Disney, Warner Bros., and other entertainment giants are shifting the focus of the copyright war against generative AI by engaging in lawsuits on audiovisual outputs rather than training data, altering both the obstacles and pathways to victory.

The studios’ assaults on AI content companies Midjourney Inc. and Minimax allege copyright infringement through mass production of image and video by users. This targets a different activity than most previous copyright suits against AI developers, which have focused on the use of protected works to train large language models.

This strategy raises new legal questions compared to text-based AI cases, especially around fair use, platform liability, and whether AI companies can claim protection under the DMCA safe harbor. Attorneys noted that unlike YouTube, which merely hosts user content, Midjourney itself generates outputs. As Robert Rosenberg of Moses & Singer LLP put it:

“YouTube didn’t create the work. Midjourney created the work. Midjourney is selling subscriptions to the infringing machine. They are selling you access to the capability to create all these infringing works. If they hadn’t hoovered all of this intellectual property in the first place, they may be able to show a mouse that doesn’t infringe, but the output is Mickey Mouse.”

By emphasizing infringing outputs, the studios may find it harder to secure large-scale damages tied to training data but could strengthen claims that these AI tools are marketed in ways more like Grokster’s unlawful file-sharing service than Sony’s lawful VCR.