Disney, Universal and Warner Bros. v. Midjourney: Why Midjourney Wants Discovery Into the Studios' Own AI Use
A guide to the studios' copyright case against Midjourney: the claims, the consolidated timeline, the magistrate judge's ruling limiting discovery to consumer-facing AI apps, Midjourney's fair use and unclean hands arguments, related cases like Andersen v. Stability AI, and what the outcome could mean for AI copyright litigation.
Midjourney, sued by Disney, Universal and Warner Bros. Discovery for copyright infringement, is trying to turn part of the case toward the studios’ own use of generative AI. In June 2026, a magistrate judge limited Midjourney’s discovery to the studios’ consumer-facing AI applications, finding broader requests irrelevant or protected. Midjourney then asked the district judge to overturn that limit, arguing that the studios’ internal AI practices bear on its fair use and “unclean hands” defences (TechCrunch, Variety). The dispute will not decide whether Midjourney infringed, but it will signal how far AI defendants can probe plaintiffs’ own AI use, an issue likely to recur across AI copyright cases.
Case at a glance
| Item | Detail |
|---|---|
| Case | Disney Enterprises, Inc. v. Midjourney, Inc., No. 2:25-cv-05275 (C.D. Cal.), consolidated with Warner Bros. Discovery’s action |
| Plaintiffs | Disney and Universal (filed June 11, 2025); Warner Bros. Discovery (filed September 4, 2025) |
| Core claims | Midjourney’s service can generate images of the studios’ characters, such as Darth Vader and Bart Simpson, and was allegedly trained on their works without permission |
| District judge | John A. Kronstadt |
| Discovery ruling | Magistrate Judge Joel Richlin limited discovery into the studios’ AI use to consumer-facing applications (June 2026) |
| Midjourney’s response | Motion asking the district judge to overturn the limit |
Sources: CourtListener docket; Variety; Law360; The Next Web.
The studios’ case
The studios argue that Midjourney trained its models on copyrighted material and that its service readily produces images of characters they own, making it liable for infringement. Their complaints include side-by-side examples of generated images resembling famous characters. The claim follows a pattern seen across AI copyright litigation: models trained on copyrighted works can reproduce recognisable elements of them, and the companies behind the models should bear responsibility.
Midjourney’s strategy
Rather than only defending its own training practices, Midjourney sought discovery into how the studios use generative AI, including internal tools, training data practices and policies. Its argument is that if studios themselves use or train image models on unlicensed material, that is relevant to industry custom and to whether training is fair use. It also invokes “unclean hands”, an equitable defence that can bar relief to a plaintiff who has acted improperly in relation to the same matter.
The studios’ position, accepted in part by the magistrate judge, is that their own practices do not determine whether Midjourney infringed. Ordering production only about consumer-facing AI applications reflects that view: some studio AI use may be relevant, but broad requests into internal tools went too far.
Key legal terms
- Fair use: a U.S. defence that weighs the purpose of a use, the nature of the work, the amount used and the effect on the market. AI developers argue training is transformative; rights holders argue it substitutes for licensed uses.
- Discovery: the pre-trial exchange of documents and testimony between parties.
- Unclean hands: an equitable doctrine that can deny relief to a party whose own misconduct relates to the claim.
- Magistrate judge review: a district judge can modify a magistrate judge’s ruling on non-dispositive matters such as discovery if it is clearly erroneous or contrary to law.
Why discovery matters
In U.S. federal litigation, discovery lets each side obtain evidence relevant to claims and defences, subject to proportionality and privilege limits. Discovery fights often shape cases as much as the final ruling, because documents produced can affect settlement leverage and public perception. A broad order would expose studios’ internal AI use to scrutiny; a narrow one keeps the case focused on Midjourney’s conduct.
The dispute also highlights a tension in the industry. Studios have publicly discussed experimenting with generative AI in areas such as concept art, previsualisation, visual effects and marketing, and their research arms publish work on AI-assisted animation and effects. None of that is unlawful, but Midjourney is betting that it complicates the narrative of AI image generation as inherently infringing.
Related cases to watch
| Case | Parties | Status and relevance |
|---|---|---|
| Andersen v. Stability AI | Artists v. Stability AI, Midjourney, DeviantArt, Runway | Proposed class action in the Northern District of California; key copyright claims allowed to proceed in August 2024 (The Verge) |
| Getty Images v. Stability AI | Getty v. Stability AI (U.S. and U.K.) | Concerns training on Getty’s images and reproduction of watermarks |
| Studio cases against other AI companies | Various | Will test how training-data and output-similarity arguments fare across media |
For a broader overview of the legal landscape, see our guide to AI art and copyright law.
What comes next
If the district judge broadens discovery, the studios would have to produce records of internal AI tool usage, training data practices and policies, which could become evidence on fair use and industry practice. If the limit stands, the case proceeds largely on the original copyright claims, centred on Midjourney’s training data and outputs. Either way, rulings on fair use and on output similarity will matter far beyond this case. Readers following the case can check the public docket for new filings and orders.
Why it matters for creators and smaller companies
A ruling that favours the studios on copyright while keeping their own AI use out of view could produce an uneven landscape, in which large rights holders adopt AI freely under licences and internal policies, while smaller AI developers and users face greater legal exposure for similar tools. Conversely, a strong fair use ruling for Midjourney would reduce exposure for AI developers but leave artists and rights holders seeking compensation through licensing deals or legislation.
For independent artists and studios using AI tools now, practical steps include reviewing each tool’s terms on training data and indemnity, avoiding prompts that target recognisable copyrighted characters, and keeping records of how AI was used in client work.
Our analysis
Midjourney’s discovery push is a smart litigation tactic with a limited chance of changing the core outcome. Whether a plaintiff also uses AI does not answer whether a defendant’s training and outputs infringe, which is why the magistrate judge drew the line at consumer-facing applications. The more consequential questions remain fair use for training and liability for outputs that closely resemble protected characters. Still, the dispute exposes a real tension: the companies most aggressive about protecting their IP from AI are also adopting it. Clear licensing norms for training data would serve both sides better than years of case-by-case litigation.