(Thread IKs:
Nuns with Guns)
|
get his rear end! kill him!!!
|
# ? Apr 27, 2023 19:26 |
|
|
# ? Jun 6, 2024 23:49 |
|
IShallRiseAgain posted:I disagree. It still leans into the narrative that AI is just a collage and its just searching up images and combining them. AI learns art in a similar way to people. All art is just combining visual concepts in novel ways. It also mostly ignores that there is always a human behind the AI creating the art, aside from one token mention of it. The issue is that AI art is still in its infancy and fine control is still not very accessible for the average user yet, but there is a lot of progress being made in this area. The technology still isn't quite at the level where its very practical to integrate into a project yet, but the technology is progressing at such a fast rate its hard to keep up. In a few years, I can see it causing a boom in independent projects allowing people to create the type of art that previously required a large group of people that could mostly only be created by a corporation or other large organization unless a person is willing to work themselves to death. There is a flood of low quality AI art because people who previously who didn't have the ability to communicate via art due to a lack of technical skills have now given been the tools to do so. I'm going to take a different tack than other people are doing: I don't think that video is actually about the values/lack thereof of AI art. It's a video about the kinds of people depicted in The Player, a film that has nothing to do with AI but everything to do with the content mill industry that seeks only to churn out as much content as possible as cheaply as possible. Eyebrow Cinema doesn't say that AI generation doesn't have value - he even positively mentions an art piece that heavily relies on it - it's that the shift of online art towards being "content" whose only value is to be aggregated was already extremely harmful, and AI generation is only going to exacerbate it. You mention the flood of low quality AI art by people excited about the tech; it's important to note that if there is potential profit to be made in it, that flood will never stop. Search Engines have already been swamped to nigh unusability by the flood of AI/form articles, and the AI algorithms are getting so good that now they're threatening everything. And the thing is, I actually love AI generation. It's behind many extremely fun videos of recent years, like The Dark Age of Gooby. But this clearly isn't going to be limited to goofy streams and gimmick twitter accounts; powerful moneymen want to see if they can supplant as many artists as possible, and a particularly obnoxious potential future is an entire media landscape of big budget AI-made blockbusters starring AI copies of long-dead actors and social network timelines churning with AI-made dross, occasionally sprinkled by art a human threw in to drown in the flood. It would be an era of "total artistic liberation" and it would, frankly, suck rear end. Pigbuster fucked around with this message at 20:09 on Apr 27, 2023 |
# ? Apr 27, 2023 20:04 |
|
BlankSystemDaemon posted:EDIT: Beaten, at least partially. I think you did a much better job of summarizing than I did, honestly. What we call "AI" is ultimately reliant on huge amounts of human-curated data and then being handheld through that data by humans until it combines those pieces of data into answers (or "art") that is agreeable to those humans training it. The "advances" in AI lately have all just been in service of improving the statistical model and its ability to land on agreeable answers - as well as coding in checks to paper over its more notorious flaws and quirks - but nothing I've seen indicates we're moving towards anything resembling a true intelligence. The facsimile is just getting more polished.
|
# ? Apr 27, 2023 20:15 |
|
Pigbuster posted:You mention the flood of low quality AI art by people excited about the tech; it's important to note that if there is potential profit to be made in it, that flood will never stop. Search Engines have already been swamped to nigh unusability by the flood of AI/form articles, and the AI algorithms are getting so good that now they're threatening everything. i've already seen artists lamenting that the places they use to find photo references are increasingly being polluted by AI generated images of unknown providence that may or may not reflect reality, and that's just the images they realise are AI generated using a wide-spanning scrape of the internet as a representative sample of human generated content worked okay up to this point, but things are going to get weird as the internet increasingly gets filled with AI generated content that the next generation of AIs, and even humans that aren't paying close attention, are trained on
|
# ? Apr 27, 2023 20:22 |
|
Somebody already mentioned the usage of underpaid 3rd-Worlders to make AI (ChatGPT for now sure but I can easily see the extrapolation to AI Art too) more palatable, but there's already been human precedence for AI Art: https://www.youtube.com/watch?v=h4hP6nOB1dc
|
# ? Apr 27, 2023 20:27 |
|
Doesn't it show that the artwork was generated with the help of an "AI" somewhere in the code or even metadata, couldn't it easily be blocked from being uploaded to these art-sites if that's the case, sure, there's ways around that but it would stop the most lazy of creators.
|
# ? Apr 27, 2023 20:29 |
|
openAI made a token effort to distinguish DALL-E generated images by putting those little squares in the corner, but i don't think any of the other generators do anything like that most AI generated images aren't even coming from DALL-E now because it costs money, and the other generators are either free to use (because they're burning VC money) or free to run locally (stable diffusion)
|
# ? Apr 27, 2023 20:34 |
|
repiv posted:i've already seen artists lamenting that the places they use to find photo references are increasingly being polluted by AI generated images of unknown providence that may or may not reflect reality, and that's just the images they realise are AI generated It just feels like another step towards the apocalyptic vision conjured up by Dan Olsen's video about those weird kids' videos that were floating around YouTube. It's all the product of systems blindly chasing "growth" so hard that they forget why they're actually doing it. It's monetization grey goo. Just visions of an internet inhabited only by algorithms creating content based on the algorithmically generated content that's getting the most views from algorithmically controlled bots to make the numbers in a monetization algorithm go up.
|
# ? Apr 27, 2023 20:38 |
|
Apropos of AI stuff this looks horrifying. Like Angela Anaconda does Bebop. https://twitter.com/RataUnderground/status/1651149275944873985?t=7x0o0LInLwFMRfDLJweWPA&s=19
|
# ? Apr 27, 2023 20:43 |
|
repiv posted:openAI made a token effort to distinguish DALL-E generated images by putting those little squares in the corner, but i don't think any of the other generators do anything like that That's good on DALL-E at least, just hope it becomes standard when coming to other generators but I doubt it, anyway, I'm no coder so I wouldn't know but couldn't you create an algorithm that recognizes if something was made by an "AI" at least, fighting algorithms with algorithms
|
# ? Apr 27, 2023 20:44 |
|
BlankSystemDaemon posted:The algorithm doesn't think. I didn't say it was alive lol. I said it learns to create images in a similar way that humans do (not identical). There is a huge difference. It doesn't require sentience at all. Its artificial intelligence not actual intelligence. It learns because it process training data to find patterns from images (like human artists do even though they wouldn't describe it like that). Every piece of art in existence is based on existing art or something from reality. You can argue that it also doesn't attach any meaning to the art (that's true) and doesn't actually understand what it is making (that's also true), but it learns the actual mechanical production of art in a similar way to humans (again not identical). It doesn't just store the art, and its actually impossible to do so considering the size of the image gen models and the size of the images in the training set. IShallRiseAgain fucked around with this message at 20:54 on Apr 27, 2023 |
# ? Apr 27, 2023 20:49 |
|
"not identical" is underselling it
|
# ? Apr 27, 2023 20:54 |
|
fun hater posted:as someone who draws for a living and also actually understands how stable diffusion works, this is literally the stupidest and most braindead poo poo anyone could ever genuinely put forth. ive got a mechanical turk out back you might be interested in investing in stable diffusion is, in fact not a collage of existing art, nor is capable of "training" on the data its generating. people who are pro-AI art (and im not even strictly against it conceptually, i just cant believe that poo poo came flying out of your mouth for real in defense of it) love to bring up these misconceptions as epic gotchas because artists are really stupid cowards who think that learning about how a computer program works will kill them, or something. so any number of these insane random assumptions are the only points that ever get argued and im so loving sick of it. i am not going to debate whether the capacity for "thought" exists in a 2023 computer, because im not a loving moron and can easily see that humanity cant even make electronic translators that work accurately. use google translate and then have the audacity to tell me a computer can think lol. anyway we arent talking about google. we're talking abt stable diffusion. im pretty sure it was a video linked in here that had a great description of how it works. this is the most simple explanation that anyone who who has enough computer knowledge to make their way to something awful dot com can manage to understand. https://www.youtube.com/watch?v=F1X4fHzF4mQ when a prompt is entered, the computer is just pulling words and captions its crawled from the internet and assembling these concepts into a single "coherent" (if steered in the right direction after enough hours) image from a base of black and white static. the static base is why there's a "house style" for something that should extremely varied instead of churning out the most generic conceptart.com circa 2006 rear end art. its why the lighting never deviates from the soft, yet omnipresent perma-glow that comes from no direction; the computer doesnt know what lighting is, what lighting does to the mood/tone of a piece, or even how to apply it. the colors of every image are the same palates of earth tones because dall-e doesnt know how to use colors and it doesnt understand how the human eye perceives them. the computer can index hex codes, but it doesn't understand what they look like. its creating data and then plays "hotter/colder" with you until you get a result that YOU like. dall-e keeps creating people with too many fingers because its a loving computer that does one thing well; there's no intelligence at play or else the computer would be able to know the basic fact that all humans know about their own default anatomy. its not thinking. thats what you and the computer have in common saying that the human behind the prompt is the artist is also laughable; if anyone is the artist its the scientists who released dall-e, which i truly do think is a really interesting little bit of tech. but the people typing in garbage didnt make anything lol. they're just using a commercial art product. its like thinking the etch-a-sketch will change the game for artists and non-artists everywhere because it's easier than ever to erase. or that typing a google prompt makes you a writer. people are not going to willingly choose a medium that looks like dog-poo poo (it really truly does and the more accessible it gets the less desirable the "style" will become) and it will end up being used exclusively by the same people who made fetish comics using poser models in the mid-2000s
|
# ? Apr 27, 2023 20:57 |
|
IShallRiseAgain posted:I said it learns to create images in a similar way that humans do (not identical). [...] It learns because it process training data to find patterns from images (like human artists do even though they wouldn't describe it like that). but it learns the actual mechanical production of art in a similar way to humans (again not identical). just a stunningly ignorant series of statements that exposes a fundamental misunderstanding of how humans create, what artistic intuition is, how skill is developed, and the purpose of why humans identify/create patterns in art lol e: like. a computer is creating an arm by copying other arms its seen before from a big folder labeled "ARMS" and creating a pastiche of human anatomy from those examples. i draw an arm by conceptualizing the bone and muscle structure underneath as a result of a decade of thinking about how the muscles work and observing my own arm as reference. its not even remotely the same. fun hater fucked around with this message at 21:06 on Apr 27, 2023 |
# ? Apr 27, 2023 21:03 |
|
There's probably potential for the development of actual artistic tools, that are new and actually useful for human artists. I think it's just that this tech is at the earliest steps. Like if everyone was saying computers will revolutionize all kinds of art in 1983. Sure, given enough time it would be true, but the apple IIe wasn't going to be the step that does that.
|
# ? Apr 27, 2023 21:11 |
|
can't believe im saying this but would anyone like to talk about you tube content creators
|
# ? Apr 27, 2023 21:45 |
|
Ngl ai chat is pretty interesting. I like it as a tool for comedy and the people who put it up as their only means of expression interest me in a conversational way. I say let them fight
|
# ? Apr 27, 2023 21:50 |
|
Feels Villeneuve posted:can't believe im saying this but would anyone like to talk about you tube content creators Does Vox's Youtube vids count as that platform's content creators? I, uh, I thought their recent AI videos were interesting: https://www.youtube.com/watch?v=y2BVTW09vck
|
# ? Apr 27, 2023 22:01 |
|
Feels Villeneuve posted:can't believe im saying this but would anyone like to talk about you tube content creators I just found a person who does videos on Scene bands from like 2008, they did a 3 parter on some guy I never heard of, but it was entertaining, also bonus points that they still have the 2008 hair in this video: https://www.youtube.com/watch?v=YiVfAO24bxc I never followed any of these bands but I’m suddenly fascinated by their inner politics. If you watch all parts be aware the subject is a total scumbag and there’s a lot of sexual/mental abuse this person inflicts on their partners in part 3.
|
# ? Apr 27, 2023 22:03 |
|
IShallRiseAgain posted:I didn't say it was alive lol. I said it learns to create images in a similar way that humans do (not identical). Dude, we know basically gently caress all about how human brains learn.
|
# ? Apr 27, 2023 22:12 |
|
IShallRiseAgain posted:I didn't say it was alive lol. I said it learns to create images in a similar way that humans do (not identical). There is a huge difference. It doesn't require sentience at all. Its artificial intelligence not actual intelligence. It learns because it process training data to find patterns from images (like human artists do even though they wouldn't describe it like that). Every piece of art in existence is based on existing art or something from reality. You can argue that it also doesn't attach any meaning to the art (that's true) and doesn't actually understand what it is making (that's also true), but it learns the actual mechanical production of art in a similar way to humans (again not identical). It doesn't just store the art, and its actually impossible to do so considering the size of the image gen models and the size of the images in the training set. Again, dumb as hell.
|
# ? Apr 27, 2023 22:30 |
|
https://twitter.com/waypoint/status/1651702008074629121 With Vice's staff getting shorted, this included the death of Waypoint.
|
# ? Apr 27, 2023 22:43 |
|
Kunster posted:https://twitter.com/waypoint/status/1651702008074629121 To be honest I didn't even know what waypoint was before now
|
# ? Apr 27, 2023 22:54 |
|
Kamrat posted:To be honest I didn't even know what waypoint was before now lol same
|
# ? Apr 27, 2023 22:57 |
|
Kunster posted:https://twitter.com/waypoint/status/1651702008074629121 Well that sucks, I just started listening to them.
|
# ? Apr 27, 2023 22:58 |
IShallRiseAgain posted:I didn't say it was alive lol. I said it learns to create images in a similar way that humans do (not identical). There is a huge difference. It doesn't require sentience at all. Its artificial intelligence not actual intelligence. It learns because it process training data to find patterns from images (like human artists do even though they wouldn't describe it like that). Every piece of art in existence is based on existing art or something from reality. You can argue that it also doesn't attach any meaning to the art (that's true) and doesn't actually understand what it is making (that's also true), but it learns the actual mechanical production of art in a similar way to humans (again not identical). It doesn't just store the art, and its actually impossible to do so considering the size of the image gen models and the size of the images in the training set. Also, have you ever worked with statistics? If you have, you know that people have no "innate" concept of it - so unless you think we somehow subconsciously do something we can't consciously do, and can provide an explanation for the mechanism of it, I don't think there's much to go on along those lines. And you're doing pattern seeking again; stop mistaking it for intelligence. It's not artificial general intelligence as depicted in science fiction - it's a simple algorithm with a huge training set. The field of "AI" was born in 1956 (when Frank Rosenblatt implemented an early demonstration of a neural network), and it hasn't fundamentally changed since ELIZA in 1964, which was the first time someone thought that a computer was exhibiting human-like characteristics - the biggest reason we haven't seen this before now is simply that until fairly recently, servers couldn't fit enough memory to handle the data without paging things to and from disk. Doing any kind of paging is impossible if you're working with anything that needs to scale, because the fastest auxiliary storage is still many orders of magnitude slower than the CPUs cache or main memory.
|
|
# ? Apr 27, 2023 23:04 |
|
Look, I don't want to continue this thread derail, but I never once said it was an AGI, or it having any sort of actual intelligence. Its not. You'd have to be an absolute idiot to believe that. Stop saying I believe that because its not true at all. I call it learning because its able to use training data to implement new functionality without being explicitly being programmed to perform that functionality. It can detect image patterns, and re-use them but its not just searching for images and combining them. All I wanted to do is point out that its not a "collage" like that video seems to imply. I am well aware of how it actually works. Its a chinese room situation, its not aware of anything. Its only capable of having a rudimentary facade of intelligence, which I believe can be useful in the future with a lot of refinement. It will be a very long time before we ever approach a true AGI and probably not in the lifetime of anybody posting here. Here is a new PushingUpRoses video about a pizza romance movie to get the thread back on topic https://www.youtube.com/watch?v=LP8tGRmSsDw
|
# ? Apr 27, 2023 23:35 |
|
BlankSystemDaemon posted:We have no loving idea how we learn, so how do you propose to compare it with something using statistical modeling? Also, feeding data it generates into itself is a the equivalent of sticking a microphone in front of a speaker. https://youtu.be/QMIKLp8eGvM https://youtu.be/oUhfkaVUPY8 I'm not reading the rest of that.
|
# ? Apr 28, 2023 00:07 |
|
ai art is for creating novel pictures of things you could never capture in real life, like steven seagal as a bloated sheriff. https://www.youtube.com/watch?v=Hh-i3FTUZzA
|
# ? Apr 28, 2023 00:10 |
|
Feels Villeneuve posted:can't believe im saying this but would anyone like to talk about you tube content creators Hell no. AIshallriseagain
|
# ? Apr 28, 2023 01:07 |
|
fun hater posted:and it will end up being used exclusively by the same people who made fetish comics using poser models in the mid-2000s Not sure why you're speaking of those people in the past tense..........
|
# ? Apr 28, 2023 01:58 |
|
John Murdoch posted:Not sure why you're speaking of those people in the past tense.......... Cause they've already moved on to AI
|
# ? Apr 28, 2023 02:12 |
|
On the topic of AI, Criken has been doing streams of the Pokémon Infinite Fusion mod while talking to an AI Professor Oak, who's mostly there to criticize the ethics of every single thing Criken does. They're my favorite streams of his in a while. https://www.youtube.com/watch?v=p-E-AYm7PtM
|
# ? Apr 28, 2023 02:26 |
|
John Murdoch posted:Not sure why you're speaking of those people in the past tense.......... they either died as a result of beginning their project at the ripe age of 50 or are still chugging along on topwebcomics.com
|
# ? Apr 28, 2023 03:11 |
|
I think AI has its place as an artists tool, to create a rough idea of what you want or a general idea that you can use as kind of inspiration. Maybe some real small studios who can barely afford artists, I mean its not like some people don't just rip textures from whatever they can find through google anyway. But the way the big studios want to use it to save costs and churn out barely coherent garbage like that AI "anime" thing is p bad.
|
# ? Apr 28, 2023 08:18 |
|
I saw one artist in the gbs thread...not arguing, per se, but just saying that AI is useful for them for quickly visualizing something in their head they can then use for concept art to make their own thing from. The gbs AI thread is pretty neat with people futzing around with ai stuff either for stupid bullshit or making neat looking stuff. This post by barons cyber skull and thr ones that follow it are fascinating to me, taking a smaller image and blowing it up while adding greater detail to things: https://forums.somethingawful.com/showthread.php?threadid=4000251&pagenumber=277&perpage=40&userid=0#post531149316 Pigbuster posted:On the topic of AI, Criken has been doing streams of the Pokémon Infinite Fusion mod while talking to an AI Professor Oak, who's mostly there to criticize the ethics of every single thing Criken does. They're my favorite streams of his in a while. It's not a mod, it's an rpgmaker thing I think? Neat though, really ambitious to try and make hundreds of thousands of combined form sprites for every single pokemon combination. I think they're currently at around 25,000 custom made sprites or so, and a bunch of them look really neat(and a bunch of them are just memes like big chungus) Captain Invictus fucked around with this message at 09:08 on Apr 28, 2023 |
# ? Apr 28, 2023 09:05 |
|
better a pig than a fascist https://www.youtube.com/watch?v=7BMaxdrCg2o
|
# ? Apr 28, 2023 09:08 |
|
Ghostlight posted:better a pig than a fascist I always found funny when people hyped that line when Myazaki had done a manga about cute german pig tank crew based on Otto Carius memoirs
|
# ? Apr 28, 2023 09:31 |
|
Ai quick visualization is weird because it can kind of get it but sometimes it janks it up. When I tried to do some concept art for film project just to see if I could make something showable it made some weird poo poo lol
|
# ? Apr 28, 2023 09:42 |
|
|
# ? Jun 6, 2024 23:49 |
|
That xenomorph has moobs
|
# ? Apr 28, 2023 10:18 |