Midjourney is flipping the script in its copyright battle with Hollywood giants Disney, Warner Bros. Discovery, and Universal Pictures by demanding they disclose their internal AI training data. The AI image generation startup has asked a federal court to force the studios to hand over documents covering datasets, model parameters, and AI development materials. Midjourney’s argument: if these studios are themselves training AI on copyrighted content, the rules of the copyright clash might look very different.
The studios sued Midjourney accusing its AI of reproducing recognizable characters like Batman and Superman, alleging copyright infringement. Midjourney counters that training on publicly available images qualifies as fair use, and now it wants to peek under the hood of how Hollywood’s own AI efforts handle copyrighted material.
Specifically, Midjourney is seeking not just surface-level info but internal AI strategies, research reports, datasets, model tuning details, and even board presentations. So far, the court has limited document disclosure to consumer-facing AI products, but Midjourney is pushing for expanded access.
This defense hinges on consistency: if studios insist that generative AI cannot be trained on protected works without licenses, their own AI experiments-if they use similar methods-cast doubt on their stance. Midjourney argues that access to these withheld documents is necessary for assessing how the industry actually uses AI, which could reshape the legal evaluation of its own practices.
The copyright clash between Midjourney and Hollywood studios
This lawsuit isn’t just about Midjourney and the three studios. Over the past two years, courts in the US and UK have seen a surge of cases testing whether copyrighted texts, images, and videos can be used to train generative AI models without explicit permission. Notable disputes include Getty Images versus Stability AI and The New York Times versus OpenAI and Microsoft. The judiciary has yet to establish a unified standard for these emerging conflicts.
Meanwhile, Hollywood is no stranger to AI, viewing it as both a threat and a tool. Studios have experimented with AI for storyboarding, localization, marketing content, dubbing, and archival searches. Midjourney’s document request strikes a vulnerable chord: if the major rights holders themselves use AI models trained on contested material, it undermines their tough stance against Midjourney and other AI creators.
For Midjourney, this case is a precedent. Should the court agree to broaden disclosure, defendants in similar lawsuits can more often demand media companies and publishers reveal their AI projects-moving the spotlight from just model developers to the content owners’ own AI activities. This shifts the debate from ”Did Midjourney copy Batman?” to how deeply courts can probe the corporate AI labs of industry giants.
The financial stakes help explain the intensity. Bloomberg Intelligence forecasts the generative AI market could surpass $1.3 trillion by 2032. For studios, it’s about protecting franchises and licensing control; for AI companies, it’s about preserving the ability to train models on vast data troves without needing separate deals for each asset. Every court ruling on discovery and data disclosure will be closely watched by both sides of the industry.
The immediate battle now centers on discovery-the exchange of evidence. If the court broadens disclosure requirements, the case will transcend questions about specific characters and focus on how studios themselves develop AI. This could set one of the first legal benchmarks for what internal information about AI models and datasets must be shared in copyright disputes worldwide.