This might be one of the most disaster lawsuits ever. How is THIS what every artist online is hitching their wagon to? They're basically saying ALL stable diffusion art is just an infinite number of ctrl C + ctrl V of copyrighted images and they just compressed all the billions of images really hard, and even one part says "just open stable diffusion's brain in the downloaded version and you'll find what it's infringing it's just compressed really hard it's not a black box lol".
Even then, they later say that even if the artwork coming out is totally and absolutely and entirely different, it was still based on copyrighted artwork so it's still illegal, it's a wild lawsuit, how did they get the entire artist community to agree to "if you've learned anything from anything copyrighted, everything you do is instantly copyright afterwards, fruit of the poisoned tree", i've never heard this in my life.
Initial Complaint: https://stablediffusionlitigation.com/pdf/00201/1-1-stable-diffusion-complaint.pdf
Response: https://storage.courtlistener.com/recap/gov.uscourts.cand.407208/gov.uscourts.cand.407208.58.0.pdf
Re-response: https://stablediffusionlitigation.com/pdf/00201/65.pdf
 Nothing Ever Happens Shirt $21.68 |
 UFOs Are A Psyop Shirt $21.68 |
Nothing Ever Happens Shirt $21.68 |
It's funny because the whole idea of copyrighting styles is essentially Disney's idea. "Learning is stealing" just means that any art created by any artist who ever looked at something owned by Disney should be owned by Disney as it. And artists are falling for it and endorsing it.
>artists are falling for it and endorsing it
I fricking hate artgays more than aigays because of this don't they see how this reasoning will frick them over in the future?
They do, they talk about it sometimes. They just literally don't care and hate AI more than they fear consequences of their own actions basically.
>hate AI more than they fear consequences of their own actions
You know every time I think the average person can't possibly get any more moronic..
Steamboat Willie should've entered the public domain ages ago, but Disney went ahead with lobbying for extended copyright
95 years is way too long
>Other people's work should be free to steal because it's old
The passage of time doesn't automatically make theft moral, anon
you better pay up every time you use fire to compensate the caveman who invented it
I've always thought it was hilarious how the moment AI appeared SUDDENLY it became super important to uphold those copyright laws people have spent literal actual lifetimes complaining about.
Almost as if it wasn't about freedom of artistic expression at all but hey.
I'm amazed to see people fighting so hard for all use, distribution and storage of copyrighted material to be treated as theft when the only actual logical way to enforce anti-AI policies for these companies is to ban all fanworks, fanarts and commission works with their property which would dry up the patreon well exceedingly quickly.
>the moment AI appeared SUDDENLY it became super important to uphold those copyright laws people have spent literal actual lifetimes complaining about.
Artist cucks only care about copyright when it benefits them. The rest of the time they b***h about not being to draw their favorite pokeyman or their shitty oc donut steel from their favorite show.
Although their arguments are stupid it is by no means certain they won't win. The courts don't exactly have a great track record interpreting technology. Though they do get it right occasionally.
The plaintiffs literally say in their own original suit that they can't provide any specific examples of works that were infringed. Generally that does not make for a successful copyright infringement case, no matter how much time you spend trying to convince the court that they totally have the technical capacity to infringe via wild misinterpretations of the technology.
>can't provide any specific examples of works that were infringed
your honor, i believe the case makes itself. i'll leave the rest to you.
Ha, ha, how relatable. The Evil Ai clearly stole her art, tee he. She's literally me, she's litterally me and so relatable. .
Seriously though, how the frick did bot stan that moronic goblin creature and her laughable "art" for so long?
At least the Hark a Vagrant b***h had a modicum of talent under all the reddit shit, and insufferable personality.
>Seriously though, how the frick did BOT stan that moronic goblin creature and her laughable "art" for so long?
We never "stanned" her, we made based edits of her work.
moar
yeah, i like the dog ones
>how the frick did BOT stan that moronic goblin creature
she's pretty cute
talentless hacks are often cute
ever heard of avril lavigne?
Even if this was produced by base SD and not by over-fitting a model derived from it, it's not infringing unless it matches a specific original work and isn't transformative.
that doesn't look anything like her art style
It wouldn't matter if it did, you can't copyright a style. Something has to match an original work in a non-transformative way to be infringing.
BOT what does this july 19th court date mean? Is it the date everything will be decided(all of it)? Cuz if so that would be the most happening day in years, actually see twitter up in flames and finally this moronic copyright bullshit ends.
It's probably just oral argument for the motion to dismiss. If they actually go to trial this shit is going to take a long ass time, so don't hold your breath.
Essentially, the defendants said
>This case should be thrown out because the plaintiffs failed to show any actual basis in law to hold a trial
And that date is for the plaintiffs to argue that the case should be allowed to continue.
Now, the defendants are clearly right here. There's no basis in copyright law to find that Stable Diffusion infringes copyright unless they can show it reproduces original works, which the plaintiffs readily admit they can't do. There's still a good chance the court allows this obviously-stupid case to continue, though, because they filled the suit with technobabble and courts are notoriously bad at understanding technology.
The Supreme Court just ruled on that gay web design case despite the fact the entire case was made up.
> did the company have the licenses for training machine learning algorithm using the copyrighted data as input?
> no but...
> case closed.
You don't need a license.
lmaoing.
StabilityAI directly stated they wont be using label music as training data for DiscoDiffusion because they know they would be raped in courts, they literally say as much.
But the copyright violation is exactly the same as using deviantart or artstation as training data for their product.
I mean, it would be funny if the US and EU just gave up on all the copyright system that supports the billionaire entertainment industry so a few tech startups can make a billions laundering data that doesn't belongs to them.
>because they know they would be raped in courts, they literally say as much.
because the system is fricking broken and you're presumed guilty until you defend yourself in court (which costs hundreds of thousands of $ per case).
Precedent is that large-scale use of images posted publicly to the internet without reproducing them is not infringement. That's why they spend so much time arguing that it reproduces the input images (which it does not).
It's pretty clear that if the model is trained entirely on copyrighted work and that the "human input" is simply putting in a few key words like a search engine and the output is some mathematical approximation of all the images that are relevant to those keywords, that there is no transformative aspect to the work whatsoever. It's simply a mechanical means to blend potentially thousands of copyrighted works together for the sole purpose of rendering the copyrighted works themselves obsolete which, is precisely what copyright laws are intended to prevent. I understand that you don't want to it be copyright infringement just like you don't want stealing movies and music to be copyright infringement, but that doesn't change what it is.
nice gpt ai post you fricking homosexual
How come then that, even with same seed and key words, the images output are different
Because the seed is not actual seed. It is a seed for just the noise, not for the other steps.
You don't understand how transformers and diffusion works. It looks at a corpus of work (copyrighted and non) and deduces *patterns* from them, and that's what it learns.
Do you really think ChatGPT is copying and pasting from some esoteric PhP BB post when you ask it how to transport Bigfoot from Seattle in order to get water out of his ear and drip it into the ocean?
You're the type of moron they probably rely on in that lawsuit.
>Do you really think ChatGPT is copying and pasting from some esoteric PhP BB post when you ask it how to transport Bigfoot from Seattle in order to get water out of his ear and drip it into the ocean?
This is what ai hating programmers who are adamant about never using it to assist in generating code think.
what a moronic post
This is bait
Maybe not; after all, some people were stupid enough to file a lawsuit on the same basis.
Its correct.
Companies violated copyright by training their AIs on copyrighted content without paying a license to do so, even more by releasing a commercial product.
When StabilityAI released a model with an open license they didn't have the right to alter the license because the model itself depends on copyrighted data.
It's cut and dry.
bros, are we all just flesh automatons?
>ALL stable diffusion art is just an infinite number of ctrl C + ctrl V of copyrighted images
In this, they are correct.
So that means your entire brain is a huge copyright violation, given how many memories you have of copyrighted information
neuralink hotfix when?
>Compressed to less than one bit per image
Imagine if this was real. You could recreate any image ever created with "yes" or "no."
not all compression schemes are completely reversible, you can get a 50 to 1 compression ratio with JPEG but you will never get the original image back
Okay, but how close can you get to the original if you compress it to less than 1 bit?
what if that's actually what the universe is. entropy being simply a compression algorithm and the singularity at the beginning and end being the single bit.
it's counterintuitive to the notion of entropy being disorder, but if you think about it energy is conserved so for the same amount you get a system of more states with the same total energy while the expectation of the Shannon information increases over time (and isn't reversible according to the second law of thermo). this would be compatible with the holographic principle too.
so from the inside it looks like you have more bits of information and you've increased the total randomness
but from the outside you've compressed it down
thermal equilibrium != zero information
this post basically btfos their whole case http://www.stablediffusionfrivolous.com/
It really doesn't though, the counter argument that AI is somehow sentient enough to know what a cat is and draw one just because it's seen enough of them, is actually absurd
That's not what the counter-argument says at all. But if you think a model can't learn pixel distributions for labeled data with enough input you're being obtuse. It simply refutes false and technologically inaccurate claims made by those afraid of technology and want to sue it away
Total copyright death.
Total, complete, utter AIgay, artgay, and janny obliteration. Tear it all down.
It’s all over anyhow.
Google have now normalised taking it all for their AI
https://www.searchenginejournal.com/google-updates-privacy-policy-to-collect-public-data-for-ai-training/490715/
Google Says It'll Scrape Everything You Post Online for AI
https://gizmodo.com/google-says-itll-scrape-everything-you-post-online-for-1850601486
Imagine not range banning all Google IP's.
Good morning, sirs.
>Sarah Anderson
Ha, ha, how relatable.
what the frick did i just read? do these people have any remote understanding of how technology works at all? the entire thing is blatant falsehoods
SAI should sue these moronic homosexuals for libel
>do these people have any remote understanding of how technology works at all?
If they did they probably wouldn't be artists and they absolutely would not have filed this lawsuit. But even considering the first part I can only agree, what the actual frick did I just read.
So, how are they wrong on this one?
ai doesnt store all the images, it's trained on 5b images and the download is like 3gb, it learns, not stores and distributes, it's completely and totally wrong
I hope they win, my job depends on it
>stability has embedded and stored compressed copies of the Training Images within Stable Diffusion
Wow, even the premise is wrong.
you saw what happened in fentanyl floyd case, reality doesn't enter the court room only daring deceptive description matters.
didn't the police chief throw that cop under the bus by saying "we don't train them to do that" with the knee or something along those lines. in other words they threw him to the wolves.
Time to abolish copyrights
>abolish copyrights
yes, please
Copyright is dumb. Every other year I find an article about a hentai artist tracing another artist's work. I compare the pictures. All I see is the characters being in the same pose, sometimes characters having a similar style. But so what? It's still a different piece of art. Similar but different. Yet people treat it like literal theft.
This AI garbo is even more moronic. AI is trained on artwork and then makes its own unique "art". Unless you tell the AI to copy a style directly it won't do it. AI art is very similar to those random image generators that were always popular. You fed them an image and would get it back edited.
AI art sucks and anyone who pays for those works is a huge moron but I don't see a copyright infringement or anything. AI is like a baby, it's a blank slate with potential. You feed it data and teach it then it becomes able to function individually (sort of). It's like telling a kid to draw a cow and the kid asks "what is a cow?" and you show it a picture of a cow, the kid then draws something similar to that and calls it a cow. Neither the kid nor the AI CTRL+C CTRL+V's the cow image.
>artist tracing another artist's work
plagiarism is a completely separate issue from copyright infringement. everyone is against the former, most don't give a shit about the latter.
>unless
no, it works on a space, and you constraint/limit/force/guide or give freedom to pick so you can put on your negatives ~~*~~*them*~~)) but still get a israelite
nothing wrong with tracing all the greats do it
>Post art to DeviantArt which gives it a creative commons licence
>NOOOOOO, YOU CAN'T HECKIN USE IT!
If Stable Diffusion is just compression then its inventor should receive an Ada Lovelace medal or whatever they give out to jack off computer scientists, because it outdoes all previous algorithms by a factor of a hundred thousand.
7zipbros... we got too wienery
>Compressed 5 billion images into 3 GB
>Mathematically means that the average image inside of stable diffusion is literally 4.8 Bits, not even a full 1 Byte, not even a thousandth of a Kilobyte
>If we assume the average image on the internet is 1 Megabyte, this is a compression ratio of ONE MILLION and 666 THOUSAND
>7Zip's compression ratio is about 1.25
>You can fit 19000 images in stable diffusion in the same space you can fit pic related, and magically you can get them back perfectly and use them for perfect collage
Did artists all flunk math too? This is insanity, they think you can compress a whole entire image into less than a thousandth of a kilobyte and still be able to use it for info and it's also copyright infringement
It's heckin Kolmogorov compression!
I think training AI is fair use, and I don't think image AI models are just a compressed database of the training images, but I don't think this is a good argument.
If you're compressing more than 1 file, data that is identical (or in the case of images, similar) across multiple files can be discarded safely all but once and still be reconstructed. Just dividing the file size by the number of images doesn't tell you how much of each file's data remains as long as there is overlapping data.
I mean, that's fine I guess, but I am almost certain the compression still doesn't add up, you need some really ridiculous levels of compression to fit all the images and still maintain any sort of usable data and I dont think you can overlap your way into turning petabytes into gigabytes, you're eventually going down to using 1 kilobyte or less per image
that must be the work of the VAEs I guess
Reminder that this is not some major rebuttal argument that people talk about in a forum where they think about it that hard, it is simplified massage meant to be easy to spread and convince people who refuse spending any time learning about AI. I think far better argument would be that the essence of each word or prompt comes from just imperfect copying and weighting the averages of each aspect. For example you ask some models to generate you image in style of starry night. That image has been across the internet repainted in so many ways and so many personal artstyles, but always contained dark sky with weird stars and some building on the left. AI being the moron that it is and not fully understanding anything, especially smaller models that are less then billion parameters will think that in style of starry night means making the image blue and putting weird stars in the background, rather then doing the image in style of Van Gogh, or at least in badly curated models. You ask it to do goblin and it will give you the most generic goblin there because it understands that goblins have green ears, big noses and look a certain way. This is what is meant when copying. It is not actual copying, just that when some artist does the same thing or similar, it will find patters and probably save more then few bits for the prompt. Basically if there is natural overfitting in the dataset it can do something close to copying
It was literally one random dude’s blog article that made the rounds on Hacker News making the argument that it was lossy compression, which even the Orange Redditors agreed was fricking stupid
Are you only know realizing that artists are stupid, anon?
There’s a part of me that wants this to go through because I desperately want to see what the online art community does when Disney/Nintendo/Cartoon Network/etc. start suing fan art sites for copyright infringement and arguing this decision as prior case law.
>Are you only know realizing
>only know realizing
fricking moron ahahah
You got me, I forgot to run my post on BOT.org by an editor, I won’t make that mistake again, take my gold kind stranger!
GRYFFINDOR
lmfao did you expect artists to have a working brain?
>i've never heard this in my life.
because it's not a thing for human artists, duh.
In any case I hope she wins. AI "art" is cringe, as evidenced by the public to whom it appeals.
someone sue israelitersen for plagiarising stickmen drawing
Last time I read it it had b***hes a lot written in it.
>i've never heard this in my life.
This insanity started on deviantart, then went Tumblr back when it was popular and went to Twitter. Basically these morons will even say if you draw you art in the same pose it's counted as copyright theft of a freaking POSE. Even if you draw in the same style they scream copyright theft.
>copyright theft of a freaking POSE
There are yoga poses that have been copyrighted (or was it trademarked).
Some yoga was being patented.
https://abcnews.go.com/International/story?id=3200043&page=1
Yes, this is insane as it sounds.
>yoga
>copyrighted
Seriously doubt it. It's exercise, not a creative work, like a dance coreography.
I personally don't mind tracing, but gib credit.
damn americans only live from suing the hell out of each other
I see the memes i made reposted on this site, one time even with a @twitter watermark, I'm happy people is sharing it but i get that talentless people are so cheap and dont deserve credit. One even remade the meme improving it, so I guess that's real effort inspired by previous work & should be encouraged
early life check?
i for one am shocked that artists don't understand things like laws and how a trial works
Wouldn't it be Fair Use though?
Dumb argument.
DeviantArt requests the right to distribute copies of its users' art as part of the terms of service. They aren't grabbing every image on the internet without attribution.
They should have just sued them for copying the data into the training set and into memory during training.
All copy steps are individually protected by copyright laws, if it's not fair use and there is no implicit license it's infringement.
>Sarah Andersen
what did she mean by that?
>They're basically saying ALL stable diffusion art is just an infinite number of ctrl C + ctrl V of copyrighted images and they just compressed all the billions of images really hard
Considering stable diffusion to be extremely lossy compression is useful as a thought, but a fricking stupid thing to base a legal case around.