Hollywood's fight against alleged AI copyright infringement has only just begun

It was only a matter of time before the major Hollywood studios started taking the fight to the artificial intelligence industry over its alleged abuse of intellectual property. Now, it’s on. Last week, Walt Disney Co. and Universal Pictures sued AI firm Midjourney in U.S. District Court in Los Angeles, accusing the popular image generator of blatantly copying and profiting from copyrighted images of characters from franchises such as “Star Wars,” “Minions,” “Cars,” Marvel, “The Simpsons” and “Shrek.”

The complaint cited numerous examples, illustrated with dozens of striking photos, of San Francisco-based Midjourney’s technology being used to generate virtually indistinguishable copies of Darth Vader, Iron Man, Bart, Woody and Elsa, sometimes in frames quite similar to scenes from the actual movies and TV shows. The lawsuit says Midjourney employed such images to promote its subscription service and encourage the use of its image generator. The companies are seeking unspecified monetary compensation, as well as a court order to stop Midjourney from further infringement, including by using studio-owned material to train its upcoming video tool.

Implications of the Lawsuit

“Midjourney is the quintessential copyright free-rider and a bottomless pit of plagiarism,” Disney and Universal’s lawyers wrote in the 110-page complaint. “Piracy is piracy, and whether an infringing image or video is made with AI or another technology does not make it any less infringing.” The stakes of this battle are high, according to the studios. The AI company’s misuse of Disney and Universal’s intellectual property “threatens to upend the bedrock incentives of U.S. copyright law that drive American leadership in movies, television, and other creative arts,” the court document said. Midjourney has not responded to requests for comment.

Arguments and Concerns

AI companies have typically argued that they are protected by “fair use” doctrine, which allows for the limited reproduction of material without permission from the copyright holder. Midjourney founder David Holz in 2022 told Forbes that the company did not seek permission from copyright holders, saying “there isn’t really a way to get a hundred million images and know where they’re coming from.” This battle is a long time coming. Artists — including screenwriters, animators, illustrators and other entertainment industry workers — have been raising the alarm for years about the threat of AI, not just to their actual jobs but to the work they create.

AI models are trained on anything and everything that’s publicly available on the internet, which includes copyrighted material owned by studios or the artists themselves, they argue. The Writers Guild of America last year called on the big entertainment companies to take legal action against tech giants and startups in order to put a stop to such “theft.” But this is the first time any of the major film studios have gone after an AI company for copyright infringement. They may not be the last.

Industry Response

The studios are following the lead of the New York Times and other publishers, who sued OpenAI and its backer Microsoft over alleged plagiarism. The major music labels have also taken AI firms to court over the use of copyrighted music. Studios are in an awkward position because they’re weighing the possibility of licensing their content to AI firms or using the technology for their own purposes.

Reid Southen, a Michigan-based film concept artist whose research on AI was cited at length in the lawsuit, said he hopes Disney and Universal’s complaint encourages others to take a similar stance. “Hopefully, I think other studios are looking at what’s going on with Disney and Universal now, and considering, ‘Hey, what about our properties?’” said Southen, who has worked on studio films including “The Matrix Resurrections,” “The Hunger Games” and “Blue Beetle.”

Southen's Involvement

Southen became part of the story in December 2023, after the release of Midjourney v6 started making waves online. He saw someone use the tech to generate an image of Joaquin Phoenix as the Joker, and he started messing around with it himself to see what kinds of copyrighted material he could prompt it to rip off. He posted the results on social media, which led AI researcher Gary Marcus to reach out.

Marcus and Southen published an in-depth article for IEEE Spectrum in January 2024, making the case that Midjourney and other well-funded AI firms were training their models on copyrighted work without their permission or compensation and spitting out images nearly identical to the studios’ own material. That article illustrated how simple prompts could produce nearly exact replicas of famous film and TV characters.

Concerns about AI Technology

The prompts didn’t necessarily need to ask for a particular character by name. The researchers were able to coax uncanny images from AI with prompts as basic as “animated toys” (resulting in pictures of “Toy Story” characters) and “videogame plumber” (which turned up versions of Mario from “Super Mario”). According to Marcus and Southen, all it took was the phrase “popular movie screencap” to evoke a picture similar to an actual frame from “Batman v. Superman: Dawn of Justice” or “The Dark Knight.” “It shows that they are very clearly trained on hundreds, if not thousands, of movies and YouTube videos and screen caps and all this stuff, because I was able to find matching screen caps and images, not just from trailers, but from deep in movies themselves,” Southen said.

Future of AI and Intellectual Property

The Midjourney examples were the most egregious, Southen said, but the company was not the only offender. For instance, OpenAI’s image generation technology DALL-E was also capable of producing “plagiaristic” images of copyrighted characters without prompting them specifically by name, Southen said, echoing the findings of his and Marcus’ IEEE Spectrum article. OpenAI did not respond to a request for comment. The Disney and Universal lawsuit did not name OpenAI, which is also responsible for the video generator Sora that is trying to take the film business by storm.

Many chatbots and text-to-image tools have guardrails around intellectual property, but they clearly have limitations. Ask ChatGPT to create an image of Kermit the Frog, and it will flatly reject the request. However, for example, I was recently able to request a picture of a Muppet-like female pig character, and the result was not unlike Miss Piggy, though I wouldn’t quite say it was a one-for-one copy.