/ic/ - Artwork/Critique

>>6554688
>The images on google already existed, the ones made using AI don't
It does, it's not fabrication, it's remixing
It's been documented many, MANY times, that """AI""" cannot create because it literally needs data to produce. Art doesn't. There's literal examples of art that could not have any source attrached to it, because beauty is a genetic tract that exist without memorabilia, aka, some iconic image used to product the final result.

>Unless you're taking some kind of bizarre metaphysical "everything in the latent space already exists" take.
A: It's not bizarre, it's fact
B: That's literally what probability means, yes. It's latent space because the data can only product the mix results of that data, not something entirely OUTSIDE of that space. If you give a CPU data of circles, it will NEVER create a square. If I saw a circle and was tasked to make a triangle, I'd still produce a triangle.

>>6525994
>ignores everything else said
I'm against it being shilled on this board you projecting retard. Everyone already knows it's an auto-encoder. The point was you Pajeets spamming it on our boards when you already have generals for its discussion just makes you a colossal faggot.

>>6473833
Photoshopping one copyrighted image over the other with 50% opacity is not "transformative", either. SD does essentially the same, except with a much larger dataset of images and a different interpolation function. It's not creating anything new, it's creating in-betweens (and in some cases lossy copies) of existing data, also fair use exceptions apply to humans, not machines.

>>6472541
>t. doesn't even know how his favourite toy works

>>6460592
Your reply options are little outdated, Pajeet.

>>6451904
Well well well, so this is what the "latent space" looks like.

>>6454401
>they're basically Dadaists, except their Ready-Mades are pretty looking instead of a urinal so it's okay.
Ahahaha, exactly this. Turns out that, essentially, every image that an imagegen algorithm craps out ALREADY EXISTS within the dataset's latent space, and all that proompters are doing is the equivalent of a Google image search. They have the notion that they're "creating" something but it's merely an illusion granted by their ignorance and limited perspective. Their "art" doesn't belong to them in any sense of the word.
Maybe the mods should move all of the /sdg/ posters to /h/ or /e/, since that's technically what they are in reality. That one retard that posts generated celeb pics could go to /s/, I guess.

>>6450250
No.

>>6450098
>Correct perspective
>Everything within the picture is legible and clear
>There is intentional design, a unique style, and a clear idea being presented
Gee, I'm having a really hard time figuring this one out, guys
>>6450103
>and it won't get better, because it all comes down to how it's made - data in, remixes out. but there isn't enough data in the world to communicate your ultra-specific niche desires
exactly right. You actually don't realize just HOW limited that this thing really is, because it doesn't "draw" in any real sense. Everything within the latent space of a dataset is a fixed set of interpolations, an archive if you will, and when you proompt you're essentially just doing an image search to pull up relevant data points from within this archive. In a sense, it's just Animorphs all the way down.

alright, I'm gettin' me mallet

>>6448425
Due to the very nature of how imagegen works, it's never going to get there.
All it can do is manipulate images in a purely 2D manner. It has no conception of three dimensions, and it cannot create anything truly new, only interpolate between the data in its dataset.
In short, (You) will be saying "two more weeks" until the heat death of the universe.
Cope, seethe, dilate, etc.

God, you fags would be so much more effective in combating the AI shills if you didn't let the AI shills lead you around by the nose in pointless debates. Here you go, since nobody posted it yet.
Image generators can only output the original images fed into them + all possible interpolations between them.
When you prompt the image generator, all you're doing is essentially an image search- it transforms your words into a vector, or a set of coordinates, inside the latent space, and decodes said coordinates into an image.
Each point in the latent space is fixed and immutable from the moment the dataset is created. Datasets are static archives. AI "artists" simply perform a search within that archive. They have created absolutely nothing whatsoever.
There you go. This simple, relevant piece of information is all you need to nullify pretty much everything that the AI shills throw at you. You're welcome.

>>6448160
>designed on similar principle
What principle might that be?
Humankind don't even understand how their own brains work, yet we're designing algorithms based on them? Bullshit. You're just parroting the marketing. The moniker "neural network" is based off of metaphor, not actual working principle. You cargo cultists are all idiots just as ignorant of the actual working of the technology as the people you call "luddites".

>>6444102
For someone with the unlimited power of AI in your hands, you all seem to be doing a lot more talking about doing things than actually getting anything done :3c

>>6443204
>t. tech cultist that believes that an algorithm is intelligent and learns like a human

>>6442745
This entire thread is a circlejerk of false-flagging shills. Disgusting. This cannot stand.
"Style" cannot and should not be copyrighted, and has nothing to do with what we're dealing with now. This is all a false argument deflecting from the real issue(s), which can actually be dealt with in a practical manner and will not need extensive or draconian changes to copyright law, merely cases settled to set precedent.

What we're trying to do is prevent companies from mining our data under the pretense of non-profit, research-only usage and then turning right around and using it for profit and screwing US over with the very data that WE produced. This is the core issue. This doesn't affect just artists, it affects everyone. Photographers, coders, writers, anyone who creates content that can be fed into an algorithm.
Secondly, due to the nature of how image generators work, EVERYTHING that they produce is an interpolation of the massive sets of images fed into them. There is no "new" data created, what you're seeing is essentially a huge library of images being animorphed together. What this is is data laundering on a massive scale. Fair use does not and should not apply to this, and if you'll take a minute to read fair use laws you'll see that without any changes whatsoever the AIfags are very likely to land in serious legal trouble. The shills can gaslight us all they want, we are not in the wrong nor are we causing a "legal apocalypse" for standing up for ourselves.

>>6442224
You mean like pretending their latent copyright scam is a magical genie?

>>6440071
I agree. Once you do some research on how it works you see how limited it really is. All it does is interpolate images at a mass scale, there is no magical learning process that teaches our robot friends to paint like a real boy.
The technology has remained practically unchanged for a good long time, the only major "advancements" that's allowed for this to explode recently was that the training & generation was made much cheaper and companies like Stability AI managed to get ahold of a staggering amount of training images, amounts numbering in the billions at this point. All this for a glorified photobashing program shitting out knockoffs of other people's work.
Anyone that tells you that this technology is going to explode in growth, that it's going to REVOLUTIONIZE ART and REPLACE US ALL are either lying to you or are idiots that have fallen for the marketing campaign. On the contrary, this technology is going to plateau. It's already plateauing, in fact. No amount of TWO MOAR WEEKS!!! is going to change the fact that these machine learning algorithms- because that's what they are- are dumb brute force calculation and nothing else, and we are nowhere near any kind of real artificial intelligence that can work with general fundamental principles and create any meaningful output, like, say, a human can.

It makes it easier for the kinds of parasites who used to trace and photobash other people's work to rip people off, but that's about it. For any practical work, it's not really that much help. Maybe if they actually tried to make a tool that could help with, say, rendering or texturing your paintings to speed up the concept art fag pipeline or something, then we can talk. Replacing artists wholesale? Pff. Lol and lmao. Those fags are getting high off their own supply if they really think that.

>>6439086
That simply wouldn't be true even if I don't draw, since computers can't make anything at all.

>>6438125
Here, let me help you out.

>>6437917
>attacking the wrong aspects of this problem
I agree. Here's an attack against the right aspect of this problem, namely that everything that AI creates is just a mishmash of the data scraped from the internet and fed into it.

>>6437926
I mean, gee willikers, when you see multiple machine learning experts saying the exact same thing, you kinda wonder if you should be listening to some anonymous asshole who's been shitting up our board for months and doesn't want any honest discourse, or whether you should be getting the info straight from the horse's mouth.
I mean, do you have any actual rebuttal besides DUUUH THAT'S NOT HOW THE TECHNOGOLY WORKS SWEATY, UHHHH HOW COULD YOU POSSIBLY STORE A BIG NUMBER IN A LITTLER NUMBER THAT'S IMPOOOOOOSSIBLE HA HA HA PLEASE STOP RESEARCHING JUST LET IT GO ALREADY AAAAAAAH
If you aren't going to post any real information to back yourself up then don't say anything at all.

Oh cool, that means I can post this here.

Holy shit, here it is, this is exactly what I've been saying this entire time! It really does store all of the images and interpolate between them! That's all that they do! FUCKING YES! HOLY SHIT GET IN HERE GUYS, I THINK WE'RE GOING TO BUST THIS THING WIDE FUCKING OPEN

Source is https://towardsdatascience.com/understanding-latent-space-in-machine-learning-de5a7c687d8d