Here Is What "Fair Use" Looks Like When Darth Vader Is the Evidence
In June 2025, Disney, Universal, and DreamWorks filed a 110-page copyright complaint against Midjourney, the AI image-generation company. The complaint alleges that Midjourney trained its model on protected works without permission and that its outputs, the images the system generates in response to user prompts, infringe those copyrights. It includes side-by-side comparisons of Midjourney outputs next to Darth Vader, Homer Simpson, Shrek, and Elsa. The message was pretty clear.
On August 6, 2025, Midjourney filed its answer. The central defense is fair use. Specifically, transformative fair use.
Fair use is a defense under copyright law that allows someone to use a protected work without permission in certain circumstances. Courts evaluate it using four factors, the most important of which is whether the use is "transformative.” Does it add something new rather than simply reproducing the original (sort of). Midjourney's argument is that training an AI model on images is transformative because the model does not store or copy those images. It extracts statistical information about patterns and relationships, then uses that information to generate new images. The CEO has compared it to an art student studying thousands of reference works before painting something original.
Whether that argument holds is genuinely uncertain, and the uncertainty runs in both directions.
Midjourney has some supportive precedent. Courts in two recent cases — Kadrey v. Meta and Bartz v. Anthropic — found that training AI models on copyrighted text could qualify as transformative fair use. Those cases involved text, not images, and the facts differ in ways that will matter to the court. But the doctrinal structure is similar, and Midjourney will cite them.
The harder problem is what happens on the output side. Even if a court accepts that training is transformative, that does not automatically resolve whether a user prompt for "Darth Vader in a forest" that produces an image resembling Darth Vader in a forest infringes Lucasfilm's copyright. The complaint is specifically constructed to push the analysis there. The side-by-side comparisons are not about training data, they are about outputs. That is where Disney's best facts are, and Midjourney's fair use argument is weaker on that theory.
Midjourney also raised an argument that is worth noting separately: Disney uses generative AI. Several Disney productions have used AI-generated imagery. The implication, that the most aggressive plaintiff in this case is also a significant consumer of the same technology, is not going to win the lawsuit, but it is a real observation about where the industry sits right now.
The case is pending in the Central District of California. When the court rules on the output infringement theory at summary judgment, that ruling (and the Ninth Circuit appeal that will no doubt follow) will set the first significant appellate precedent on whether AI-generated images that resemble protected characters infringe those copyrights. That is the ruling the entire AI industry is waiting for, and the entire entertainment industry is waiting for, and they are not waiting for the same answer.