Midjourney Seeks to Force Studios to Reveal Their Own AI Usage

Midjourney filed a motion on July 4, 2026, asking a federal judge to overturn a ruling that shielded Disney, Universal, and Warner Bros. from producing their internal AI documentation. The AI image generator argues the studios may be doing exactly what they have accused Midjourney of doing: training AI systems on copyrighted content without authorization.

That framing turns the case on its head. The defendant is using a discovery dispute to demand that three of the largest studios in Hollywood explain on the record what their own AI tools are built on.

Judge John Kronstadt has not yet ruled on the motion. The outcome will determine whether this case stays a narrow copyright dispute or becomes the first federal proceeding to treat major studio AI development as a matter of public legal record.

The Lawsuits That Started This

Disney and NBCUniversal filed suit against Midjourney in 2025, alleging the platform generates recognizable likenesses of their protected characters from text prompts alone. The complaint named Bart Simpson, Darth Vader, and other iconic characters as examples of outputs produced without consent or licensing. Warner Bros. Discovery joined the action months later with parallel claims over Superman and Batman.

The studios' core argument is that Midjourney trained on copyrighted visual material and built a system that lets users reproduce protected characters on demand. They are seeking damages and an injunction against specific outputs. The case is before Judge John Kronstadt.

The suits represent the entertainment industry's most concentrated legal action against an AI image generator to date. Unlike earlier cases brought by individual photographers or stock image companies against generative AI companies, this litigation aggregates IP claims across three major studios and thousands of protected characters under a single proceeding. The legal theory, that training on copyrighted images without licensing constitutes infringement, has not yet been tested at trial.

Midjourney has argued in court that its training process constitutes fair use under copyright law, drawing on precedents from search engines and other transformative technologies. That defense has not yet been tested on the merits. The discovery phase, currently underway, determines what evidence each side is allowed to present before any trial on the merits begins.

A parallel case, Getty Images vs. Stability AI, is working through federal court on a similar training data theory. That case involves a single plaintiff and a more contained set of images. The Midjourney litigation, with three major entertainment conglomerates and a character catalog spanning nearly a century, carries broader implications for how courts will ultimately define the line between lawful AI training and infringement.

Multiple other AI copyright suits are proceeding simultaneously, including actions by authors against large language model developers and by music companies against AI music generation platforms. The Midjourney discovery motion is the most direct attempt by any AI defendant to flip the lens back on a plaintiff. If it succeeds, defendants in other pending cases will consider the same strategy, particularly in suits where the plaintiff is itself an active developer of AI systems.

A Magistrate Narrows Discovery

In June 2026, a magistrate judge ruled on a dispute over what documents Midjourney could obtain from the studios. The ruling limited discovery to AI materials related to "consumer facing" uses, meaning publicly released content where AI was visibly involved. Internal AI used for development, preproduction work, research systems, and visualization done before filming fell outside that definition.

The order protected a significant portion of each studio's AI activity. Internal storyboarding tools, script analysis systems, or generative models used during creative development would not need to be disclosed under the June ruling.

That boundary is what Midjourney is now asking Judge Kronstadt to erase. The company argues the line the magistrate drew is precisely the one that prevents the most legally relevant documents from being produced.

In federal litigation, a party challenging a magistrate's discovery ruling must show the ruling was clearly erroneous or contrary to law. Midjourney's motion argues both. It contends the magistrate applied too narrow a definition of relevance and protected material that goes directly to the fairness of the proceeding.

The July 4 Motion

The studios' lead attorney called the request a "fishing expedition." The studios have also stated they "do not seek to stop AI technology or even shut down Midjourney's business," framing the lawsuit as a narrow claim about specific character reproductions rather than a challenge to AI training as a practice.

That framing is precisely what Midjourney is contesting. The motion asks whether the studios can maintain that position while shielding documentation of their own AI development from the proceeding.

What Each Category Would Reveal

The five document categories Midjourney is seeking each target a specific legal question.

Business plans for AI development would reveal whether any of the three studios are building or licensing generative AI systems, what those systems are designed to do, and what competitive analysis the studios conducted before filing suit. Internal research reports would show what technical AI work studios have funded and what their own engineers and researchers understand about how large AI training pipelines operate.

Training datasets are the most significant category. If Disney, Universal, or Warner Bros. have assembled visual datasets for AI training purposes, that material is the direct analog to what those studios accuse Midjourney of building. Model weights would confirm whether any studio has deployed internal generative AI, and would make it possible to analyze what the model knows and how it was built.

The prompt and output data from Midjourney is a separate category entirely. Midjourney is arguing that records of studio employees actively using Midjourney before the lawsuit was filed, and keeping the outputs, would demonstrate that studio personnel valued and relied on the same system their legal team is now seeking to shut down.

Taken together, the five categories form a coherent argument. Midjourney is not fishing for unrelated material to embarrass the studios. It is seeking to establish, through the studios' own records, whether the production and development practices being litigated are specific to Midjourney or widespread across the industry, including at the companies doing the suing.

What the Studios Say and Why

The studios' response to the July 4 motion rests on two arguments. The first is procedural: the magistrate's June ruling was correct, and internal AI development falls outside the scope of what is relevant to a claim about specific character reproductions. The second is substantive: even if studios use AI internally, that use has no bearing on whether Midjourney's training process infringed specific copyrights.

Studios told the court they "do not seek to stop AI technology or even shut down Midjourney's business." That formulation is deliberate. By framing the case as a narrow claim about output, not a broad challenge to AI training, the studios argue that Midjourney's demand for internal documents is simply not responsive to the case as pleaded.

That argument has a structural problem that Midjourney is exploiting. If the studios' case is genuinely limited to stopping specific character reproductions, then Midjourney's training data defense should not be in play at all. But the studios' lawsuit explicitly challenges training on copyrighted material as a basis for liability. That opens the door to questions about what training practices are and are not standard in the industry.

The studios have argued that any internal AI tools are development tools used in planning and production logistics, not systems that generate copyrighted content for public release. That position, they say, is why the magistrate correctly excluded those tools from discovery. Midjourney's counter is that the distinction between internal and external use is legally arbitrary when the question before the court is what the underlying training pipeline looks like and what data went into it.

How Judge Kronstadt resolves that definitional dispute will determine the scope of the trial record. A narrow reading keeps the case focused on Midjourney's specific outputs. A broad reading makes it a referendum on how the studios themselves build AI.

The Industry Custom Defense

Midjourney's demand for internal studio documents is not incidental to the defense. It is a specific legal strategy built around copyright law's recognition of "industry custom" as a factor in infringement analysis. If a court is shown that training AI on unlicensed third party content is standard practice across the industry, including at the studios bringing the case, that evidence directly challenges the claim that Midjourney's approach is uniquely wrongful.

Three studios that have sued for copyright violations over AI training could be compelled to explain, on the federal record, what data went into their own AI systems. No prior major AI copyright defendant has attempted this strategy at this scale. Judge Kronstadt has not yet ruled on the motion.

The argument also exposes a structural tension the studios would rather not navigate in open court. Studios have spent two years publicly positioning themselves as defenders of creative rights against AI companies. The discovery motion asks whether that position holds when applied to their own internal AI programs.

Studios have made some AI disclosures in other contexts. During SAG-AFTRA and WGA contract negotiations in 2023 and 2024, studios were required to describe certain AI applications to union representatives as part of bargaining over AI protections. Those disclosures were bounded, not made public, and limited to uses directly affecting covered workers. Midjourney is now asking for a category of disclosure that is broader in scope and entered into the public federal court record.

The distinction matters for the long term. Union negotiations produced contractual guardrails on AI applied to human performers. Federal litigation could produce a precedent about what studios must reveal to any party they sue over AI, not just the unions whose members they employ.

Courts assessing copyright claims sometimes consider whether a practice was standard in the relevant industry at the time the allegedly infringing work was created. The question is not whether the practice is technically legal but whether the defendant had reason to believe their conduct aligned with what others in the field were doing. If studios are themselves training AI on unlicensed material, that pattern is legally significant. Not as an excuse for Midjourney, but as evidence that the harm the studios are claiming is not specific to Midjourney's behavior.

That argument, if accepted, does not automatically win the case for Midjourney. Copyright law does not permit widespread infringement to legalize more infringement. But it can affect how courts weigh the credibility of claimed harm and whether broad injunctive relief is appropriate when the conduct being challenged is common across the industry, including among the plaintiffs.

Beyond industry custom, Midjourney is also preserving a fair use defense. Courts apply a four factor test that includes whether the use is transformative, the commercial nature of the use, the amount of protected material taken, and the effect on the market for the original work. The internal studio AI question is directly relevant to that fourth factor. If Disney, Universal, and Warner Bros. are building competing AI image systems trained on similar datasets, their own conduct becomes evidence in the market harm analysis.

What a Disclosure Order Would Mean for Hollywood

If Judge Kronstadt overturns the magistrate's ruling, it would mark the first time any major studio has been required to produce internal AI training documentation through federal litigation. Model weights, training datasets, and AI development plans are precisely the documents studios have chosen not to disclose to the public, to union negotiators, or to awards bodies making eligibility decisions.

Hollywood studios have repeatedly disclosed AI use only after press scrutiny forced it out. Secret Invasion's AI title sequence, The Brutalist's voice processing, and Emilia Pérez's vocal work all surfaced through journalism rather than studio statements. A court order requiring three studios to produce AI training records would apply a fundamentally different kind of pressure.

The motion also intersects with other legal efforts to define AI's boundaries in entertainment. The NO FAKES Act, which cleared the Senate Judiciary Committee by unanimous voice vote in June 2026, creates authorization requirements for AI replicas of individual performers. The Midjourney case targets something structurally distinct: what went into building AI systems, not what those systems produce. Both legal tracks will run simultaneously through the second half of 2026.

The Google and A24 partnership announced June 22 made one studio's AI investment explicit and publicly accountable, with Demis Hassabis of DeepMind named alongside A24's leadership in press releases and interviews. The Midjourney motion is aimed at AI investments studios have not chosen to make public. Whether Judge Kronstadt orders production of those documents is one of the more consequential pending questions in entertainment law this year.

The disclosure question extends beyond the three studios in this suit. If studios are compelled to reveal AI training practices in litigation they initiated, every studio with an active AI program will have to consider whether filing a copyright suit exposes its own internal AI infrastructure to scrutiny. That calculus could shift how and when studios choose to litigate AI copyright claims in the future.

The outcome of this procedural dispute could force a reckoning with AI disclosure standards that legislation has not yet required and contract negotiations have not yet resolved. It would, for the first time, treat a studio's own AI practices as legally relevant material rather than proprietary business information.

If the motion fails and the magistrate's ruling stands, studios retain the ability to conduct internal AI development behind a discovery shield. The case would then narrow back toward its original question: whether Midjourney's specific training and output practices infringe specific copyrights. That is still a consequential question, but a much smaller one.

The path of this single procedural motion will determine which version of the case the industry actually has to reckon with. A court decision on the motion is expected before the end of 2026. Whatever the ruling, the motion has already put the studios on record about what they believe they can and cannot be asked to reveal. That record does not disappear when the procedural dispute is resolved.

The pattern across the case so far suggests both sides are preparing for trial rather than settlement. Three major studios coordinated on the same legal theory, and Midjourney responded with a discovery posture aimed at upending the evidentiary record before any hearing on the merits. Neither posture is consistent with a party looking for an early exit.

Any AI company sued over training data after this case will study how Midjourney's discovery strategy played out. If the approach succeeds and forces studios to produce internal AI documentation, it becomes a model for defendants in future copyright cases. If it fails, studios will have established that their internal AI practices are off limits in copyright litigation even when they are the ones bringing the case.

Filmmakers building with AI tools today can generate images and video through AI FILMS Studio while the courts continue to define where the industry's legal limits actually are.

---

Sources

TechCrunch | Variety | The Hollywood Reporter | The Wrap | IndieWire