Don’t let Disney have a monopoly on AI-generated art

Disney and the rest of Hollywood have been eerily silent about the launch of Stable Diffusion, despite the fact that this open-source AI software will happily spit out high-quality images of copyrighted iconic characters from comics, animation, and movies in response to text prompts.

But this does not mean that they sit idly by and do nothing. There’s a legal battle forming, though it doesn’t involve any of the biggest players — at least we can’t see it publicly. Big first lawsuit Against Stability AI, the company that makes Stable Diffusion, It’s topped off by friendly “indie artist” faces who are featured on a pretty neat website as fighters for a fair shake for indie creators. Their lawsuit takes a direct legal aim at the heart of how generative AI works.

Now Getty Images has joined the fray, too, with a lawsuit against Stability AI Tell the edge It is mainly about seeking legal clarifications and not so much about damages.

the The mouse that ate the public domain Watch carefully, because what’s at stake is my existence.

To put it in terms that sound like sci-fi but definitely aren’t: a future release of Stable Diffusion will not be a replacement for some of the artists whose work you love; can replace full studios and intellectual property stores such as Disney, Pixar, and Marvel. You’ll get the creative and technical talent of a movie studio on your laptop and can keep you endlessly entertained with your favorite characters and worlds without sending a dollar to the copyright owner of those characters.

Here’s the biggest reason: Under current law, none of this would obviously be illegal.

A glimpse into the future

It’s easy to misunderstand the stakes in these battles if you think of generative AI solely in terms of who can do what with images of Mickey Mouse, Spider-Man, and other copyrighted characters and settings. Even if you follow generative AI closely enough to understand that a major part of the battle is about the so-called program Transfer pattern Capabilities—the ability to imitate the style of a given artist well enough to produce an endless stream of new work in it—you still pull a single thread from a broader tapestry.

The potential danger to Disney and other large intellectual property owners is much greater than the mere influx of new user-generated images, memes, and videos derived from their copyrighted work, and thus arguably devalues ​​the original article.

To find out why, let’s play this up.

For now, if I give you a prompt, some parameters, and a prime number, you and I can independently use Stable Diffusion to generate the exact same image, pixel for pixel. On a practical level, this means that with just a bit of text, I can effectively “send” you a very large image file. You might even say if I posted this text (router settings + seeds + stable propagation), then I posted that image.

Now let’s fast-forward a few years to the time when full-size text-to-video conversion was in full flower. In this world, a single text prompt and initial combination might be fed into a ChatGPT clone, which would then produce text, which would then be fed into a text-to-animation form, which would then produce a video.

If we have access to the models required to make this work – let’s say all relevant models are completely open source, like Stable Diffusion – then I can “publish” a long cartoon by posting an initial text prompt and initial set along with the workflow details related and form settings. Anyone with that text information and access to renderers can watch my animations.

Now imagine that this cartoon stars Mickey Mouse.

Will Disney sue me for posting a text prompt of a few hundred characters, an integer, and a handful of key/value pairs that define model settings and workflows? That would be pretty ridiculous, even by Disney standards. It’s hard to see the courts going along with it or anyone being able to enforce it if they did. We’re talking about an amount of text that’s probably so small that I could hand it out as a Notepad.exe screencap.

At the point where the above is possible — a point that comes much sooner than you can imagine — it’s game over for Disney’s ability to maintain its position as the exclusive global provider of fresh Mickey Mouse content at scale. (Before you quibble that Disney’s inner workings will be marked by the quality and creativity of the writing, I invite you to watch any random episode of Mickey Mouse Club And ask yourself how much quality or creativity a typical mid-level Mickey project on a Disney watch gets.) We’d all be able to get into Mickey’s work by just passing some text and JSON around and getting it into forms that are open-source and widely available.

derivative works

This script — a prompt, a seed, and some parameters go to a bunch of AI models, and the movie comes out on the other end — isn’t clearly illegal under current law. It should be the same legally as creating a picture of Mickey Mouse with Procreate or Photoshop and then hanging it in your bedroom: just a private, personal, non-commercial creation of some derivative work that is seen only by the creator and is not stored, transmitted, published, distributed or profited in any way.

In short, if I make a new photo (or video, 3D rendering, audio file, graphic novel, etc.) of Mickey Mouse on my laptop, It stays on my laptopthen it doesn’t really seem possible to say that I broke any laws, no matter what software I used to create that image.

So the legal arguments in the Stable Diffusion lawsuit go to great lengths to frame this scenario as illegal. This isn’t the place to go into the plaintiffs’ argument in detail, but the gist of it is that the Stable Diffusion model’s weights file – the file containing the trained neural network that Stable Diffusion uses to generate the images – is itself a derivative of the work of all the billions of images in the training data, and that Stability The AI ​​benefited from this work without compensating the copyright holders of the training data.

There are plenty of other influential parts of the technical and legal arguments outlined in the class action complaint, but this one, which appears early in the document on page 3, is the deadliest of the entire generative AI project: Here to the output of AI Image products In addition to the AI ​​Image products themselves—which contain compressed copies of the copyrighted works on which they have been trained” (emphasis added).

If they can make this charge consistent, they have a chance to force tech platforms to give these model files a try Same treatment They are currently offering other forms of digital bootlegging: child pornography, pirated movies, cracked software, malware, 3D printed weapon files, and so on. It would be very difficult to host a form file like this publicly, and anyone caught doing so can expect a takedown notice.

This, then, would be the end of generative AI — or, at least, generative AI in any kind of decentralized form. Closed companies and models like OpenAI will still be around, because they are centralized, controlled, censored, willing and able to filter what users can and cannot do with their products. Here in the US, and possibly in any country with compliant copyright laws, all AI models will be locked securely behind APIs, and innovation in this area will slow dramatically.

Disney will be able to work with Microsoft, OpenAI, Google, and other big tech platforms to use these large, closed-source models to replace the teams of artists they currently employ. They can release most of their talents and exchange them for AI on the cheap It generates infinite new content from their extensive catalog of existing intellectual property. Meanwhile, independent creators and non-commercial users would be prevented from using these same software tools to compete with the Disney of the world — they would be stuck in an era of making art the old-fashioned way.

That seems to be what the plaintiffs want in this lawsuit. But I don’t think they fully realize what it would mean for them if they left all generative AI to just big content.

Leave a Reply

Your email address will not be published. Required fields are marked *