/ic/ - Artwork/Critique

I have to hide and report AI bait threads now but that's all really

>>6950486
commission based artists. And it will only get worse

>>6950754
>will you be watching the digital circus anon...

>>6950760
Your reply is nonsensical

I was affected by AI art alright.

Having to see so many bait thread and doomposting and how anons here go full retardo with the exact same arguments and bullshit... it made me realize how stupid the average user of this board really is and how I should not take anything said on this board seriously under any circumstance.

So, I guess AI art affected me positively...

>>6950787
>Having to see so many bait thread and doomposting and how anons here go full retardo with the exact same arguments and bullshit... it made me realize how stupid the average user of this board really is and how I should not take anything said on this board seriously under any circumstance.
>So, I guess AI art affected me positively...
if come to actually enjoy the AIjeet treads having hour long semantic debates with cretins change of pace from an otherwise extremely slow board.

it's only "taking artists jobs" in that instead of commissioning an illustration for some magazine article now they just ask the intern to do it with the company midjourney account or whatever, nobody is getting an actual job generating AI "art"

>>6950794
I made one ai fag(might have been trip before he tripped) debate bing a bunch, it was pretty funny.

>>6950754
Can you elaborate why it affects mostly commission based artists?

>>6950486
I haven't heard of artist losing out on work outside of china, and that's china so nobody cares.

>>6951096
NTA but I thoroughly disagree with them. Sure you can probably imagine some people who just want their OC illustrated for cheap, so commission artists are likely losing out on those guys, but only the lower end comission artists.
MOST people who would commission an artist would do so because their a fan of their stuff and can see it working for the idea/project. An AI doesn't suddenly make people non-fans of these artists.

So it's still early days yet, but it doesn't looks like AI has done much damage to the art industry/career, BUT I wouldn't be surprised if it has done some heavy damage to side hustles like shirt or poster designs that artists would make and sell - etsy and such are likely being flooded with trash. That said, it wouldn't ONLY be artists effected, all sorts of stores are likely having a hard time of over-saturation from AI hustler dickheads.

>You will buy a 3090 with 24GB VRAM.
>You will train your own Checkpoints, LORA's and LyCoris models.
>You will install the AutoCharacter and AutoDetailer.
>You will AI-Bash artworks.
>You will use ControlPose.
>You will paint over the AI-Bash with a Latent Consistency Model assisting you.
Companies will find a way to generate their own loop hole for copyrighting content generated with their licensed models.


The industries haven't been hiring Juniors for the last 2-3 years so it has impacted us all plenty. Stable Diffusion ruins anything that made Art a desirable career, they can have it.

How has it affected (you)
Nobody will see your hard work due to the AI noise generated.

>>6952484
why gozira has booba ??

>>6952485
>trending on artstation

>>6952484
>Nobody will see your hard work due to the AI noise generated
Fucking this. It truly is over, isn't it?

>>6951315

>that's china so nobody cares.

until everyone start outsourcing to china because... CHEAP.

>>6952484
>The industries haven't been hiring Juniors for the last 2-3 years
Not that I don't believe you but do you have any source for this claim?

>>6951315
>MOST people who would commission an artist would do so because their a fan of their stuff and can see it working for the idea/project. An AI doesn't suddenly make people non-fans of these artists
What you fail to see is that this technology has given fans of the artist an opportunity to BECOME said artist. If I want to commission Dave Rapoza, I don't need to contact or pay him, like a few years ago. All I need to do is to feed ai with his work and that's it. It's faster and basically free.

>>6952847
>until everyone start outsourcing to china because... CHEAP.
That would be the case regardless, so the point is mute.

>>6952927
>Having an imitation instead of the real thing
I doubt you would have commissioned him anyway regardless, he's lost nothing.
>this technology has given fans of the artist an opportunity to BECOME said artist
What, like shad? lol.
If they try hard enough, they may one day reach his level.

>>6952927
>this technology has given fans of the artist an opportunity to BECOME said artist
this is the peak of delusion on their part.

>>6952941
It is. But still, it's going to undercut ze value of original artists. Blus, ziz>>6952842

>>6952907
Fair question, Ashlan Kahn* and other artists on LinkedIn has been vocal about it, especially this year.
Last time I checked open jobs on art station there were 400 Senior jobs, 200 Mid and 15 for entry level.

You can also find a similar problem in Juniors getting hired for coding and such.

*Not certain how to spell his name, just read it in passing.

>>6952986
Are you a fucking Homestuck character?

>>6953017
What do you mean? Yes, I am stuck at home with cold again, but how do you know that?

>>6952842
Afraid so. You can't draw people in with captivating visuals as AI works are quite spectacular at first glance.
In the attention economy you're always losing to volume and noise, even if you do value adds and Unique Selling Points.
If you manage to build a unique style you become a LORA model like King Jung GI.

Build a popular blog or newsletter I guess...

Like >>6951080 said, the only people whose work is being replaced by AI are precisely those who had nothing to offer besides just making pretty pictures.

Stop making AI threads.

>>6953049
>Build a popular blog or newsletter I guess...
This, but how can you stand in said noise of ai garbage? Or maybe recording your whole process for YT might be a good idea too. But who knows, maybe in 2 years from now ai will be able to generate fake videos of "original certified human painting footage verificated video™" too.

>>6953065
*stand out
Ze entire internet was folded with ai shit. Damn it, A half of posters in this very thread are probably ais bots

>>6951080

Whenever any company or anyone for that matter decide to use "ai" instead of hiring an artist in their project. That's taking an artist job. Especially entry jobs that new and upcoming artist needed to break into the industry.

>>6953052
>Stop making AI threads
Sticking your head in the sand won't do a thing. Talking about how the new artist ecosystem will function will function with AI is not the same as slop posting.

>>6953065
They already have programs that generates the different stages as a fake.

>>6950486
likely will start replacing character & porn commission artists now if not pretty soon, at least ai artists will take their place in abundance, it will be so much easier for someone that doesn't know how to draw & paint to pick up ai and generate dnd oc's. it will likely be a base for any commercial art in the future. might replace us, might just be apart of the workflow like reference but I'm pretty sure getting good at ai is going to be important as using reference from now on.

>6953083
>Sticking your head in the sand won't do a thing.
Ah, the human gnat's approach to discussion.
>Talking about how the new artist ecosystem will function will function with AI is not the same as slop posting.
If you wanted to talk about the "new artist ecosystem" you would have presented an argument instead of starting with an idiotic question.

>>6953003
Link? Couldn't based on that name

>>6952484
isnt commercial art already bankrupt? i used to watch FDZ videos and he taught photobashing for the past decade. now artists can do this easier with ai, so how are people losing jobs? would people just be more productive?

>>6953003
>You can also find a similar problem in Juniors getting hired for coding and such.
i am in tech and the reason why there is a credential creep is because there are so many retards with fake degrees or leiing about their degrees. so clueless middle management people hire these guys that have no idea what they are doing (indians) and then they think "man we should have hired someone with 5 years of expereince because grads are crap!". also everyone advertises themselves as an "engineer" now, even if they are the most basic coder that just translate requirments thats in a jira ticket and makes a pull request. is the same thing happening in art?

>>6953087
>likely will start replacing character & porn commission artists now if not pretty soon
After seeing some of the new NAI v3 stuff I fully agree with that statement. Not only is it far better at doing specific artist's styles but it's also better at doing porn in general. Why bother trying to commission your favorite artist for 100$+ when they're always full anyway and put out works a few times a week at most when you can pay 25$ for a NAI sub and infinitely generate in their style

>>6953157
ayo we holodeck now

Why are building throwing fire upwards?. Man this looks terrible

>>6953168
maybe there's an oil rig ontop lmao

>>6953157
Is that an Aetherion lora?

>>6953173
It's not a lora it's just the new novelai

>>6953107
I'm not the OP I'm >>6952484.

>>6953118
Arsalan Khan

>>6953148
That sounds about right, all job listings are asking for BA or MA, shipped titles and AAA experience.
Also ideally a custom made 8-15 piece portfolio made for that specific studio,
And a custom cover letter they may still never respond to.

Universitets are spitting out Beg Tier MA's the AI far out performs on any level.
The universities are ignoring everything but ChatGPT atm.they flip flop between making that not cheating and cheating.

>>6952927
But then you can't say you bought Dave Repoza. Guarantee there's someone better than Dave Repoza on the internet but they're too autistic to get famous without an agent.

>>6953522
Speaking of agent. How do you get one if you are better than Dave but too retarded to shill your self out?

>>6953049
Good thing I draw because I enjoy drawing not because I'm neurotically trying to peddle my shit

I am not trying to be a pro-ai or anti-artist but some of the ai arts are really impressive

>>6953937
That's because they were stolen from some artist retard

>>6953937
really impressive? It's ok, if you ignore the million ai artifacts/ai logic. I've seen a lot better AI imagery, it was "impressive" early on.

>>6953937
It's not about being impressive, it's more that they would take an ungodly amount of hours for a human to render

>>6953937
if you really think that abomination you posted is impressive better go check /sdg/ - Stable Diffusion General
>>>/g/97577789
or
/de3/ - DALL-E 3 General
>>>/g/97577957

>>6954176

>>6951096
They all draw generic sexy girl poses which is the only thing AI is good at.

>>6954250
Okay, can you please post an example of nongeneric (in your head) art?

>>6953157
You can just ask for characters fucking and it gives you mostly fappable outputs. This was just: nsfw, 1boy, 1girl, son goku, nami (one piece), dragon ball z, one piece, spiked hair, crossover, orange hair, sex, pussy, penis, huge breasts, sketch, spot color,

>>6954271
He can't. Verification not required.

>>6953953
no matter how often you repeat it, that's not how this works

>>6954250
AI is plenty good at creating backgrounds, textures, details to fill in.

>>6955048
AI is plenty good at most everything right now, it will only get better.
The visuals generated by it has no monetary value until someone pays for them.

Where is the moderator when you need him??? they are plaguing the AI shit thread,.assholes

>>6955036
>no matter how often you repeat it, that's not how this works
The only arguments you have for it not stealing is a false equivalenve and a piss poor understanding of fair use, retard.

>>6955093
you have zero understanding of what the AI is doing. and the competitive clause is the only real argument you have that isn't just pure ignorance.

.

The more I see AI art and the people using it and saying real artists are not important anymore, the more it reminds me of Napster and illegal online downloading of music that happened, what, 25 - 30 years ago.

I remember those people that downloaded music online and said the musicians were going to be destroyed because no one would pay for their music in the future. Those people remind me of the people today that generate AI art and say they don't need artists and the artists will die off because AI art will replace them.

Musicians survived Napster, will still have human beings making music. Musicians did not go away. They are still here. And artists will not go away because of AI. No matter how much these AI using people say that visual artists will go away because of AI, they will not. We will still have human artists, no matter how good AI gets.

>>6955036
actually more and more ai researchers are saying these types of ai are essentially wierd lossy compression algorithms.

>>6955176
they mean that in the context of information science.
not that the actual images are stored anywhere.
it just means that the information contained in the images gets "saved". and by information i mean the individual pieces of information, not the entire image.

take the example of a black hole. if you fall into it, it will have your "information". but you're still fucking GONE.

>>6955148
lmao please tell me how a program that needs to harvest data to work isn't actually stealing that data

>>6955188
you need information to function, to draw, to do anything. it's called learning. or do you consider all the studies you did, every piece of media you saw and consumed, every view you've seen to something you've stolen from the world and its people?
because you clearly USED that information to learn more about the world, to enrich your visual library. to learn how things look like.

>inb4 false equivalence111!!!
>bad faith!!!!
don't even bother to respond if that's your take.

and no, it does not matter that it is not sentient. it is still USING that information in the same way: to learn from it.

anti piracy retards will claim that copying is stealing, but this isn't even that. how can you justify your position at all?

>>6955186
I didn't say the images were stored you blithering retard.

>>6955198
oh really. the quote is
>That's because they were stolen from some artist retard
>from some artist
>artist (singular)
what is this implying then? enlighten me.

i love how easy it is to tell when the tripfag is here

>>6955197
Holy shit, you nigger, you actual braindead monkey.
I honest to god thought you were finally going to make a substantial argument but no, it's the same "brain = code" retardation I've heard a billion times before, my fucking sides.

>>6955209
i don't try to hide it. but you're more often wrong than not.

>>6955210
concession accepted

>>6955173
yes. but the irony here is that it's the "artists" here saying that artists won't survive this.
doomers are always retarded people who can't think their way out of a paper bag.

>>6955205
not me. I only made one post itt.
https://arxiv.org/abs/2311.13110

>>6954271

>>6954273

I've seen hostility towards proompts on other boards because they refuse to stay in their containment thread.

>>6953157
Because that means I would have to give money to a s.o.y. guzzling techfag. Now kindly kill yourself.

>>6955231
have you actually read the paper? it does not claim that the outputs are stealing from the training data.
you can ctrl-F and you can see the term "representation learning" appear multiple times in the paper".

even when you take this quote
>In a task-agnostic setting the learned representation z needs to encode almost all essential information about the distribution of the data x. That is the learned representation z not only is a more compact and structured representation for the intrinsic structures of x but can also recover x to a certain degree of faithfulness.
note specifically this part
>encode almost all essential information about the distribution of the data x
>the DISTRIBUTION of the data x
this is in line with all the things i usually say about generalization, learning patterns from across large amounts of data rather than individual images, etc.

all this paper is saying about compression is that it compresses the ESSENTIAL information of the dataset as a whole (as opposed to single images). and this is lossy compression.
and it says "transform and compress" too, to make it even more clear.

i don't see how anything in there supports the stealing argument.

>>6955265
>stealing argument.
not their data to compress. stealing.

>6955197
>debate me
>no not like that!
I don't see why you're acting like a broken record. you've been told ad nauseam how your premise is flawed and you haven't brought anything new to the table. you're just trying to win through attrition and sophistry.

>>6955313
did that argument happen in your mind? because it hasn't happened in reality. this is the common response i get: >>6955210
as you can see, there is no argument, nothing of worth is being said. and it's still stuck at the basic fact that artificial neural networks work by simulating a learning process of some sort.

i keep telling you how these things work, all i'm met with is basic denial of reality.
>brain isn't code bro! it can't learn becuz only brains can learn!

and then i go on to ask you: if it's not then HOW IS THE AI DOING IT.
and then the common response is
>it doesn't matter lol it's just taking the images lol

>>6955293
i can use take your artwork.
>download it.
>save it in my folder
>and reference the colors, pose or even aspects of your style for my own art.
>i can even directly do a study on it.
and you would never call any of these processes stealing as long as i don't copy a very large amount of information from your image.

now the AI question aside, consider WHY this isn't stealing, even though i'm directly taking elements from your artwork: it's because the things i'm talking are not things you actually own. the way you draw a hand, a color pallete, a pose in general, a composition. you own none of these things. you CANNOT own any of these things. in fact, you probably took most of these from your own visual library or references.

you can only own the entire artwork you made. you cannot stop anything from learning and analyzing your artwork.

trip schizo... I thought you were going to go for a new angle, but here you are repeating the same shit again and again. shut the fuck up and draw.

>>6955341
i drew a bunch today thankyouverymuch.
and still no argument in sight.
a coward to the end. are you actually afraid to put your argument out there?

either way, i'm outta here for today.

you just ignored the argument. Computers!=humans. goodbye.

>>6955348
that's not a real argument. only a vague and basic, simple statement.
yes, computers are not humans. and? where is the argument?

you are in fact too cowardly to put it down.

goodnight.

I boiled down the argument into a simple statement.
I afford the right to learn from (some of)my data to humans, I don't not for machines. Why should I?

The reason why people think AI art is soulless is because you can't add backstory into the AI art
For example, if the pic related is a human drawing, the artist can say "her boyfriend's soul is trapped inside the robot" or even without the artist's words viewers will still imagine some backstory for the drawing, and then people will be like "muh soul, le sad"
But with AI art if you say something like this, people just think you're cringe and making shit up, viewers won't even think about it because they know AI didn't consider it, it's all algorithm

>6955347

I swear it's this again
https://m.youtube.com/watch?v=xH9g-DT7hPU&pp=ygUceW91ciBsaWZlIG1lYW5zIG5vdGhpbmcgc29uZw%3D%3D

>>6955379
That's already not true it does storytelling just fine and the LCM painting assistant allows incompetent users to do that with a few key words copy pasted from a directory.

>>6955173
>Those people remind me of the people today that generate AI art and say they don't need artists and the artists will die off because AI art will replace them.
When was the last time you bought a CD or hired a local band to play at a Birthday party or similar private event?

People will still enjoy using pencils and Acrylics. It's the low to middle class of artistic jobs that's vanishing.

>>6955477
>bought a CD
I only buy vinyl

>>6955379
If I were to picture AI art as a person, I would picture someone who talks way too much but says very little. Basically a redditor.

>>6953937
What you posted (1.5slop) is a generation behind the latest (DALLE-3/NAIv3)
This is what the current tech looks like

>waa waa debate me

>>6955248
I think you forgot to type out your post, Anon.

>>6955463

>>6955379

>6955353
>(YOU)

le dot.

every pro-AI fag I see is lower than a dirty nigger >>6955210
>>6955331
>>6955221
so I just ignore it and hide the threads like this one

>>6953157
This shit can reproduce any style and any character in that style. It's over.
I used loliconder, a guy whose clean animesque artstyle was actually unique and hard to replicate.

>>6955783
Made me give up

>>6955363
>I afford the right to learn from (some of)my data to humans
lol. it was never your right.
when i study my favorite artworks and artists. they never once had the RIGHT to stop me. because i wasn't stealing anything, i was learning.

you're saying you're allowing humans to learn from your art. that implies you could forbid humans from learning and studying from your art. but can you really?
>I afford the right
you don't have that right. you have the right to your entire artworks. not the pieces that make it whole.

this was never a question before brainlets like you popped up. because learning is the fundamental spreading of ideas.

you don't seem to understand the hypocrisy in calling it stealing just because the AI is too powerful at doing its thing.

>keeps his gaslighting and semantics games
kek what an attention whore. keep doing mental gymnastics to convince yourself you're not a fraud

Seeing this tard talking to himself constantly bumping threads no one cares about is really fucking sad.

>>/6955789
The reason why people tell you this argument is in bad faith is because ML training and generation is in no way, shape or form on the same complexity as how humans perceive, learn, and utilize information. "It just works like a human brain" is the kind of simplified abstraction given to laymen like you, or clueless investors who have fallen for worse than this. It's the best we can achieve with current tech- (and I'm not even talking about image gen tech, which is arguably not AI and just raw probabilistic). But it's not complex at all.

It's in bad faith to compare, because we still have yet to understand the underlying process of how our brain works, and I'm pretty sure a faggot like you who genuinely has no clue on how the img gen tech he keeps gloating about functions, also has no place pretending that the two are alike. Leave the debating you're so fond of for people who actually know what they're doing. You're very, very easily impressed by what you see- (aka le cat reflection on water, completely ignoring said cats had two heads and mangled bodies, AND incorrect "reflections") and you expect us to be as retarded as you are. I haven't seen anyone fit the crab in the bucket saying so neatly before.

The reality of it is, you're desperate for others to accept you and what you're doing because you've been buck broken by AI. Yet, you fail to realize that a lot of people just simply do not care about your opinion and are simply annoyed that you shit threads constantly like a lolcow. You're wasting your time, and if you truly believed in the tech's future, you wouldn't be here desperately trying to get some other fag to debate you. You're ultimately only making things worse, as I can tell you you normies and artists wouldn't despise AI as much if you morons didn't have this cult mentality going on. Just look at the WaifuLabs threads we had before and cry.

tl;dr: You're wasting your time. And I'm probably wasting mine as well.

>>6954273
>>6955783
>>6955785
>>6953157
>>6953162
>This shit can reproduce any style and any character in that style.
Really? And yet all I see is fucking anime - the style AI is already really fucking good at. Show me it drawing in Spumco, or Ed Edd n Eddy style, if it is so goddamned great.

>Made me give up
You have to draw in the first place, in order to 'give it up'.

>>6955873
>>This shit can reproduce any style and any character in that style.
I never claimed that, just that it can make fappable enough stuff, so I'd expect demand for low hanging fruit commissions in that field to decline.

>>6955836
It's rough, I have very good rebuttals, he's wrong on the internet, but I have learned that it's not even worth my time since he will not argue in good faith.

>>6955908
Just don't waste time replying to bait.
This entire board needs to stop doing that.

>>6955870
you entire argument is a false equivalence. i never said it is exactly like a human brain, like i already said in >>6955331
>brain isn't code bro! it can't learn becuz only brains can learn!
but your assumption is that it NEEDS to be a real brain in order to do learning, because you seem to think only real brains can do learning.
but even without transformers, any deep reinforcement learning AI can obviously learn from literal TRIAL AND ERROR without being an actual human brain. (https://www.youtube.com/watch?v=Dw3BZ6O_8LY).). this proves that AI does not to work like a human brain in detail in order to learn.
a transformer is just doing it in a different way using training data.

so once again, i already adressed your points but you are just unable to see it. because you have zero understanding of any of this.

> "It just works like a human brain" is the kind of simplified abstraction given to laymen like you, or clueless investors who have fallen for worse than this.
that's so funny because i made the effort of understanding this tech in depth. partly out of my own curiosity, and partly in order to explain to people like you.
these are simplified abstractions that people like me MADE in order to explain it to you. and you think this is some idea i lifted off some "investors trying to scam me"? lol.

it's time to realize that you are in the realm of dunning kruger when it comes to this entire topic. you know so little, to the point where someone tries to offer you an explanation, you think you're being bamboozled.
and this is exactly the reason why i'm posting. i just find the discourse on AI in artist circles absymal.

AIfaggots have nigger iqs. Just report sage and hide

>>6955897
>>6955783

>pay attention to meeee
you only deserve contempt

>>6955908
you have good rebuttals that work only in your mind, because you don't put them forward.
that's the difference between you an me.
the majority of you are hiding behind
>bad faith!
>semantics!
and ad hominems now. you're a bunch of cowards who have no belief in your own arguments.

i challenge any of you retards to point out where i used semantics or bad faith here. it's genuinely pathetic at this point.

>>6955785
>>6955783
I'll have to raise my artist powerlevel to the point where AI will have a hard time copying me.

>>6955926
Not me.

>>6955922
you keep trying to compare apples and oranges and you're a sophist who then acts smug. even if you were still worth debating after the amount of thread shitting you display, do you think your attitude is in any way constructive to getting your point across? it's a rethorical question but you're too autistic to notice

>ignores every rebuttal
>insults others
>feels icky when his bad faith and semantic trickery are pointed out
>"I won"
>proceeds to project the "it was real on my mind" unto others
what a clown

>>6955935
Stop bumping the thread anon.

>>6955936
I did not, other threads stayed above. we both know who really bumps it

>6955928
T.tripnigger

>>6955569

>>6954738

>>6955931
and how is it apples and oranges?
the AI is using images of apples to understand apples, the concept, and everything that makes up an apple visually.

i already disproved your point about it needing to work like a human a brain. that is just completely irrelevant since it doesn't need to work exactly like a human brain to do learning.
the important part is what the AI is doing, and IF IT IS learning or not.

to come back to my point in >>6955789:
if you have no right to stop a human from learning from freely viewable images, then you also don't have the right when the AI is using it in the same way.

>>6955935
>ignores every rebuttal
like you're doing right now? like you keep doing?
and what did i ignore? i try to adress every point worth adressing.

>insults others
uh huh. look. this are the first responses i got itt. >>6955093 >>6955210
and then you bitch and whine about me insulting you back. do you see me doing the same bitching? grow some skin.

also you're the only people who care about bumping and front page shit. that too is pathetic. (oh nooo! more insults!)

>>6955922
>brain isn't code bro! it can't learn becuz only brains can learn!
Where have I said that? There's no point arguing if all you can come up with is some retarded strawmaning."Only brains can learn" is patently false and I even mentioned machine learning in my point- (Y-you do know what ML means right...?) You seem to have woefully misunderstood what I said. "Learning" isn't equivalent nor comparable for both- image gen models are probabilistic and human brains- or any kind of brain, really- works with different methods.

>le trackmania video
>"I obviously did a lot of research"
Watching meme videos on youtube isn't research. "Deep reinforced learning" also has no correlation whatsoever with image gen tech which is the crux of what we're talking about. It is used in other fields, though. Why are you derailing with this with random crap? Are you trying to show off your "knowledge"? And please tell me this clickbait content isn't what you used to inform yourself, there's so much more to the world of ML and computer science and this isn't a good nor representation of it if you're truly interested in learning how things work.

>and you think this is some idea i lifted off some "investors trying to scam me"? lol.
You lack reading comprehension. Quote, "(...)is the kind of simplified abstraction given to laymen like you, or clueless investors who have fallen for worse than this". Are you ESL? The problem with this field (that I used to love) is that conmen and scammers have successfully tricked non-techfag normies like you into believing their basic bitch shit is AGI level, and you think you're some kind of messiah for the little people like us when in reality, 99% of this board has better understanding of the tech than you do. I mean ffs, you've even went against quotes from Emad himself.

>in order to explain it to people like you
I study the field

>DUNNING KRUGER
kek. Sage'd

>>6950486
I feel like porn artists would be affected the most, given that coomers can just generate what they want quality be damned.
Then again, most people are lazy.

Can't really think of a positive. Maybe it may speed up some processes of genuinely talented artists that know how to use it to increase their workforce rather than as a crutch.

>>6955390
See here

>>6955928

Tripfag got so booty blasted that I called him a braindead niglet, that he (You)'d me 3 times over kek.

>>6955993
>Where have I said that?
the entirety of >>6955870 is arguing about how it's not the same because ai doesn't learn like humans do. same with the apples and oranges shit.
same with every time you say "it's not the same!" isn't this what you're ultimatively saying?
i mean you're outright saying it again:
> "Learning" isn't equivalent nor comparable for both- image gen models are probabilistic and human brains- or any kind of brain, really- works with different methods.

you tell me. how does it matter that it's not the same? how else am i supposed to interpret this point, other than that you think it has to learn exactly like a human in order for it to count?

because i'm merely saying that because it is learning, it is taking the same kinds of things from the data. things that you do not own, things that can't be stolen unless you consider gaining specific pieces of GENERALIZED knowledge theft.

>"Deep reinforced learning" also has no correlation whatsoever with image gen tech
no, i'm merely using it to make the point that it doesn't need to work exactly like a human brain for it to be OBVIOUSLY learning. it's just a very clear example to prove the point.

> The problem with this field (that I used to love) is that conmen and scammers have successfully tricked non-techfag normies like you into believing their basic bitch shit is AGI level, and you think you're some kind of messiah for the little people like us when in reality, 99% of this board has better understanding of the tech than you do. I mean ffs, you've even went against quotes from Emad himself.
i do not believe at all that this is AGI level. it''s all in your head

>I study the field
then everything i said should make sense to you.
also don't lie. please.
someone who studies this would not use the term "deep reinforced learning" instead of "deep reinformement learning". or think that the basic simplified explanations i provide are akin to scammy sales pitches. that's just utterly laughable.

>>6953157
that's base NAI? holy shit that's wild. is this paid or does stable diffusion have an equivalent yet? not spending a cent on robot art since all i use it for is pose and color reference

>>6956041
>someone who studies this would not use the term "deep reinforced learning"
nta but why not?
https://arxiv.org/abs/2302.10848

>>6956041

>>6956054
because it's not proper english
>Deep reinforced learning heuristic
>heuristic
as a standalone you would only say "deep reinforcement learning". if he was in the field then he would have heard that term enough to not make such a silly mistake.

>>6956047
It's paid

>>6956063
maybe he's phone posting. (like you, I assume, unless you are ESL like me)

>>6956067
that's too bad. i'll stick with stable diffusion then.

>>6956063

>>6953157

>>6956041
NTA but ain't reading all that you have prob a 10k word count you posted here and prob 70-80% was repeating yourself

>>6956090
He pretty much regurgitates the same shit every thread anyway

>>6956104
ironically, it's the one thing he doesn't use AI for

>>>658986350

>>6956090
he uses a lot of words to say because AI is "learning" it''s not stealing, your work is just data that exists anyway. doesn't matter how it works, it earned the privilege to copy your work because it learns. and if it learns too much that's just a bug, ignore that, ignore! learning = privilege.
it's hard work being an AI you know, it's not as simple as feeding it billions of images without permission or even the littlest respect of the creators of that data. oh it is? shit... but it's not stealing! it's just taking without permission, that's totally different. you didn't say AI *couldn't* take it, so you can't complain when it shows up suddenly and does so! if you used watermarks, that's too bad because we remove those as part of the training process (using AI!).

>6955980
>autist acts like a victim after shitting the board for months
>still begs for debate when he has clearly exhausted any good will at least 15 threads back
>attempts to construct a reality where he did nothing wrong and he's harassed by one schizo

>>6956136
Stealing is taking something so the original owner no longer has it. Copyright infringement is not stealing and it's not clear if copyright infringement is taking place.

>>6956151
a nigger is someone bumping this abortion of a thread...like you

>>6956151
my definition of stealing

Stealing:
the use and development of computer systems that are able to learn and adapt without following explicit instructions, by using algorithms and statistical models to analyse and draw inferences from patterns in data.

>>6956156
The only thing that has been aborted is your chance at ever earning money with Art.

In art we call this plagiarism and forgery. Also funny how there's been internet data protection laws but AI simply gets to ignore all those

>>6956164
Cool, I'll help myself to your wallet, since that's not stealing.

>>6956171
That's stealing too. I have multiple definitions to fit whatever situation suits me. They all stem from respect of property.

>>6956136
>it earned the privilege to copy your work because it learns.
no. it didn't have to earn that priviledge. it was never barred from doing it in the first place.
again. you cannot stop ME from learning from your work. you have zero right to do that. it is not a question of needing your permission or not, considering how it actually USES your data.

>and if it learns too much that's just a bug, ignore that
it's not a "bug", but it is not the intended purpose of the model or the goal of the researchers. it is a mistake/error no matter how you spin it.

>it's hard work being an AI you know, it's not as simple as feeding it billions of images without permission or even the littlest respect of the creators of that data.
and do you know what it does with that data?
ironically, if you were to anthropomorphize AI, it would actually be fair to call it "hard work". because what it is doing is not that dissimmilar from doing studies of all the images. over and over again.

>without permission or even the littlest respect of the creators of that data.
you do not need permission to learn knowledge, to analyze a thing, any thing. you may need permission to access some thing or to reproduce a thing, but if it's freely put on the internet, then you already gave that access away.
as for reproduction, the AI is not doing that.
it's not like the AI is recreating your work or using it directly in any way. it is only using it for learning.

> if you used watermarks, that's too bad because we remove those as part of the training process (using AI!).
because they don't want watermarks in their outputs. not because it steals of copies from the data, but because the AI will get the idea that watermarks are a crucial aspect to certain types of images.

>>6956090
>>6956104
because i'm trying to get certain ideas through your thick skulls.
i post the same things because nowadays i get the same kind of non-replies every time.

i actually think in this in thread some slight progress was being made. since someone at least managed to put down some of their points proper instead of winning an imaginary argument in their own head and just calling me names.

>>6956164
>the use and development of computer systems that are able to learn and adapt without following explicit instructions, by using algorithms and statistical models to analyse and draw inferences from patterns in data.
ok. let's play along, and how is that stealing?
which part describes the stealing, what is being stolen?

>>6956173
Your wallet being in your pocket and not mine is stealing. It suits me this way and stems from my respect for my own property.

>>6956192
don't steal my post and make it green you faggot.

>>6956182
>not because it steals of copies from the data, but because the AI will get the idea that watermarks are a crucial aspect to certain types of images.
gee it's almost like AI steals from an image 100%, something humans don't and can't do because the way we learn are fallible unlike machine learning

>>6956199
this is so deliciously ironic.
quoting is specifically described as fair use.
it just shows how far you have to go to describe what the AI is doing as stealing. you have to basically forbid any interaction with the data at all. and claim ownership of the entirety of it.

and i really mean ALL of it.
like this sentence?
>"and claim ownership of the entirety of it."
this is MINE now. you cannot use these word sequences of
>"claim ownership"
>"the entirety of it"
ever again, much less the entire sentence.
because they're mine now. and i didn't give you permission to steal my shit. i own this after all.
that's the true nature of what you're saying.
>>6956205
>gee it's almost like AI steals from an image 100%
is that how you think it works? tell me more.

>6955783
>This shit can reproduce any style and any character in that style. It's over.
If literally all you had to offer is how your work looks, then it was over way before AI.

>>6956211
>that's the true nature of what you're saying.
you wouldn't know truth if it hit you in the face.

>>6956041

>someone who studies this would not use the term "deep reinforced learning"
Linguistic error, I'm not english and I didn't study in english either. It's not my only obvious typo.
Sorry for mixing up. Proud self-taught ESL moment but,
>deep reinforMement learning
Looks like I'm not the only here who's one.

>i'm merely using it to make the point that it doesn't need to work exactly like a human brain for it to be OBVIOUSLY learning. it's just a very clear example to prove the point.

Yes? Which I agree with in the last two posts. ML is ML. I've never once refuted this, because it's retarded. I dare you to find anywhere in my posts where it says as such. There's learning in the name. The entire defining purpose of it is that it learns. What's your point? That machine learning is.... Learning? Mind blown.

>how does it matter that it's not the same?

Oh, so you do agree it isn't? The reason why I mentioned AGI is that the things you claim image models are "totally doing" like understanding the concepts and things they generate like humans do would actually require AGI. What it does is impressive (sometimes), but it's much, much stupider than what you're led to believe.
If it understood what a cat was, it would know there is no chance for it to be two-headed or have two tails, as it's not a trait that IRL cats have. Inb4 "but human artists make mistakes too" like you've claimed before, try and tell me that here.

(1/2)

(2/2)

How it works is pure math. The AI understands that graphic data is supposed to be arranged in a certain way, pixel by pixel, via tensorization- when fed image data, in the case of image gen. When prompted, it will try to re-create the ((PIXEL PATTERNS)) weighed by the keywords you prompted (which are also used in the learning process). It knows that certain patterns (aka pixel placement) are expected, but unlike a human, is unaware of the validity of said pattern in the output, because it doesn't know what a cat is, biologically.

In your research, you might've misunderstood the term "3D tensors" which is what is used to get all data from a colored image- and interpreted it as the AI learning "things" in a 3D aspect, with full spatial visualization and projecting and translation into a 2D plane like how human artists work. 3D tensors are actually just a plain image processed in raw data for learning purposes; it contains pixel intensity information for all channels and is necessary for colored images. The AI literally scrapes the image of all mathematical information it has- but it doesn't know what it is. We literally aren't capable of copying raw data to such a degree by just looking at an image- which is the crux of the problem.

(I also think I forgot to sage the last post, my bad.)

>>6956226
>>6956227
>forgot to sage the last post
forgiven, you've got some good posts here. I was thinking AI could pick up on small details like the heartbeat affecting a line you drew, even if it doesn't know the wobble is from your
heartbeat, it still stores some data related to that. so it has a vector 2345567 for heartbeat, just without a text tag for it. and would reproduce it when asked to draw in your manner.(maybe not visibly in current AI, but later on for sure)

>>6956226
>What's your point? That machine learning is.... Learning? Mind blown.
no, that learning is not stealing. yes. mind blown, apparently.

>The reason why I mentioned AGI is that the things you claim image models are "totally doing" like understanding the concepts and things they generate like humans do would actually require AGI.
no. i call it "understanding", but what it actually has are just representations of tokens. it's not wrong to call those "understanding". in an LLM you'll find these tokens include relationships with other tokens and in image gen a token will include the generalized representation of what it has learned about that token in the training data. in short: a sense of understanding of sorts.
if the token is "dog" the the token will have a representation of the features and relationships that make up "a dog" in general.
and this is something it has learned from seeing patterns across a large sets of dog images. not from individual dog images.

researchers will use words like understanding all the time. because that's essentially what it is, yes, even without a shred of sentience. because that's just what these tokens are. a representation of a concept or idea.

cont.

>If it understood what a cat was, it would know there is no chance for it to be two-headed or have two tails, as it's not a trait that IRL cats have.
you say "if it understood a cat" but realize that the image AI has a limited, VISUAL understanding only. you're saying that because it makes mistakes, there can be no understanding. and that is a simplistic, black and white view of what is actually happening.

take the example of a cat's surface features.
>it's general shape, texture, elements like ear, nose eye and the proportions they are arranged in.
and now take in its deeper unseen features
>its skeletal structure, anatomy the resulting range of motion.
the former is visually easy to grasp, the latter is not. it would take much more data to see the patterns that tell it about the DETAILED anatomy of cats.
all this is to say that just because it makes errors, it doesn't imply non-understanding.

the other example i always use is reflections. SD's reflections are rarely accurate, but the fact that it always gives certain surfaces reflections at all is already a sign of generalized understanding.

cont.

he just keeps going
it's insane

>cont.
no. don't.

>>6956263
combi breaker you're a nigger

>>6956170
>melted signature

>>6956273
It's a mistake!

>>6956273
what, isn't your own signature just a weird amalgamation of all the signatures of all the artists you learned from?

your honor I didn't kill them. I only mutilated and crippled them. it's not murder, see. so I should walk free tee-hee

>>6956273
Would you prefer a real one?

why does it matter what goes on inside the black box? why should it matter at all? the action and results matter.
the action, taking work you didn't own and feeding it to a machine, result = images built from concepts gleaned from the work inputted with no effort on your behalf. very similar action to "stealing" work and presenting it as your own. you didn't do the work, just presented others work in some form or another. in the AI's case, the form is not a mish-mash of pixels from those works, but a mish-mash of concepts and more, some kind of stuff we don't even know.

it's infringement of some kind, maybe we need a new word for it. I'm going with "Stealing" hehe gottem.

I forgot the herbs. sorry bros just hide the thread...

>>6956304
because it let's morally dubious individual act in impunity thanks to pilpul

>>6956308
lets*

>>6956227
the key here is this
> it will try to re-create the ((PIXEL PATTERNS)) weighed by the keywords you prompted
the pixel patterns you're talking about here are neither static nor are they even representing specific images in the training data.

instead these keywords (tokens) represent GENERALIZED PATTERNS seen during training. because the one token for cat includes everything the model has learned about cats in its training. meaning all cat images. and it's not the full images that get associated with the token, but merely specific features and relationships. ones that ultimatively make up a cat.

yes, at the source of this all is statistics and math, but it's for the sake of pattern recognition and generalization. this is the crucial part you a re missing.

>3D tensors
i never mentioned those nor did i mention spatial understanding. everything the AI learns is visual patterns. some papers have claimed that it has seen enough patterns to have a general sense of space in an image, but that's not really relevant here.

>The AI literally scrapes the image of all mathematical information it has- but it doesn't know what it is
it tries to nudge itself towards recreating the training image during training (using backpropagation). but that's not the learning part.
the learning part happens when it copies a large amount images and "collapses" their collective representations onto single tokens (well, in actually it is a complex multi dimensional mapping that is also associates with other features and tokens).
the result is a representation that is better at recreating not just one specific image, but ALL the images associated with that token.

what this means is that the token gets better at representing the features and relationships that are actually statistically important to the concept of a cat.

So anon, what will (you) do in this new visual art economy?

>>6956304
but it is not other people's work. it is by all means the AI's work, everything i said about it highlights this fact.
(imo the argument is different for LoRa's btw. because especially artist loras can have much more significant influence from one singular source. it depends on whether you call copying a style plagiarism. and i would. but either way it is not copying images directly. the same ideas of generalization apply.)

yes, it took other people's work in the broadest sense, in order to learn, but the result is not at all yours, even if you were in the training data. that's the point. how can you claim that it stole something that isn't yours or anyone else's?

again, what it "took" are not things that belong to anyone in the first place. transformers need large amounts of data to learn, because they learn by pattern recognition. by seeing hundreds of round apples, it knows that the apple is a round thing, with a stem. and not any specific apple. the things that it learned, the shape, colors and textures of the apple, are not things you own, even if you have a painting of an apple.

kek, lel lmaofuckinglulkek the clown keep blathering

Honestly tired of AI grifters coming out of nowhere pretending to be artists while stealing another artists style then immediately deleting their accounts when they get exposed for using AI. Gee I really wonder why artists dislike AI

>>6956261
>It's not wrong to call these "understandings"
It is an abstraction given to people laymen who have no background or understanding in ML, yes; but it's like saying a star is a burning ball of fire in the sky- it works for a kid to visualize the general aspect of it, but it's completely untrue in an academical setting; and no debate should ever be made on that sort of foundation.

But let's entertain it. Your little grade school infographic here is true in one part- It doesn't contain images, as much as pure raw mathematical processing of the data it was given via tensorization. But again, the explanation here is incredibly simplified to such an extent, I think this had led your understanding astray. Please provide the source for this image as well.

>in an LLM you'll find these tokens include relationships with other tokens and in image gen a token will include the generalized representation of what it has learned about that token in the training data. in short: a sense of understanding of sorts.
>in short sense an understanding of sorts
You're abstracting and generalizing it wrongly again. Please refrain from that when discussing something as technical, unless you're purposefully trying to throw off others with ambiguity. Again, You're basing your entire argument on your own perception of a throw-away word in an infographic aimed at laymen.

>the pixel patterns you're talking about here are neither static nor are they even representing specific images in the training data.
It isn't. For one, the training data is static. For the generation output, it's mangled by noise vectors and such. Besides, it's a null point- data also isn't considered "static" with, say, ZIP compression and dynamic transformation, yet, the content inside is the same in the eyes of the law, before and after transformation.

(1/2)

>>6956338
>Besides, it's a null point- data also isn't considered "static" with, say, ZIP compression and dynamic transformation, yet, the content inside is the same in the eyes of the law, before and after transformation.
The difference is that ZIP compression is lossless, while compressing billions of images into a 2GB model file is incredibly lossy. It is so lossy, that the model is forced to generalize. Perhaps some inputs can be "decompressed" in a recognizable manner, such as the Mona Lisa or common meme images, because it has learned that these things are their own concepts, but anything that is not extremely prevalent in the data cannot be reproduced in any kind of recognizable form. That aside, training data is not entirely static. A commonly used technique is called data augmentation, which modifies the data in some way each time it is used during training.

>You're abstracting and generalizing it wrongly again. Please refrain from that when discussing something as technical
Would you prefer source code excerpts?

(2/2)
>the former is visually easy to grasp, the latter is not. it would take much more data to see the patterns that tell it about the DETAILED anatomy of cats.

I'm gonna be honest, this sentence is written in such a way I have no idea what the fuck you're on about. But assuming your point is that "But look at this, this is so complex, surely it needs a deeper understanding to do it": This is both incredulity and attributing complexity to something that isn't that really that difficult. I know you draw, so you're probably thinking like an artist here- seeing things that you, as a human being, would naturally find difficult to draw (like folds) and attributing the AI pulling it off to the model knowing the same amount of information and skill that would be required for an artist to pull off. (btw, this kind of complexity argument is also used by creationists whish is pretty funny.) You're humanizing it.

The truth is, like I said, we do not work on the same level and as such, this kind of interpretation is false. This is why, like you've mentioned yourself, you'd need an immense amount of high-quality, diverse image data for a image gen model to be able to pull it off- It knows how to do it to a T (most of the time), because it has learned to do so by processing the raw mathematical data of images where such a thing happens- if it was tagged autistically enough it's what's it's good at. It's not that impressive.

Again, to reiterate: It does not learn the thing, but the pixel data of the thing in three different scales.

>Reflections rarely accurate
>Still uses it as proof
?
>>6956330
This is just a giant word salad. Please get some sleep

>again, what it "took" are not things that belong to anyone in the first place
That's your opinion, so you'll have to agree to disagree with literally everyone here. I believe one is entitled to the raw, mathematical, lossless tensorization of their copyrighted work.

>>6956350
>while compressing billions of images into a 2GB model file is incredibly lossy
It's funny you say that when literally very pro-AI crab has been trying to debunk that very statement since the beginning. I don't entirely agree with that quote anyways, like I said, it doesn't compress images, but their raw pixel data in three vectors. Maybe I'll try to make an infographic or whatever to illustrate my point but I'm itching to draw atm and I won't be engaging any further in the conversation.

I like your argument here - I've heard both when it comes to image model training, that it depended on models for whether or not the data was truly static- but even with information loss, the training (and I mean the tensorization) still occurs non-compressed images, do we agree on that point? Sure, for higher quality results, you'd have to graphically modify the images during tensorization, but it's stuff like flipping, cropping, and changing a few things, which I- and the law- wouldn't call fair use. Data augmentation isn't that involved of a process here. Besides, is there a statement pertaining to how truly lossy a model like say, SD 1.5 trained on pruned LAION is?

>but anything that is not extremely prevalent in the data cannot be reproduced in any kind of recognizable form
Sure, we both agree on that- yet, it doesn't change the data itself being unethically sourced. Can you really blame people for being slightly miffed about it, when none of the current AI gens would be possible without their own input, taken with both no warnings nor compensation, and with the sole intent of making them "obsolete"?

This is my last reply, time for me to draw. Also duc, find yourself a debate companion on discord or whatever, 4chan really isn't the place for that

>>6956338
>It is an abstraction given to people laymen who have no background or understanding in ML
it's an abstraction i've heard ML researchers use all the time. it's not that unusual for them to anthropomorphise AI like this in a colloquial setting.
even outside of that, LLMs have actual semantic understanding of their own context.
as in, the relationships between the words, what each word referrs to in a sentence and across senteces, paragraphs. and that is in addition to their general meaning.
i don't know what you call this but i call this understanding.
i am aware that this is not the same as human understanding or sentience of any kind.
either way the point is moot: you know what i'm talking about when i use that word.

>image
i made that image. it is literally my own understanding. tell me how it is wrong instead of blaming the image. like i said, it's just me trying to explain my position.

>It isn't. For one, the training data is static. For the generation output, it's mangled by noise vectors and such. Besides, it's a null point- data also isn't considered "static" with, say, ZIP compression and dynamic transformation, yet, the content inside is the same in the eyes of the law, before and after transformation.
are you seriously trying to say that the model being a "lossy compression" of the training data has any meaning here? to the point where they would be considered the same in a legal sense?
even though training images cannot be recovered, and you have to try pretty hard and even then only recover them less than 1% of the time?

qrd on thread?
wth is going on here

>>6956337
>Gee I really wonder why artists dislike AI
lack of nuance in the theft

>>6956403
>It's funny you say that when literally very pro-AI crab has been trying to debunk that very statement since the beginning.
It is pretty funny that people would argue against this. In a certain way, all kinds of intelligence are compression. You can make a prediction and the better your prediction is, the less data you need to store to note down the correction that makes it match the target data.

> the training (and I mean the tensorization) still occurs non-compressed images, do we agree on that point?
It's a bit more complicated than that. The training data itself may be stored as PNG or JPEG or whatever. I don't think it really matters. "Tensorization" is just uncompressing this data into RGB, which you have previously described, and storing it into a specific form of a multidimensional array, which is commonly used for ML. Typically it has the shape (batch size, height, width, channels [RGB]), with all images being somehow either cropped, scaled or padded to the same dimensions inside the batch. A batch is a number of images shown to the model at once during training, for example 1024. So the input when training something like Stable Diffusion may have been tensors with four dimensions 1024x512x512x3, because it was trained on 512x512 images. This is what a practicioner might call tensorization, but it is really just a very mechanical transformation of how the pixel data is laid out in memory.

>>6956403
>>6956511
The more complicated part is that these models are actually multiple models: A text encoder, a variational autoencoder (VAE) and the UNet. For our purpose, we can just ignore the text encoder. The VAE actually consists of two parts itself, an encoder and a decoder. It takes a batch of images and lossily compresses them. It's kind of like JPEG, except the representation was learned from data, rather than being a manually designed compression algorith. Specifically, it takes 8x8 blocks of RGB pixels, and turns each of these 8x8x3 blocks into a single 1x1x4 block. This means that a 512x512x3 image is compressed down to a 64x64x4 representation.

The VAE decoder will take this representation and lossily turns it back.

You are right that data augmentation is usually quite simple (clipping, scaling, noising and randomly cropping images, etc.), but I figured that I would mention it shortly.

The UNet is the thing that is the actual AI that generates the images and learns concepts. It doesn't work on RGB pixels, but the abstract representation of images that the VAE encoder produces and that the VAE decoder can turn back into images. During each training step, when the model is shown a, let's say, 1024x64x64x4 batch of encoded images, first a certain amount of random noise is added to these images. Then they are fed into the UNet. The UNet is basically told how much noise was added and tries to denoise the image accordingly. The output is then compared to the target data and an optimization algorithm modifies the numbers in the UNet, so that its output will become closer to the target. It is also shown the output of the text encoder during this process to "condition" on, so that it can learn how to denoise an image of a dog.

>>6956403
>>6956511
>>6956512
When generating an image with the trained model, a 64x64x4 "image" in created by filling it with random values. This is then given to the UNet, which is told that this is a very noisy image of a dog or whatever else the text prompt says. The UNet then tries to see a dog in the noise and removes a little bit of the noise in some places. Then it is given the result it just produced and told that this is a slightly less noisy image of a dog and it will try to remove some random noise from the supposed dog again. This is repeated a number of times (25-50 usually) until a more or less clear image emerges. It's a bit like watching clouds and seeing images in them, or like creating a sculpture of a bear by removing every bit of stone that doesn't look like a bear.

This isn't really an argument, I just wanted to talk a bit about what actually happens with the image data. Hopefully I haven't kept you from drawing for too long.

>>6956478
False flags.

>>6956355
>I'm gonna be honest, this sentence is written in such a way I have no idea what the fuck you're on about.
are you not an artist?
i mean the unseen stuff. the skeleton below the obvious visible stuff.
just like how humans can construct skeleton shapes or boxes/beans to help with drawing an accurate human.

same with the mirror. to depict a mirror, perfect spatial understanding is a MUST. because a mirror is a perfect reflection, so it must reflect its surroundings with minimal error. so these are additional requirements a model would need in order to depict something perfectly. it's not hard to understand that this isn't something that is easily learned, or at least is much harder to learn than the obvious stuff i mentioned. like the cat having 2 ears.
and yet it is required for perfect representation.
it could have the easy stuff included in its representations while lacking the nuance of the difficult concepts and yet you'd call it all the same: a failure to understand. because you equate this to human understanding. but that is not what this is.

>Reflections rarely accurate
>Still uses it as proof
i'm trying to tell you that "understanding" is not a simple black and white, on and off switch.
where you see failure, i see even the attempt at it as something remarkable, because you need to understand something about the concept to even attempt any of it, even if that understanding is flawed, full of holes or plain wrong.
if it was not based of its own internal representation, it would not be able to even do this much, no matter how wrong that representation is. if you understand art, then yo uknow that you can't just copy pixels.
i use reflections and mirrors as a constant example because they are strictly context dependant. you CANNOT create them on their own, without a context and their nature depends entirely on the context surrounding them.

to me it is just a good example for showcasing an abstract idea that cannot be simply copied.

cont

(............also before you make any more dumb assumptions about me, you can see roughly my skill level if you check the archives.)

>The truth is, like I said, we do not work on the same level and as such, this kind of interpretation is false.
if you say that, then you must have a better explanation. your explanation of "maths" is wholly lacking. ironically you my explainations simplistic, but you just basically handwave it all away with "maths"

>..because it has learned to do so by processing the raw mathematical data of images where such a thing happens-
first it needs to recognize what "such a thing" even is, and it's not like it can just copy pixels. colors must fit. other parts of the image must be considered. it's all interdependant. similar to how a llm semantically understands its own context.

again i stress: you cannot just copy pixels and some fucking RGB channels. it's ludicrous.
if it can even simplify an apple into a generalized edge of a round circle object, than that already is a generalization. a departure from the training data. it is no longer making the training data, it is making a circle because that is a part of its representatio of what an apple is.

>>6956403
One last thing.

>Besides, is there a statement pertaining to how truly lossy a model like say, SD 1.5 trained on pruned LAION is?
I believe SD 1.5 was trained on LAION 2B. The images are 512x512 pixels. At JPEG 90% quality, a 512x512 is perhaps about 26.76KB. Multiplying this out, we get about 50TB of data. I'm not sure about how much of the data they may have pruned, but if they only kept the "best" 10%, that's still 5TB being compressed down into less than 2GB, which is compression by a factor of 2560x. I believe they used way more than 10% of the dataset though.

>Can you really blame people for being slightly miffed about it, when none of the current AI gens would be possible without their own input, taken with both no warnings nor compensation, and with the sole intent of making them "obsolete"?
That much is understandable, I agree.

This really is a discussion for /g/ don't you think?

>i'm trying to tell you that "understanding" is not a simple black and white, on and off switch.
"Understanding" is abuse of terminology when referring to ML.
>i use reflections and mirrors as a constant example because they are strictly context dependant. you CANNOT create them on their own, without a context and their nature depends entirely on the context surrounding them.
That's a very convoluted way to say they're a linear transformation.

>>6956540
first off, i'm not a academic so yeah.
but how is that linear transformation at all? a reflection is entirely context dependant.
i don't even begin to understand what you mean by this. elaborate?

... or are you legit saying that the AI runs a separate linear transformation to flip the image upside down when doing reflections in the water?
please stop it with the jokes....

>>6955347
>either way, i'm outta here for today.
>comes back to argue without his trip name

>>6956519
I didn't read a word you said but I can tell you're terminally assblasted.

>>6956567
he's an attention whore, plain and simple and still doesn't get that we're way past entertaining his autism. he'll be back for (You)s tomorrow with trip or badly hiding it

reminder to ignore the mentally ill shizo

>but how is that linear transformation at all? a reflection is entirely context dependant.
>i don't even begin to understand what you mean by this. elaborate?
Mirroring is a linear transformation that takes coordinates and returns their equivalent when placed on another side of some chosen axis. For a simple example, you can consider mirroring around the "y axis" (the vertical line at coordinate x=0) as a matrix multiplication Mv = w, where v is the original vector, w the mirrored vector and M the transformation matrix
|-1 0|
|0 1|
Applying this matrix, you can mirror any image, no context necessary.
>... or are you legit saying that the AI runs a separate linear transformation to flip the image upside down when doing reflections in the water?
No, all the ones I've seen are worse than that, they simply reproduce a similar shape on the other side of whatever is supposed to be a mirror line in the image. Sometimes it's actually the mirror image, sometimes it's very distorted because the networks in use are not trained for abstract spatial representations, but for denoising compressed images.

>the only way his constructed reality where he's absolutely right works is if he's harassed by a single individual

He thinks after months of trying to validate himself and arguing on a mongolian basket weaving forum he comes off as the mentally sane. Actually rich.

>>6956584
Oops, forgot to remove the trip

>>6956584
You?

Reminder /g/ is the board to discuss AI
Fucking retard

>>6956585
yes, everything is trained for denoising images. but denoising images is essentially the same as creating an image out of noise. that is the entire trick to diffusion models.

but i genuinely am not sure if i should take you seriously after this. you can't be serious with the linear transformation shit.

yes, a ruler can draw you a straight line more reliably than the AI could most of the time. but the AI isn't using a ruler. even when it makes a straight line it is doing so using patterns and relationships it learned from its training data.
same when making reflections. the model is not using linear transformations when making the reflection. so what is your point even?

and you can't use linear transformations to do anything remotely advanced. not even the reflection in pic related, much less images with mirror reflections that aren't just an upside down flip.
again, are you kidding me? i think i'll stop here.

>>6956610
>>6956609
>>6956601
>>6956593
are the tears coming out? better try harder, otherwise i'll bring in the AI dystopia :)
(this is sarcasm)

>>6953049
I think the only way around this honestly is to go back to traditional art, oil painting, graphite and charcoal drawing etc...

If its digital work now you might as well not bother trying to copyright it at all...The only way you'll be able to truly say a piece is your own soon is if its an actual physical medium on a canvas or a sculpture.

>>6950490

>>6956584

>>6956617
>i'll bring in the AI dystopia
You are the last person capable of doing so, brother. If anything AI has helped you show us your disturbing narcissism.

1

> Meanwhile
https://news.artnet.com/news/refik-anadol-vs-jerry-saltz-2400275/amp-page
AIbro is severely assmad an art critic called his work a half million dollar screensaver lmao

>>6956617

>>6956519

>>6956743
>“Your words has no meaning to me,” Anadol replied, appearing to be very worked up. “You never talked to me, never visited my studio, no idea who i am, why and how I create art. But Let me tell you; I create my work from my heart!”
>“The world you coming from is changed! New world is bright, new world is inclusive, new world has no gates!” he added, concluding: “I’m everyone! You are no one!”
That guy has his shit in the moma, the most gatekeepy institution that exists in the arts, and is whining about gatekeeping. All because the critic didn't thoroughly lick his taint and give him a rimjob. Something is very wrong with these "people". Also what use does an ai bro have for a studio, is it just part of the larp or what?

>>6956516
Whomst
>>6956315

>>6956824
>Anadol
everytime it's gonna be someone whomst not American

>>6956516

>>6956619
The ability to create physical imagery without computer aid will probably still lose value. Best of bad choices that.

Everyone will be super sated on new imagery due to the deluge of AI pictures.
Visual imagery hyperinflation is the problem artists face, not ethics or morality.

>>6956917
Do (YOU) have the time to go to the 2nd page of Google or keep scrolling through pgs anon

6956617
So what I'm gathering is that it's a fancy scmancy way of photobashing parts of images it memorised from millions of artworks during this denoising process. It sees part of the noise form a shape it remembers from an artwork or set of artworks and leans into it. That's erotic.

>>6956948
Pretty much, but that output of new content will drown out your efforts.
Like artisanal soda pop vs coca cola company

6957237
I don't post my work online. good luck getting into a gallery here aifags.

>>6957246
>good luck getting into a gallery here aifags.
Gallery art will be swamped with AI shit in the next 5 years because ""Art is about expression of self"".
They can custom order that shit printed on Canvas stretched on a wooden frame.

>>6957267
Would be hilarious to see, I hope meet an aifag in person trying that hahaha
>So what inspired this peice?
>Uh 1girl,big boobs, fine art, abstract, Picasso, vangogh, Bauhaus, trending at moma

>>6957269
Normies are using it too you know.
They'll create normies watercolor Ai pieces soon enough.
They just need a webpage on their iphone.

>6957267
>Gallery art will be swamped with AI shit in the next 5 years because ""Art is about expression of self"".
Most galleries either lean into high-concept stuff that can't really be made by this or any other technique, or artisanal, hand-made stuff. There's only a few galleries in the real world where AI kitsch would fly, and they're not high up on the list of importance. Everywhere else expects the artist to be able to actually interface with the art world or have taste or both.

>>6957246
I bet your art will still look like something AI can imitate, because humans also "steal" art styles from others.
You can't compose rock music without listening to rock music.

>>6957410
ah yes
rock music
given to us by aliens

>>6957410
Stop treating the machine like its a human and the people using the machine as anything other than thieves leveraging what is basically a means to photobash existing art together with enough extra steps that it covers their tracks. You're either not capable of understanding what art really is or you want to get a free pass to be a thief too. Inspiration is healthy for art. AI is not. People like you are not. Go back to being a sociopathic zero EQ programmer or whatever the fuck you do because it sure as fuck isn't and shouldn't be art.

>>6957422
Not the guy you're responding to, but why are you against using photobashing? I'm not trolling, I just want to understand the mindset that it needs to wholly made from 0 or else it's thievery or unhealthy.

>>6952484
>The industries haven't been hiring Juniors for the last 2-3 years
Actually when they need juniors they post for seniors because otherwise they get flooded by retards with shitty portfolios.
You literally just need to be good enough, which is lacking, but there certainly are still plenty of jobs

>>6957374
>create
>watercolor

>>6957431
> I just want to understand the mindset that it needs to wholly made from 0 or else it's thievery or unhealthy.
not him, but I don't believe people when they say that anymore, because it comes down to praising the individual mark making of people, the identity of minute lines, the emotional labor that does exist in places where non artists can't see.
The only response you get is "I can't see it, so it doesn't exist"
photobashing and ai are inconsistent in the places humans are and consistent in the place humans aren't. They both produce understandable visuals, but human markmaking communicates some primal stuff that just tickles in a very organic way.
Production calls for cut corners, so it makes sense in that context, but when it really comes down to it, the best art always comes from manual labor and taking the very long path.
No great art can come from ignoring this, which is why it's unhealthy, I don't much care for "thievery"
If you want your Ghiblis, and your Akiras and your Berserks, you need to respect the artistry in them,.
In the end this will be self evident, but in the meantime... well we'll see what happens

>>6957453
you can make individual marks with AI.
that does not exclude AI, specifically using it as an assistant.

>but human markmaking communicates some primal stuff that just tickles in a very organic way.
you're just romanticizing here. though i agree AI markmaking is very rough and basic still, obviously less deliberate, and maybe that will never change.

>the best art always comes from manual labor and taking the very long path.
highly debatable. ever since photography, even master painters have used reference very liberally.
people could have taken the very very long path of learning to draw everything from imagination, but almost nobody bothers to do that. especially not for everything. here again you are romanticizing a process. but it's not the process that matters.

>>6957422
>Stop treating the machine like its a human
to me, this one line
just highlights the fundamental mistake you people make when thinking about AI.
you're thinking that we are treating the machine like a human when we say it can learn. you're thinking that being a human is a requirement for all these things that AI is starting to do.

but it's precisely because it's not a human that this is a paradigm shift.

>>6957488
Yes, you're not telling me anything I wasn't expecting you to say.
And I won't debate you, you're welcome to try and fail to create meaningful art, if you make it great, and if you don't I won't be there to tell you anything.

I don't like this thread. It's rude. Grow up u lil fags.

>>6957488
>not the process that matters.
the process affects the artist, the result affect the audience. I think it matters greatly.

>>6957488
>but it's not the process that matters
Not to you and people who give two shits how the pic was made, as long as it's made. That is what differs you from everyone else and why you're never taken seriously.

>>6957488
>even master painters have used reference very liberally.
>people could have taken the very very long path of learning to draw everything from imagination, but almost nobody bothers to do that. especially not for everything. here again you are romanticizing a process. but it's not the process that matters.
That is some seriously out of touch take holy shit. It's like you're missing the entire point why people draw and then some. You genuinely think it's just "inefficient picture production".

>>6957520
>You genuinely think it's just "inefficient picture production"
question: what do you think it is? you think they use references for shits and giggles? because i'm sure pic related can draw things from imagination.

>>6957499
the process affects the artist in how much you want to masturbate to yourself.

>>6957574
You know looking up that scene you can also look up how Rockwell worked with photographers and others to produce it.

>>6957574
how much references do you fucking actually fucking need. Even when looking up that reference for Rockwell you look up the elaborate lengths he went to into staging that.
>he wasn't copying random fucking photos

>>6957441
Normies see watercolor style and its watercolor to them

>>6957488
Cringe redditbrained tripfag

6957574

>>6957596
>Even when looking up that reference for Rockwell you look up the elaborate lengths he went to into staging that.
yeah, don't you think it's ironic that it is more in line with using AI than just taking a ref from the internet?

>>6957574
>the process affects the artist in how much you want to masturbate to yourself.
Tell me more

>>6957574
>what do you think it is?
Why is "wanting to be able to draw well" such an outlandish concept to you? People don't want to be hacks and make something themselves. I can already feel you raising an eyebrow thinking "what is this weirdo talking about, why would you spend time and work if you can do it faster and get the same result".

>>6957605
Your retarded how is staging and directing actual people the same as just throwing filters at anything.
>Rockwell actually painted those pictures you know after setting up the references

>>6957617
>"wanting to be able to draw well"
you just want to draw well to satisfy your ego!

>>6957623
why do anything if you don't want to be good at or enjoy things anon
>also you forgot your trip

>>6957623
But of course! How stupid of me for wanting to better myself. I should have made musi- oh wait shit it's the same thing. Ok I should have learned to rally cars and- wait that's not it either. I should have gone to colleg- wait no. But anon I can't do anything now. Everything is me boosting my ego. :(

>>6957617
>Why is "wanting to be able to draw well" such an outlandish concept to you?
lol. it's not. i bet you i spend way more time on that path than most people here.
you, like most here, just assume that using help means giving up on that. because you can't help but see AI in this black and white dichotomy.

again, painters could have tried to do stuff without reference. but they use it in the end because drawing from imagination is difficult. because they care more about the result than wanking over their skill.

>>6957619
>as just throwing filters at anything.
is that what AI is to you? you're delusional.

>>6957630
you can practice and get better. none of that has anything to do with shitting on AI.
what, you think i'm saying "go fill your sketchbook with AI print outs"?

These threads are stupid. AI will not replace everything but it will replace concept artists, illustrators for things like book covers, merchandising, logos, abstract graphic design and so on > N-no it won't Do you think like a big company like HBO will pay artists to make concepts for their new Game of Thrones spin off when the ceo's son can just prompt it for free in an evening? Get real. It is not a question of "will it replace us?" It is a quetion of how many it will replace Disney and Riot games were already caught using AI pictures to promote their movies/games >b-but it doesnt look good So? The general public doesnt care if Mickey has fucked up fingers on the cover of the new Disney Shorts Blu-Ray disc The same thing will happen to comissions by the way, do you really think the guy comissioning furry porn cares if Tony the tiger is cross eyed? No, he doesnt. If you want to survive, go trad, make a comic, or start learning how to clean AI art of the mistakes

>>6957632
I call it how I see it
>It's designed for everyone and their mother to use
>planned obsolescence nigga

>>6957640
Will you be watching the amazing digital circus or other popular series coming

Be friend with AI, they will help you draw

>>6957574
>question: what do you think it is
you being a fag, to no one's surprise.

>masturbate to yourself.
>wanking over their skill.
almost at the station, keep going...

>>6957647
You are a missing the forest for the trees. I've already said it is a question of who it will replace. It won't replace, animators, especially not indie ones, but it will replace concept artists that do locations for example, especially in big companies, just think for a fucking second, why, why would a company like Warner pay an artist to do a concept of Gotham City for theie new Batman movie when they can just prompt it?

>>6957640
letting CEO son try an become influencer marketing strategy
>will the 9 year old army have a new leader

>>6957662
Why is everyone going indie these days after hating on near everything the industry doing

>>6957662
what director are they hiring now

>>6957410

>>6957640
On commission what do think of this

>>6957237

>>6957674
I've already said, go trad.

>>6957662
>you need 5 or etc... years experience of working
>nothing new being made and audiences believe shits stagnating
>for some reason CGI and effects are better looking and hold up compared to newer stuff

is trip the biggest faggot this board has ever had?

>>6957686
Glad my niece finger paintings have more value than all AI or digital works

>>6957710
>or digital works
I know this is bait, but, you are right, finger painting is based. crayon is based.
I regret spending so much time on digital, if I had spent that time on trad...

>>6957635
the point was not about ai you actual schizo

>>6957635
I could ask why don't you go to /g/ instead of Mongolian basket weaving boards anon if all you care about is AI related things
>also buy an ad you fagget

>>6957574

>>6957632

I have just transitioned everyone
Please be kind

.

>>6950754

>>6957574
You need to go back

Le AI

>>6956743

Le

mc

uh oh, someone's having a melty again.

look at this shit.
and they say i'm the one with the issues.

>>6957833
Well anon were you the one making the last 3 AI threads in less than fucking day since this one nearly done

>>6957833
>they say i'm the one with the issues.
You are. this nigga is just tryna hit the bump limit while mocking you.

>>6957833
Will you be (you)ing everyone again next thread as well

>>6957833
You both got issues bruv. Just different kinds. One has schizophrenia, the other narcissistic personality disorder.

Before the threads over prob could post trip niggas word cont.
>I swear if it reached the 1k mark
>and he does this near every thread

>>6957833
Mc(you)

>>6957765
Le dot

>>6957852
huh.
(minus some green text, still gonna be around 4k words)

>>6957837
>well anon..
>*insert delusional shizo bullshit*
i rarely make threads.
the funny thing is i wanted to make a PROPER thread today. but somehow there are a bunch of AI threads already so i'm not sure now if i should.

>idiots trying to reach the thread limit in order to burrito ze truth
You know we are going to create a new threads with links to ze archives of olz ones? Right?

>>6957872
Dear God and he done this every thread.
>also about 8 pgs
>>6957886
discordfags comfirmed

>>6957888
him and the schizo are genuinely the most autistic things that have happened to this board in a while. what's even more impressive is that, like you said, it has happened almost, if not every ai related thread. for months. that's literal insanity.

>>6957903
>that's literal insanity.
one more thread and anons will see the light... one more... one... show me the blueprints...

>>6957903
what schizo?

>>6958013
the one that thinks every pro-ai post is made by trip and spams threads with dots or nonsense to spite, bumping the shit. idk this is 4chins maybe it's more than one person but in that case they're all equally as autistic as tripretard.

pic related looks soulful to me

>>6958040
>One guy spamming nonsense is enough to want have one nigga make 90pgs of debate responses

>>6958040
the tripfag has constantly been caught posting without his trip and he puts it back on after the fact so i don't blame schizo anon for saying stuff like like that

>>6958042
It's nice but
>Sunset = le soul
Is getting old

>>6957886
>we
I thought it was only one very passionate guy about ai.
The fact that you're a group makes it incredibly gay, holy shit lmao
who cares if I bump this I won't be coming back /ic/ is so garbage nowadays it's insane

>>6958137
Aren't we all a group?

I find it funny that /g/ has quite litterally created more art in what a year? than /ic/ has in a decade

How perma broken has AI made /ic/?

>>6959503

>>6955173
Musicians survived Napster because the law stepped in. I know the US and the EU are "trying" to do something about AI but it's probably going to amount to nothing. Art doesn't have mega-corp backing behind it like music does so it's a miracle if anything actually takes place. Your best bet is to just start using the AI and become a touch-up guy until your job is completely gone. This is coming from an artist, not a techjeet. Life is often bleak, sorry.

Alternatively, you can pray that Nightshade or some AI poisoning tool gets somewhere. But that's also very unlikely as most of the money lies in AI, trying to subdue would be suicide for most companies since the other option is far more fruitful.

>>6950787
Noticing the bad advice everywhere here pissed me off. Now I come back out of spite to fuck up crabs attempts to make begs draw boxes 100 times before they ever touch drawing anything else.