Disney & Universal v. Midjourney

A vending-machine theory of infringement

Disney and NBCUniversal’s complaint is built around specific images, not an abstract training argument: screenshot after screenshot of Midjourney generating recognizable characters — Darth Vader, Elsa, the Minions, Shrek, Bart Simpson — on request. They call Midjourney a “virtual vending machine,” arguing the service will produce nearly any of the studios’ characters on demand, which they say proves both that the underlying model was trained on their copyrighted films and that its outputs are themselves infringing copies rather than new transformative works.

“a virtual vending machine”— Complaint, Disney Enterprises, Inc. v. Midjourney, Inc., filed June 11, 2025, C.D. Cal.

Training and outputs, not one or the other

Most AI copyright suits emphasize either training (was the model built on copyrighted works?) or outputs (does what it generates infringe?). This one presses both simultaneously, and revenue is part of the argument: the studios point to Midjourney’s roughly $300 million in annual revenue as evidence the company profits directly from generating their characters, which goes to the commercial-use factor in a fair-use analysis. The complaint also flagged Midjourney’s then-forthcoming video-generation product as an extension of the same alleged problem into a new medium.

Midjourney’s fair-use defense

Midjourney answered on August 6, 2025, denying the allegations and leading with fair use — arguing that training on publicly available images, including the studios’ promotional and released material, transforms the underlying works into a new kind of creative tool rather than substituting for the films and shows themselves. An answer, unlike a motion to dismiss, commits a defendant to litigating the substance rather than seeking early dismissal on the pleadings; Midjourney is defending this one on the merits.

Consolidation with Warner Bros.

On November 4, 2025, the court folded Warner Bros. Entertainment’s separate suit against Midjourney into this action, designating the Disney case as lead. Consolidation lets the studios share discovery and expert work on questions common to both suits — how Midjourney’s training pipeline was built, what data it used, and how outputs are generated — while keeping distinct damages claims tied to each studio’s own catalog.

Where discovery stands

The case has moved through active discovery into 2026, including reported disputes over Midjourney’s requests for material about the studios’ own internal generative-AI projects — a signal Midjourney may be building a defense around the idea that the studios use similar tools themselves. Dispositive motions are reportedly due by late November 2026; no summary judgment or trial ruling has been issued.

Why it matters for training-data licensing

This is the entertainment industry’s biggest litigation bet on generative AI, and it shifts the fair-use fight from artists’ individual styles to instantly recognizable, heavily merchandised characters — the kind of content where courts have historically been least sympathetic to transformative-use defenses. A ruling that character generation defeats fair use would push every image and video model toward output-side filtering as a legal necessity, not just a product feature. For data suppliers, it confirms that studio-owned imagery is close to unlicensable without a direct deal; for buyers, it means guardrails against reproducing protected characters are becoming as important to due diligence as the training data itself.