictures "in" an AI model than there are bits in the file, and storing a full-res image in literally just a single 1 or a single 0 isn't possible.
The (main) reason AI art is so samey and regurgitated is mainly because people are just lazy shits that don't bother making a decent prompt. It's like commissioning a handful of artists on repeat and all you do with each is say "krystal big boobs anime" and sending over the money and not clarify further when the artist asks what pose she'd be in or what anime the art should be based on, rather than giving them more instructions like "I want art of Krystal from Star Fox, make her wear a purple kimono and have her pull it back to reveal her hugely inflated tits, and there should be a small electric pump hidden in the kimono, implying she's still inflating. Also make the whole pic in the early Naruto art style, and have Krystal leaning against the Leaf village's main gate, looking like she was waiting for you. Also put a pink flower behind her ear."
AI has posing tools, it's got inpainting, upscaling, loads of refining options, different controlnets and LoRAs to use, thousands of models to pick from, different types and versions of generators, even a whole-ass node system for the really nerdy guys, but most people just don't use any of this because AI lowered the skill floor way too far and people are lazy.
Lowering the skill floor of something is good and all, but then there's also no filter since now any rando can crap out 99999999999 different low-effort images in 10 minutes, so like, this whole thing is a double edged sword.