Disney and Universal sue Midjourney over copyright

Disney and Universal have filed a lawsuit against Midjourney in a federal court in California, alleging copyright infringement. The studios say the image generator allows users to create copies of protected characters and franchises — including Darth Vader, Shrek, the Minions and Homer Simpson — and that its operation relies on works used without permission.

The case matters because it brings a central question for generative AI into the courtroom: whether a company can train and commercialize a model capable of reproducing recognizable intellectual property without licensing the source material. Until now, the most prominent lawsuits against AI companies had focused on text, photographs, music or code. Hollywood is turning up the pressure on a product that transforms a written instruction into images ready to share or use commercially.

The accusation: recognizable characters at the prompt of a command

The lawsuit, filed on Wednesday, alleges that Midjourney generates images of protected characters when users request them by name. The plaintiffs say these are not merely results vaguely inspired by a particular aesthetic; they allege reproductions of figures, costumes and visual universes associated with specific films and television series.

The action brings together Disney and Universal entities tied to franchises including Star Wars and Marvel, as well as The Simpsons, Shrek and Despicable Me. That combination broadens the lawsuit’s symbolic reach and brings together some of the most recognizable characters in both groups’ catalogs.

The legal dispute has two core parts. On one side, the studios challenge the use of their images to train the system. On the other, they allege that Midjourney facilitates the creation and distribution of outputs that infringe their rights. The second issue could be especially significant: while the debate over training remains unsettled and depends on how judges interpret the fair use of copyrighted works, outputs that directly resemble specific characters are easier to put before a court.

A subscription-based model facing growing pressure

Midjourney is one of the most popular image-generation services. It rose to prominence through Discord and offers paid plans for creating images from text. Its appeal has been precisely its visual quality and the ease with which it turns familiar cultural references into illustrations, posters or new scenes.

That same ease has become a problem for companies that control large catalogs of intellectual property. A prompt combining a character’s name with an action, style or setting can produce content that retains identifiable traits of the original work. For a fan, it may be a way to experiment. For a studio, it opens the door to unauthorized commercial uses, confusion over an image’s origin and a loss of control over brands built over decades.

The plaintiffs are seeking damages, the amount of which will have to be determined in the proceedings, as well as court orders to prevent further infringements. They are also asking for a jury trial. The lawsuit does not mean Midjourney has been found liable: the court will examine what data the company used, how the model was trained and the extent to which its service contributes to the alleged reproductions.

The line between inspiration and copying becomes commercial

Generative image models typically produce new pixels from statistical patterns learned from large datasets, rather than operating like a search engine that retrieves a specific image from a conventional library. But that technical explanation does not by itself resolve the legal conflict or establish which works a system may have used or memorized. If an output retains protected elements that make it unmistakably Darth Vader or Shrek, the debate is no longer abstract.

The studios are not merely challenging users’ creativity. They are questioning the design of a commercial service that, according to their complaint, may have offered protected characters without effective filters or licensing agreements. That detail distinguishes this case from a theoretical debate over whether AI can draw inspiration from existing works: here, the accusation targets the practical availability of recognizable copies within a paid platform.

The litigation comes as other AI companies try to negotiate licenses with publishers, media organizations, record labels and visual archive owners. For Midjourney and its competitors, an adverse ruling could raise the cost of training models and force them to strengthen barriers blocking requests involving well-known franchises. For Hollywood, the goal is to preserve its ability to decide who can exploit its characters and under what conditions.

The first battle will not by itself decide the future of generative AI, but it could set an important precedent on two very specific questions: what responsibility a platform bears for the images it produces and when an AI-generated output crosses the line between a cultural reference and copyright infringement.