A San Francisco image generator facing a copyright suit that could bankrupt it now wants the studios suing it to hand over their internal AI experiments. The fight over what counts as evidence could reshape how every AI copyright case plays out.
The AI startup at the center of Hollywood's biggest copyright fight has stopped playing defense. In a motion filed on July 4, Midjourney asked a federal judge to force Disney, Universal, and Warner Bros. Discovery to reveal how they use artificial intelligence inside their own operations, arguing that the studios accusing it of theft may be running the same playbook behind closed doors.
The filing lands in the U.S. District Court for the Central District of California, where three of the most powerful entertainment companies on the planet have spent more than a year trying to prove that Midjourney's image models amount to industrial-scale piracy. Now the company wants the court to look in the other direction.
What Midjourney Is Actually Asking For
The demand is sweeping. According to the filing, Midjourney wants the studios to produce their AI business plans, research reports, training datasets, and model weights. It also wants something more unusual: the presentations executives showed their own boards about generative AI.
There's a second demand that stings more. Midjourney is asking the studios to reveal every prompt their employees typed into Midjourney's own service, along with the images those prompts produced. Not just the ones that generated infringing characters. All of them.
The logic is straightforward. If Disney or Universal staff were quietly using Midjourney to spin up concept art, and if the studios were building their own image generators trained on unlicensed material, then the accusation that AI training is inherently unlawful starts to wobble. Midjourney's position is that you cannot sue a company for an activity you practice yourself.
Bobby Ghajar, the attorney representing Midjourney, put the stakes plainly in the motion. He wrote that if the studios are doing the very thing they want to punish, that evidence strikes at the core of Midjourney's fair use and unclean hands defenses.
The Ruling Midjourney Is Trying to Kill
This motion exists because Midjourney already lost a round.
In June 2026, Magistrate Judge Joel Richlin drew a line around what the studios had to disclose. He ruled that they only needed to turn over information about "consumer-facing" AI applications, the tools and outputs that reach the public. Internal AI projects, the experimental systems used for storyboarding or pre-production, stayed off-limits. Richlin found that broad information about the studios' AI use was not relevant to the narrow question of whether Midjourney infringed their copyrights.
The studios accepted that boundary. They agreed to hand over material tied to public-facing AI, and nothing more.
Midjourney's July filing asks Judge John Kronstadt to overturn Richlin's decision. The company argues that the restriction lets the studios control the narrative. In its own words from the filing, the limitation "unfairly" allows the studios "to cherry-pick only those documents they believe support their market harm claims while depriving Midjourney of documents that would support its defenses."
The company pushed the point further, claiming the withheld documents are exactly the ones that would show whether the studios, behind closed doors, are doing precisely what they are suing over.
Why Internal Storyboarding Matters So Much
One specific example runs through the whole argument.
Midjourney contends that if the studios are developing image-generating models trained on unlicensed, third-party copyrighted data for internal work like storyboarding or ideating content for film and television, that fact would prove something valuable. It would establish an industry custom. If even the plaintiffs download and train AI on copyrighted content they don't own, then the practice Midjourney stands accused of starts to look less like rogue behavior and more like standard operating procedure across the entire business.
That framing is deliberate. Fair use analysis in the United States weighs several factors, and one of the heaviest is market harm, whether the accused use damages the market for the original work. If the studios themselves treat AI training on copyrighted material as normal, the "harm" they claim becomes harder to isolate and pin on Midjourney alone.
The Studios Call It a Distraction
David Singer, the lead attorney for the studios, has a name for what Midjourney is doing. He calls it a fishing expedition.
Singer has argued that the demand is an attempt to change the subject and pull attention away from Midjourney's own conduct. His framing of the studios' goal is careful and worth reading closely. He wrote that the studios do not seek to stop AI technology or shut down Midjourney's business. What they want, he said, is for Midjourney to stop copying their movies and TV shows and to stop distributing, publicly displaying, publicly performing, and creating derivative works that include copies of their famous characters without authorization.
He added that these are the same rights any copyright holder would assert against any infringer, whether AI-powered or otherwise. The phrasing does real work. It positions the studios not as enemies of technology but as ordinary rights holders protecting property, a stance designed to keep the case from becoming a referendum on AI itself.
How the Fight Started
To understand the current skirmish, rewind to June 2025.
That month, Disney and Universal filed a 110-page complaint against Midjourney, calling the company a "bottomless pit of plagiarism" and a "quintessential copyright free-rider." The suit came packed with side-by-side comparisons: a copyrighted character on one side, a Midjourney output on the other, often nearly indistinguishable. Homer Simpson. Darth Vader. Shrek. Yoda, lightsaber raised, dropped into a scene that never appeared in any film.
The most damning claim in that complaint was about what Midjourney would produce without being asked directly. The studios showed that generic prompts like "animated toys" or "popular movie screencap" could return highly specific depictions of protected characters, suggesting the models had absorbed those characters deeply during training.
Warner Bros. Discovery joined the pileup in September 2025 with its own 87-page complaint, represented by the same legal team. Its roster of characters read like a century of pop culture: Superman, Batman, Wonder Woman, Bugs Bunny, Scooby-Doo, Tom and Jerry, and Rick and Morty among them. Warner Bros. accused Midjourney of stripping away guardrails that had briefly blocked infringing video outputs, framing the removal as a calculated, profit-driven choice.
The numbers behind Midjourney help explain why the studios see a threat rather than a nuisance. The company reported surpassing $200 million in revenue in 2023 and $300 million in 2024, with a user base in the tens of millions, all built by a startup that once described itself as a small self-funded team of eleven.
Fair Use and the Human-Learning Analogy
Midjourney's central defense is fair use, and specifically the idea of transformative use.
The company argues that training a model on billions of images is fundamentally different from copying and reselling someone's work. It compares the process to how a human artist absorbs influences over a lifetime, learning visual ideas and then producing something new. In this telling, the model doesn't store and spit out copies; it learns patterns. Midjourney has also pointed to its terms of service, which forbid users from creating infringing content, and has argued that responsibility for any bad output sits with the person typing the prompt.
The studios reject the analogy outright. Their position is that a machine generating downloadable, high-quality replicas of Batman on demand is not an artist finding inspiration. It is a vending machine for infringement.
The Disney Wrinkle
There's an irony baked into this case that Midjourney's motion seems designed to exploit.
Of the three studios, Disney has been the most public about its own AI ambitions. In late 2025, it announced a $1 billion investment into OpenAI that would have brought hundreds of Disney characters to the Sora platform. Those plans came apart earlier this year when Sora was shut down, and Disney said it would keep engaging with AI while embracing technologies that respect IP and the rights of creators.
That history is precisely the kind of material Midjourney wants dragged into the open. A studio that publicly courted an AI image and video platform, then sued a different one, is a studio whose internal AI files could complicate its own arguments about harm and custom.
A Ruling That Reaches Beyond One Case
Judge Kronstadt now has to decide whether to uphold Richlin's limit or blow it open. His choice carries weight far past Midjourney's fate.
American courts are already split on how AI training interacts with fair use. The Thomson Reuters v. Ross decision narrowed one common defense, and every new ruling adds another data point to an unsettled area of law. How judges handle discovery, what evidence they let defendants dig for, will shape what future AI companies can gather when they find themselves in court.
If Midjourney prevails on this procedural question, it gains a weapon that other AI defendants will want to copy: the ability to force accusers to prove they aren't practicing what they preach. A discovery order that opens the studios' internal AI operations would give Midjourney the chance to argue that the entire industry, plaintiffs included, treats training on copyrighted material as the cost of doing business.
The studios are seeking statutory damages that could reach up to $150,000 per infringed work for willful violations, along with injunctions to halt the alleged copying. For a company Midjourney's size, a loss on the merits could be fatal. That existential pressure is exactly why its lawyers are fighting this hard over a question that, on its surface, is only about which documents get handed over.
Comments
Join the discussion and share your perspective.