Getty Images Promises Its New AI Contains No Copyrighted Art (technologyreview.com) 41
Getty Images is so confident its new generative AI model is free of copyrighted content that it will cover any potential intellectual-property disputes for its customers. From a report: The generative AI system, announced today, was built by Nvidia and is trained solely on images in Getty's image library. It does not include logos or images that have been scraped off the internet without consent. "Fundamentally, it's trained; it's clean. It's viable for businesses to use. We'll stand behind that claim," says Craig Peters, the CEO of Getty Images. Peters says companies that want to use generative AI want total legal certainty they won't face expensive copyright lawsuits.
[...] The legal challenges have sparked many attempts by others to benefit from generative AI while also protecting intellectual property. Adobe recently launched Firefly, which it claims is similarly trained on copyright-free content. Shutterstock has said it is planning on reimbursing artists whose works have been sold to AI companies to train models. Microsoft recently announced it will also foot any copyright legal bills for clients using its text-based generative models. Peters says that the creators of the images -- and any people that appear in them -- have consented to having their art used in the AI model. Getty is also offering a Spotify-style compensation model to creatives for the use of their work.
[...] The legal challenges have sparked many attempts by others to benefit from generative AI while also protecting intellectual property. Adobe recently launched Firefly, which it claims is similarly trained on copyright-free content. Shutterstock has said it is planning on reimbursing artists whose works have been sold to AI companies to train models. Microsoft recently announced it will also foot any copyright legal bills for clients using its text-based generative models. Peters says that the creators of the images -- and any people that appear in them -- have consented to having their art used in the AI model. Getty is also offering a Spotify-style compensation model to creatives for the use of their work.
There has always been a market for soulless art. (Score:2)
There has always been a market for art without a soul, but it used to be that you had to hire real artists to make it. Now you don't need that anymore.
Newspapers and magazines used to agonize over the fact that they had to pay writers and artists for their work, and thanks to AI, they won't have to hire real writers or artists any more. And thankfully, due to the internet, I can interact with primary sources directly, people who actually witnessed the events in question, rather than having to read brai
Re: (Score:3)
artists used to have to learn a lot of technique to paint realistically in oils, because portraits was one of the few regularly paying jobs for non-famous artists. after photography took off, we started seeing a lot more art that discarded traditional realism and technique. That's not to say there weren't abstract painters before cameras. But rather that abstract painting became the dominate form shortly after the camera. Photography isn't soulless either, there is quite a bit that goes into composition tha
Re: (Score:3, Informative)
The irony is that the model made by Getty (one of the few firms that can
AI training is going to be Fair Use (Score:3)
There is literally no difference what so ever between what programmers have created to "Train" AI systems from copyrighted works and what literally every single human artist has always done. No one creates something that is truly brand new. Every human, from the dawn of consciousness, has only ever modified what they have seen before. Any "new" thing is just good marketing. Every single one of your favorite artists, musician, composer, writer, etc. created their works by copying and re-mixing other works that they have seen before.
That is all these AI systems are doing.
Re: (Score:3)
Problem here is that it doesn't allow establishing control over the product. This is why we're seeing this massive push from everyone to completely revolutionize copyright into it's final form.
That where everyone will be required to pay just to learn from something, and prove that whatever it is that you know hasn't been learned from something that already has been copyrighted. It's the ability to collected a rental fee for the very existence of your functional brain, because human brain is designed by evol
Re: (Score:2)
I draw a distinction between human interpretation and feeding terabytes into a matrix of coefficients. Quite literally a difference.
Re: (Score:3)
If there is a difference, then I would argue that the human artist is more guilty of copyright violation than the AI. The AI's memory doesn't store a copy of the actual picture. It's not there, you won't find it. The human actually remembers and intentionally references actual memories of the the original. Making such intentional citations in the art itself used to be a sign of a well skilled artist.
Re: (Score:2)
That's like saying there's no difference between a nuke and a stick of dynamite since they both go bang.
AI is capable of digesting and training on a level which is humanly impossible, just as nukes have the capacity to create explosions on a scale which is impractical to achieve with conventional explosives.
Re: (Score:2)
it's a bit morbid of an analogy, but that's correct.
If you are holding a stick of dynamite in one hand and a suitcase nuke in the other, and they go off at the same time, there is no difference to you.
Re: (Score:2)
Re: (Score:2)
There is literally no difference what so ever between what programmers have created to "Train" AI systems from copyrighted works and what literally every single human artist has always done. No one creates something that is truly brand new. Every human, from the dawn of consciousness, has only ever modified what they have seen before. Any "new" thing is just good marketing. Every single one of your favorite artists, musician, composer, writer, etc. created their works by copying and re-mixing other works that they have seen before.
That is all these AI systems are doing.
Human artists also occasionally get in trouble for copying other artists, sometimes without even realizing it.
Paul McCartney famously wasn't sure that Yesterday was an original composition, so he spent weeks playing it for people to see if anyone would recognize it [medium.com].
This is an issue for current generative AI systems, they could regurgitate a fairly complete image/song/text from their original training set and the user would never know. It's probably possible for them to add in a "is this too similar to an ex
AI is now good enough (Score:2)
Re: (Score:2)
This is not true, yet. It's still pretty easy to tell when something is created by AI, at least not without a human artist on the back end making final corrections. That might change soon, but it's not quite there yet.
Re: (Score:2)
Re: (Score:2)
That is something that should have fixed soon. Common watermarks like Getty and Shutterstock have certain looks, so they just need to make sure to account for it.
They could either do it by telling it what the actual watermarks in those images are from each source. Or the better way would be to train it to recognize and remove them. If it has millions of known clean images of a broad range of most kinds of images, like photographs, paintings, and CGI of both real things, abstract art, and things like diagram
Re: (Score:2)
We'll always be able to find bad AI generated images, no matter how advance the technology gets. Because the human beings selecting the images are lazy and bad at art.
Not Copyrighted (Score:2)
Saying "No Copyrighted Art" is like saying your food has "No Chemicals" in it.
Unless it's from the 1920s or earlier, or the copyright was badly mismanaged, it's most likely not in the public domain.
Re: (Score:3)
Oh it is most definitely copyrighted, just by Getty (and they'll definitely sue you if you use it without their permission). And IIRC Getty has been sued in the past for selling others' work without their permission, so there's still possibly non-Getty images feeding into this AI.
Re: (Score:2)
They're not saying that. They're saying "no disputes", i.e. they own the relevant rights.
Re: (Score:2)
Re: (Score:2)
For what it's worth, I don't recognize AI generated images as having a copyright, as lest in my country (US). As long as the image doesn't strongly resemble an existing work, then it's in a gray area of uncopyrightable work. This is something the courts haven't ruled on definitively yet, and won't be recognized internationally without some new treaties. I'm sure everyone investing in AI wants the various legislative bodies of the world to recognize the ownership of data their algorithms barf out at a phenom
Re: (Score:2)
The US Courts have been very hostile to anything not created by humans having copyright. (works created by monkeys, cats, and machines are public domain)
Naruto v. Slater is probably the most famous case in this regard.
What if it was human generated? (Score:2)
I swear, I wrote this book without ever reading any other book, so it's surely not going to infringe any copyright.
Same goes for music composed without ever listening to any music or images drawn without ever seeing anything.
I think the copyright office and copyright owners still haven't understood what training an AI model means.
Either no AI work is derivative or all original content is derivative.
Re: (Score:2)
> or all original content is derivative
Paramount has entered the chat.
I dunno, remember all the Microsoft code leaks? (Score:2)
NT and 2000 back in, what, 2003? XP and Server 2003 a few years back, and Cortana last yearsloppy sloppy! Anyway, every time one of these drops there's massive, urgent, HAIR ON FIRE ZOMG warnings for coders to not even let the mere thought of reading them even begun to speculate about the merest possibility of crossing their minds, since if they did take so much as the tiniest glance at them, MS could claim that any code they ever wrote again was tainted by the stolen IP.
It's hardly an obscure legal concept
Hurry up and pollute Getty (Score:2)
Easy, just flood the engine with copyright stuff (with minute modifications if necessary).
Just to fuck with these people.
Getty doesn't have a copyright anyway... (Score:2)
As I understand current US law on the subject, Getty also lacks a copyright on the imagery
"Aug 21 (Reuters) - A work of art created by artificial intelligence without any human input cannot be copyrighted under U.S. law, a U.S. court in Washington, D.C., has ruled."
https://www.reuters.com/legal/... [reuters.com]
So strip out the watermark, and use the images freely!
Re: (Score:2)
I'd wait for that D.D.C. ruling to make through D.C. Cir. and SCOTUS before I'd agree that it's a settled issue.
It's a financial bet (Score:2)
it will cover any potential intellectual-property disputes for its customers.
This is the real crux of what Getty is doing. They are trying to eliminate any worries their customers might have, about legal liability. Will it *actually* have no copyrighted material? Who knows, but it doesn't matter. If you're a customer, Getty is saying they will take the legal hit. That's a pretty significant obstacle removed for many potential customers.
Getty is not needed anymore (Score:2)
Just like photographers, lighting specialists, models, camera-makers....
Opinion (Score:2)
For reference, this is the same Getty Images that charges piles of money for public domain images because of the convenience of pulling them from their catalog instead of hunting for the public domain originals.
On a scale of trustworthiness, I'd place them at or slightly below the RIAA or the MPAA. At least the latter two usually have exclusive rights for the works they want to charge to use.
If you can't beat em join em (Score:2)
This whole thing reminds me those old lifelock commercials where the CEO went around posting his SSN on billboards... suspect outcome here won't be much different.
If you can reproduce a copyrighted work using MS Paint there is no way in hell people are not going to be able to create prompts that do the same regardless of whatever the heck the system was trained on.
Actual copyright risks are in the resulting image not modalities of an images creation.
Misleading title or claim (Score:2)
This is what bugs me about the way people address copyright in these debates; they make it about copyright status, saying things like "train without copyrighted works," or "ban copyrighted works being used," ignoring that because of how copyright wo
Blood sucking lawyers (Score:2)
I'm surprised, impressed actually, that their lawyers could cobble together a generative AI model.