Sony AI, Sony’s research and development division dedicated to AI, has developed a new technology called Protective AI (PA), as reported by The Nikkei. Its purpose is to prevent copyright infringement in videos and music generated by AI and create a system in which creators can receive fair compensation.
Apart from preventing blatant infringement by prohibiting, say, “Ghibli-style” outputs, Protective AI is apparently trained to avoid imitation even through indirect prompts. This works by deliberately feeding the source material, in this case Studio Ghibli data, to the model, and instructing it to generate outputs that diverge from that style.
Additionally, the technology is meant to allow creators and rightsholders to be compensated when their work contributes to output generated by the AI model. Sony has previously mentioned such efforts when it comes to music specifically, citing the experimental use of “unlearning algorithms to measure how much each training example contributed to the generated output.” It seems Protective AI will also approach video output in a similar way, pursuing attribution and compensation to rightsholders.
Protective AI (PA) is currently in its R&D phase, so Sony still hasn’t adopted the technology internally.
So to have their content not used.. every company just has to let Sony train its AI using their content? That seems like an extremely dubious exchange at best. AI has to generate things that look like something anyways, it will always resemble someone’s style, it can’t do novel work by merit if its very design.
Tbf, there’s literally no other way to tell the AI what not to do. The AI will never know the Mona Lisa, for instance, is someone else’s work that shouldn’t be copied unless you specifically tell it so, much like how a million monkeys on typewriters could stumble onto writing Shakespeare and never know. Both examples are famous by human standards, but AI is about as knowledgeable of that as the million monkeys unless you tell it so, much like how a child would never know unless you show them those works.
While i do mostly agree with this. I will say that in an environment that seems to push more and more sketchy this is at least a step in the right direction.
This system is being developed not to stop AI from copying art styles or specific themes. Its to compensate the share and right holders, as in this example, Studio Ghibli—their animators, when their style gets copied unlawfully. And as for the AI training, this entire thing is still in its Trial and Error phase. I’m sure the companies who sees AI copying them as a threat will be willing to spend a fortune to Sony to do this to protect them.
As though every single AI company out there isn’t *already* training on their content without consent. As though every single AI company out there wouldn’t still be training on their content without consent of Sony had never thought of this.
As if Sony wouldn’t also already be training on their content without consent (and surely currently is) if they didn’t see this as a profitable public image opportunity.
Sony, as one of the world’s prime content distributors and holders of art related IP, has every interest to protect creators. This is exactly what they state it is, hardly about training AI on other people’s stuff.
They do this to protect creators, yet their Playstation store is awash with AI generated fake copies of existing games that pretend to be valid, high quality products and, when purchased by trusting customers and identified as being fakes, Sony refuse refunds and all the slop to remain.
Ohhh, that’s gonna backfire hard.
Either having to give Sony all of your content will be an issue (one far worse than new content just being generated via AI) or it’ll ve completely useless in terms of generation- it’s simply impossible, as all ideas are inspired by other things and converge in some way- that’s how creativity works in the first place. So you either have to hand Sony all of your copyrighted material (bad idea), or it “works” and… can’t generate a stick figure without twisting itself in knots and creating nothing, out of anxiety of stepping over a single copyright. Styles can’t be copyrighted for a reason. If they could, nobody would be able to draw anything.
That is a slippery slope, that only works in obvious situations. You can’t copyright a style, and art is a gradient not a collection of specific looks. Sure South Park, Simpsons, Disney, etc… have distinct looks but what happens when one artist starts to look similar to another, who gets the credit for commercialized, minimalistic art? It devolves to the point where Google gets money when you use green yellow red, and blue. Parody, satire, homage, all can be considered original but derivative works.
Styles can’t be copyrighted, so “Ghibli-style” outputs aren’t infringement, much less “blatant infringement”.
How do you copyright a “style?” Characters yes, logos, even catch phrases and music… But an art style?
“preventing blatant infringement by prohibiting, say, “Ghibli-style” outputs”…
So we’re just lumping “style” in with copyrights now? That’s our future? Bickering over compensation because the picture you made looks like it’s inspired by the picture I made? Can’t wait to sit down as an artist and try to make art that hasn’t been influenced by any other art that exists in the world… 🙄
I hate A.I., but not as much as I hate our B.S. copyright laws, and this is only strengthening their case. It completely throws the idea of fair use out the window. “Style” isn’t even copyrightable in the first place, and fair use gives artists a wide berth for how they use and transform other works. We’re NOT going to defeat A.I. slop with copyright law, we’re only going to further hamstring that last remaining human artists.
So let me get this straight: Sony is spending who knows how much money developing an A.I. model that won’t listen to its users, and will instead spit out A.I. slop art that avoids any known art style — while also identifying all the copyright holders that could hold you legally liable for generating art with Sony’s model, because their art was used to train the model on what to avoid? …Did some Sony executive ask ChatGPT what kind of A.I. model to create? 🤔
Next time just give that R&D money to real human artists to make you some decent content, Sony… 🙄
So the people that want to infringe on copyright and use studio ghibli just won’t use it? I dont understand the purpose of this?
The probably is, EVERYONE copies someone’s style to create art. It’s the whole point of studying other pplz style and being “inspired” or “influenced” by your favorite artist in your field. Illustrators regularly infringe on other artists’ styles and other companies IP with “fan art.” Writers infringe with “fan fics.” Hollywood’s mantra for decades is creating something that “X IP meets Z IP.” Singers regularly infringe on others artist’s songs to get notice by do “covers” of them. The AI can’t do what humans can’t even not too. Compensation would be cool. But will they really consistently pay out anything worthwhile? Doubt it.
This is just impossible work being done. Like dude was saying about the monkeys working and accidentally Writing Shakespeare. Every artist has a style so at some point you would infringe on every artist, to the point no art would be usable. This is AI being used to self counter AI for a profit.
People are misunderstanding the “style copyright.” There’s a difference between being inspired by other artworks and considering them when you do an artwork as an artist vs what AI does when given spefically prompts to copy a style. When a generative AI user puts in prompts, instructing the AI to deliberately make it look like the works of any artist or any studio is a different scenario than those not specifying it looks like an artist’s or company’s style but those works still being used by the AI to create an image.
It’s not the style that’s being copyrighted so much as the prompt instruction by the user to make the generative AI use their name/brand as the source because it’s impossible for the gen AI to regulate how much it uses. Your brand/ identity is copyrightable even if a style technically isn’t! The idea is that if someone is using your name/company specifically in the prompts and getting an output because it was trained on your work enough to be able to spit things out as well as it being without your consent, you should be compensated. Especially considering so much of generative AI slop is being used for profit. Even though it’s not copyrightable, people are making money off which is where it gets even messier beyond the AI companies profiting off training on works without creator consent.
However, it’s true a style can’t be copyrighted. This is all about the technicalities prompting language and names like “Ghibli”, “Disney”, etc having copyright status.
That said, though, I actually am quite confused by this as some other’s have said. The ones who want to directly copy will just use other software.
I guess maybe this is a strange marketing idea hoping to feed off of the feeling of generative AI users who will feel more comfortable using this Gen AI because it is supposedly made not to infringe on other artists? Im not really sure that is going to help either side, though. It’s not like anyone seeing the generated work is going to immediately know what AI tool made it unless it’s on that tool’s website.
The AI will still work and produce images just fine for those thinking otherwise, by the way. This kind of thing can actually only come into play if the prompts specifically use the artist/studio’s name in prompt instructions. Which is where Sony would benefit from being given the works by artists/studios. Sony AI would technically prevent anyone from deliberately using your brand without your consent, but the works are still used in a general pool of training data to make output.
So I’m also skeptical if it will be able pinpoint exactly what artworks were used by the AI generating an image. This may be a really convenient loophole for Sony most of the time. As far as I can tell you don’t get compensated for unlawful use of your work because someone trying to do so simply can’t. They just stop it from happening, supposedly, by making direct prompt directing it to look like x or y works make the AI deviate even further from x or y influence than they might normally use
On paper the idea is a valid idea. Whether it can work in practice without problems anticipated by other comments can’t fully be weighed on until we see some development.