Midjourney's Defense Against Disney: "Show Us Your Homework Too."

Disney, Universal, and Warner Bros. are suing Midjourney for copyright infringement, arguing its AI image generator can produce copyrighted characters like Bart Simpson, Darth Vader, Superman, and Batman on demand. Midjourney's defense strategy just took an interesting turn: instead of only defending its own AI training practices, it's trying to force the studios to open up about theirs.

Here's the legal chess move. A judge already ruled the studios have to hand over documentation about their own generative AI usage, but only the parts that led to "consumer-facing" content, meaning finished videos or images the public actually sees. Midjourney wants that limitation removed entirely.

Its argument is pointed: that narrow scope lets studios cherry-pick only the documents that support their own case, while conveniently shielding anything that might undercut it. Midjourney wants access to internal use too, specifically tools studios may be using behind the scenes for storyboarding or content ideation, not just what eventually reaches audiences.

The logic underneath this is essentially: if the studios are quietly training or using AI models on unlicensed copyrighted material for internal creative work, that's not just an inconvenient fact, it's potential evidence that downloading and training on unlicensed content is standard industry practice, not the reckless behavior Midjourney is being accused of. Midjourney is also asking for something more specific: every prompt the studios themselves typed into Midjourney, and every image that came out, not just the small subset that happens to look infringing.

The studios' attorney, David Singer, isn't having it. He's calling this a fishing expedition, and his framing of the actual complaint is narrower than headlines suggest: the studios say they're not trying to shut down AI technology or Midjourney's business at all. Their claim is more specific, that Midjourney should stop reproducing their movies and shows and stop generating unauthorized versions of their copyrighted characters.

That distinction matters. This isn't really a referendum on whether AI image generation itself is legal. It's a fight over one narrower question: does training on and outputting copyrighted characters cross a line, regardless of who's doing it or why.

Which is exactly why the discovery fight is so revealing. If it turns out major studios have been using generative AI internally in ways that look uncomfortably similar to what they're suing Midjourney for, that's not a technical footnote. That's the entire moral high ground of the lawsuit up for debate.

Should the standard for "acceptable AI training on copyrighted material" depend on whether the output ever reaches the public, or should it apply the same way regardless of who's using it?