On June 11, 2025, Disney and Universal did something the two studios almost never do: they filed a lawsuit together. Their target was Midjourney, the AI image generator that will, on a simple text prompt, hand you a picture of Darth Vader, a Minion, Elsa, or Yoda. Months later Warner Bros. and DC Comics joined the fight, bringing Batman and Superman with them. The consolidated case — Disney Enterprises Inc. v. Midjourney Inc., No. 2:25-cv-05275, in the U.S. District Court for the Central District of California before Judge John A. Kronstadt — has become the single most closely watched test of whether a generative-AI company can train on, and reproduce, the characters that Hollywood has spent a century building.
The plaintiff roster reads like a studio-lot directory: Disney Enterprises, Universal City Studios, Warner Bros. Entertainment, DC Comics, Marvel Characters, Lucasfilm, DreamWorks Animation, Twentieth Century Fox, Cartoon Network, Turner Entertainment, and Hanna-Barbera. When that many rights holders line up behind a single complaint, it is not really about one product. It is about where the law draws the line between learning from creative work and copying it — a line that matters just as much to a filmmaker with one registered screenplay as it does to a studio with ten thousand.
What the lawsuit says
The studios' theory is unusually concrete for an AI case, and that is what makes it dangerous for the defendant. Most AI-copyright suits live or die on the murky "input" question — whether training a model on copyrighted works is itself infringement. The studios press that theory too, alleging Midjourney copied their films and characters wholesale to build its training set. But their strongest card is the "output": they say Midjourney's service will generate near-identical images of their protected characters on demand, and they filed the lawsuit stuffed with side-by-side exhibits to prove it. The complaint famously called Midjourney a "bottomless pit of plagiarism."
That framing is deliberate. A picture of Darth Vader that a user can generate in seconds is not an abstract idea about a "space villain in a black helmet" — it is the specific, copyrighted expression of a character Lucasfilm owns. By leading with recognizable outputs rather than the invisible mechanics of training, the studios put the most sympathetic possible facts in front of the judge: here is our character, here is the machine reproducing it, here is no license.
Midjourney denies liability and has signaled it will lean on fair use — the doctrine that permits unlicensed use of copyrighted material when the new use is sufficiently transformative. It argues, in essence, that training a model is a transformative act of learning, not wholesale copying, and that the company is not responsible for every image its users choose to generate. Nothing on that central question has been decided. These are allegations and defenses, not findings, and being sued is not evidence of wrongdoing.
What has happened is a bruising discovery fight. Plaintiffs have moved to compel Midjourney to produce the source code behind its newer model versions — the "V8" training code — arguing they cannot prove what the model ingested without it. Midjourney, in turn, went on offense: in mid-2026 it demanded that the studios disclose their own internal use of AI, seeking to show that Hollywood quietly relies on the very tools it is suing over. A magistrate judge narrowed that demand, ruling the studios need only produce material tied to consumer-facing content, and Midjourney has asked the district judge to review that order. The IP Feed's case meter currently reads the overall coverage as roughly balanced, with the early procedural rulings tilting toward the plaintiffs — a signal, not a verdict, and the kind of thing that can flip the moment the fair-use question is actually briefed.
The hard part: characters are property, training is unsettled
Two legal ideas sit underneath this case, and understanding the difference between them is worth more to a working creator than any headline.
The first is character copyright, and it is more powerful than most people realize. Copyright does not just protect a particular film or comic book; a sufficiently distinctive, well-delineated character can be protected in itself, across every work it appears in. That is the principle — built over decades of cases from Mickey Mouse to James Bond to the Batmobile — that lets Disney sue over an image that appears in no specific film still. The character is the protected expression. The flip side, as always in copyright, is that the generic stays free: no one owns "a wizard," "a superhero," or "a yellow cartoon helper." You own the specific, developed version you created.
The second idea is the one nobody has resolved yet: whether training an AI model on copyrighted works is fair use. This is the trillion-dollar question running through every AI case on the docket right now, and Disney v. Midjourney is one of the cleanest vehicles to answer it, precisely because the outputs are so recognizable. If the court treats the reproduction of a protected character as straightforward infringement, the fair-use defense has to carry enormous weight. If it treats training as transformative and the outputs as the users' problem, the calculus shifts. The outcome will ripple far past Midjourney — into every tool that was trained on the open internet, which is to say, nearly all of them.
Why it matters beyond one lawsuit
It is tempting for an independent filmmaker to read this as a clash of giants with nothing to offer the little guy. That is exactly backwards. The studios can afford a twelve-member coalition and a source-code discovery war; you cannot. But the rules this case is about apply to your work at a fraction of the scale, and the habits that put Disney in a strong position are the same ones that protect you.
A few concrete takeaways, drawn straight from the pressure points in this docket:
- Register the work, and register it early. The studios can sue over their characters because ownership is documented down to the frame. For you, timely registration with the U.S. Copyright Office is what keeps statutory damages and attorneys' fees on the table and gives you a dated, official record of what you made. A character lives inside the works you register — so register the screenplay, the series bible, the finished film.
- Know your chain of title cold. Notice that even Disney's biggest early battle in this case is evidentiary — proving what was copied and what was owned. If your rights, options, and assignments are not papered and traceable, you have a great idea and a weak claim. Clean chain-of-title documentation is not back-office busywork; it is the thing that makes a copyright enforceable, and the thing a collection account or distributor will demand before a dollar moves.
- Lead with the output. If a generative tool is reproducing something recognizably yours, that concrete, side-by-side evidence is far stronger than an abstract argument about training data. Document the specific infringing outputs — screenshots, prompts, dates.
- Watch the fair-use line, don't bet your plans on it. The transformative-use question is genuinely unsettled. Until a court draws the line, assume your registered work can be ingested by AI systems, and make your licensing and distribution decisions with that risk priced in rather than wished away.
- Keep the records while you're hopeful. Every registration certificate, assignment, and release. The paperwork you save when a project is exciting is the evidence you need if someone later copies it.
The uncomfortable truth underneath the spectacle is that the law here is still being written, in real time, by the parties with the deepest pockets. The creators who come out of this era protected will not be the ones with the best ideas. They will be the ones whose ownership was documented, registered, and traceable before the dispute — so that if a machine, or a studio, or a distributor ever reproduces their work, they can prove it was theirs first.
We'll track this docket as it develops. The next real signals will come as the discovery fight resolves and the parties finally brief the fair-use question head-on — the ruling that could, for the first time, tell every creator where the line between training and copying actually sits. Follow the filings, parties, and coverage on the case page.
This post is editorial commentary on public court filings and news coverage, not legal advice. The allegations described are unproven, the defendant disputes them, and being named as a defendant is not evidence of wrongdoing. Details are drawn from the docket and press reports and may be refined as the case develops.