Midjourney wants Hollywood to open its AI files

Why it matters

Midjourney is trying to turn Hollywood's AI copyright case back on the studios by making their internal AI use part of the evidentiary record.

Midjourney asked a federal judge to order Disney, Universal and Warner Bros. to turn over broader records on their own generative AI use, turning a Hollywood copyright case into a test of how much the studios must reveal about the tools they use.

The dispute, reported Saturday by TechCrunch, sits in a Central District of California lawsuit over whether Midjourney's image and video models infringe studio copyrights when users generate characters tied to franchises such as The Simpsons and Star Wars. In a recent motion, the company asks the district judge to review a magistrate judge's discovery order.

For Midjourney, the court fight targets what it sees as a contradiction in the entertainment industry's AI posture: studios are suing over models trained on copyrighted media while also experimenting with generative tools for production, pre-production and internal research. Midjourney's legal team is trying to make that contradiction discoverable.

The discovery fight

The magistrate judge had already required the studios to produce some information about their generative AI use, but only when that use led to "consumer-facing" images or videos. Midjourney wants that limit removed.

In the filing, Midjourney argues that the consumer-facing line lets the studios "cherry-pick" documents that support their market-harm theory while withholding documents useful to Midjourney's defenses. The company says internal AI use could matter even if the resulting work never reaches an audience. Its example is direct: if a studio uses image-generating models internally for storyboarding or ideating film and television content, and those models are trained on unlicensed third-party copyrighted works, Midjourney argues that evidence could show industry custom around AI training.

That framing connects to Midjourney's core defense of fair use. The studios say Midjourney copied protected works and lets users create unauthorized derivatives of famous characters. Midjourney says training on copyrighted images can be permitted under fair use and that the studios' own practices may undercut claims that such use is outside accepted industry practice.

The filing also attacks the practical ambiguity of the consumer-facing standard. Midjourney asks whether internal scene preparation later used in a movie counts as consumer-facing, whether an internal tool for writers or illustrators counts if it shapes a confidential pilot, and whether a stealth AI research project becomes relevant only when it becomes a product.

Prompts are now evidence

Midjourney is also asking for all prompts the studios or their agents entered into Midjourney and the resulting outputs, not only the prompts behind the images cited in the complaints.

The argument is simple enough for any founder who has watched a demo get reverse-engineered in a boardroom: a prompt is part of the experiment, and the output alone does not tell the whole story. Midjourney says it needs the full prompt record to test whether the studios are presenting curated examples of infringement and whether those examples came from investigators acting on the studios' behalf.

The filing says the magistrate judge treated undisclosed prompts and outputs as attorney work product. Midjourney disputes that characterization, arguing that prompts such as a simple character request are factual inputs into a third-party platform, rather than legal strategy. Midjourney also says the studios submitted prompts under its terms of service, which it argues made the inputs public by default and gave Midjourney rights to use them.

An attorney for the studios, David Singer, previously called Midjourney's discovery push a "fishing expedition," according to TechCrunch's summary of Variety's reporting. Singer also said the studios are not trying to stop AI technology or shut down Midjourney, but want Midjourney to stop copying and distributing works tied to the studios' movies, shows and characters. Variety coverage

The lawsuit widened after Disney and Universal

Disney and Universal sued Midjourney in June 2025, accusing the company of training on their content without permission and generating images of studio-owned characters. TechCrunch reported at the time that the startup’s models could create images of characters such as Bart Simpson and Darth Vader. Warner Bros. followed in September 2025, alleging that Midjourney enabled users to create images and videos of characters including Superman, Bugs Bunny, Batman, Wonder Woman and Scooby-Doo. The Associated Press reported that Warner Bros. sought up to $150,000 in damages per infringed work and that Midjourney had denied copyright infringement allegations in the Disney and Universal case.

The discovery outcome now matters beyond this case. If Midjourney wins broader discovery, the record could expose studio AI workflows that have remained private. If Midjourney loses, the studios preserve a narrower record focused on public-facing outputs and the examples they chose to plead.

How Midjourney describes itself

Midjourney describes itself as a community-funded research lab of 60 people and says it is "lean, self-funded" and distributed. It lists image and video models as current work, with software, medical and hardware projects also on its roadmap.

The immediate question before the court is procedural: what documents and prompts the studios must produce. The strategic question is larger: whether an independent AI company can force Hollywood to show its own AI work before the industry defines the boundaries for everyone else.

Sources: TechCrunch, CourtListener filing, TechCrunch (2025), AP (Disney/Universal), AP (Warner Bros.)