The Hollywood Characters That Could Break Midjourney’s AI Magic Spell
Midjourney calls it imagination, Warner Bros., Disney, & Universal are calling it theft.
Hollywood studios are finally getting around to filing lawsuits against some of the biggest AI image generators on the market. And while the AI space is a seemingly target-rich legal environment, the primary focus from movie studios right now appears to be Midjourney.
That makes sense for two reasons: (1) It’s the AI tool many users consider the best in class in terms of rendering aesthetically human-like imagery, and (2) Midjourney finally launched its image-to-video function in June, effectively seeding the beginnings of a new kind of movie studio where no soundstages or actors are needed.
Warner Bros. filed its 87-page lawsuit yesterday. And yes, I read every single page. In form and overall substance, the filing is largely a mirror of the lawsuit Disney and Universal jointly filed on June 11, which also alleged copyright infringement on the part of Midjourney. Both legal filings exhaustively list their most prized characters they allege were found in Midjourney’s outputs. The characters listed read like a who’s who of the last half-century of cinema hits, including Superman, Batman, Spider-Man, The Simpsons, The Avengers, Star Wars, and many more (see the chart below for the blow-by-blow breakdown).
Another similarity the two lawsuits share is that they both mention a 2022 article in Forbes in which Midjourney founder and CEO David Holz answers a few pointed questions about whether he thinks his AI tool will negatively impact working human artists. In fact, both lawsuits use the exact same quotes from the article:
In particular, Midjourney CEO David Holz stated in a Forbes interview that Midjourney has “a famous character designer using [its] product.” Holz further opined that the entertainment industry will use systems like Midjourney to “try to cut costs” and that “some people will try to cut artists out.”
I found the selection of these quotes curious because when you actually read the nearly three-year-old interview, conducted by Rob Salkowitz, the aforementioned quotes are hardly the most damning (at least from the perspective of anti-AI working artists).
Question: How was the dataset built?
Holz: It’s just a big scrape of the Internet. We use the open data sets that are published and train across those. And I’d say that’s something that 100% of people do. We weren’t picky.
Question: Why would you point this technology at that level of the economy as a kind of business focus and priority for the stuff that you're doing?
Holz: It’s important to emphasize that this is not about art. This is about imagination. Imagination is sometimes used for art but it's often not.
Question: Nevertheless…this is very disruptive of [the artist] economy.
Holz: I think it's like we're making a boat, and somebody can race with the boat, but it doesn't mean that the boat's about racing. If you use the boat to race then maybe like, yeah, sure. In that moment it is.
Ghosts in the Dataset
So why didn’t the lawsuits, which are both being administered by David Singer on behalf of the studios (hence the similarity in filings), focus on these parts of the interview, which seem to indicate a lack of regard for the human-created artworks used to train Midjourney? I think the answer is, well, that ship (or boat, as Holz might call it) has already sailed.
If the studios were hoping to stop a company like Midjourney from training on its Internet-hosted images, or in some other way rooting out the problem at the source, the time to do that would have been in 2021 or 2022. But now there are so many data sets in the wild, including open source data sets, like LAION (Large-scale Artificial Intelligence Open Network), there’s really no way to put the training data genie back in the proverbial bottle.
However, I don’t think this waiting game was a deliberate gotcha strategy on the part of the studios. I think the truth is that they didn’t fathom just how rapidly AI image generation tools would improve and then give rise to AI video. So now there’s only one move left in their arsenal: sue for copyright infringement. It’s unclear whether the ultimate goal is to force a damages payment from Midjourney or perhaps gain legal leverage that would make any potential licensing deal overly weighted in favor of the studios.
The Imitation Artists
Following the Disney/Universal lawsuit, Midjourney responded confidently, denying any infringement and instead framing itself as a mere tool for its users.
“Midjourney’s AI models had to be trained on billions of publicly available images, which are paired with text descriptions, in order to learn visual concepts and how they correspond to language,” Midjourney’s legal response read on August 6. “This process is akin to how humans learn to draw or paint—not by memorizing individual artworks, but by internalizing patterns and techniques through repeated exposure and practice.”
Translation: Our AI should be treated like a human who consumes several years' worth of their favorite artist’s style on the internet, and then proceeds to mimic that style, but in a wholly original way that isn’t simply a digital copier machine. But there are a couple of problems with that notion (Note: I’m not a lawyer, so these aren’t legal arguments, but observations I’ve made since I began working with AI tools in early 2022.)
The first issue came up in a 2023 case in which Getty filed a lawsuit against Stability, alleging infringement, and as proof, detailed outputs that included partial Getty watermarks on theoretically “original” images. Although Stability later managed to fix that output problem to eliminate the watermark issue, it at least cast doubt on the AI industry’s claim that AI outputs are always transformative.
The second issue revolves around the idea of framing the training and outputs of an AI system as similar to the training and renderings of a human. I understand the reasoning of making such an analogy, but that then raises several new questions: Are humans just flesh-covered machines that are simply less efficient than computers when it comes to reproducing our influences, or are humans imbued with a special, undeniably unique and human element that distinguishes our work, be it visual art, writing, or music, from that of algorithms housed in metal boxes?
This might sound like existential philosophy more suited to a 3 a.m. wine-soaked house party layered in incense smoke, but I think these are the questions the courts, AI companies, and humanity as a whole are being gradually pushed toward. What is the legal difference between a thinking machine in a box and a human mind and soul? Is there a difference? And if there is, in the face of AI, what new legal parameters can we possibly design to clarify all of this?
I happen to be an avid Midjourney user, as well as an admirer of David Holz’s approach to technology, all the way back to when I covered his first company, Leap Motion. And I have a deep love affair with movies, and believe the studios have the right to protect the franchises they fund and help to produce. So I’m hoping there’s an outcome that makes everyone happy. But I don’t honestly expect these lofty ideas about the differences between humankind and AI to gain much purchase in a courtroom fight between Hollywood studios and AI startups like Midjourney.
Meanwhile, Midjourney faces the legal wrath of three studios that collectively, via their parent companies, represent a market cap of roughly $370 billion, or just over a third of a trillion dollars. Just on financial might alone, it is difficult to envision a reality in which Midjourney simply “fights it out” against these studios indefinitely. Either time and dwindling funds, a backroom settlement, or a fundamental change in AI law are what will likely decide these cases.
My prediction is that, like other lawsuits, this latest legal skirmish is more likely to lead to a legal détente that ends with an agreement that somehow satisfies the studios, and makes generating certain images on Midjourney a little tougher.