Midjourney wants Hollywood studios to reveal the details of their AI usage

Midjourney Seeks to Force Hollywood Studios to Disclose Their Own AI Training Practices

In a move that could reshape the legal landscape for generative AI, Midjourney is asking a court to compel three major Hollywood studios to disclose how they use artificial intelligence internally. The request, filed in the ongoing copyright infringement lawsuit brought by Disney, Universal, and Warner Bros., goes beyond standard discovery demands: Midjourney wants the studios to reveal whether they are doing exactly what they accuse the startup of doing—training AI models on unlicensed copyrighted content.

For developers and builders working with generative models, this dispute is more than a Hollywood legal spat. The outcome could define the boundaries of fair use for training data in the United States, set precedents for what constitutes acceptable internal AI use, and influence how every AI company approaches data sourcing. The case has already produced a key ruling: a judge limited discovery to consumer-facing AI-generated images and videos. Midjourney now seeks to overturn that restriction, arguing it unfairly allows the studios to cherry-pick evidence.

What the Filing Actually Seeks

Midjourney’s latest motion asks the court to require the studios to produce all documentation related to their internal generative AI activities, not just those that resulted in publicly released content. Specifically, the startup wants to see:

• Details of any image-generating AI models the studios have developed for internal use (e.g., storyboarding, ideation)
• The full set of prompts the studios used in Midjourney itself, along with all outputs—not just the prompts that allegedly produced infringing images

The legal argument is strategic: if the studios themselves are downloading copyrighted material to train AI models behind closed doors, that fact could demonstrate that such training is an “industry custom.” Under fair use doctrine, industry norms can weigh heavily in the analysis of whether a use is transformative or commercially harmful. In other words, Midjourney is trying to use the studios’ own behavior to build a fair use defense.

The studios’ lead attorney, David Singer, has previously described the request as a “fishing expedition.” He also stated that the studios “do not seek to stop AI technology or even shut down Midjourney’s business,” but rather want Midjourney to stop reproducing their characters without authorization.

Why This Matters to AI Developers

The core question here—whether training on copyrighted content without permission is fair use—is the same one facing every AI company that scrapes the public internet. Midjourney’s approach of asking a judge to scrutinize the plaintiffs’ own AI practices is a novel defense tactic. If successful, it could create a powerful precedent: companies that sue for copyright infringement while engaging in similar training internally might find their own operations exposed and their claims weakened.

For developers, the practical takeaway is that transparency about training data may become a legal necessity. If the industry norm becomes “everyone does it, including the rights holders,” that could lower the risk for using broad scraped datasets. But conversely, if the judge denies Midjourney’s request and upholds the narrow discovery scope, it would signal that internal AI use is irrelevant to fair use analysis—meaning developers cannot rely on studios’ behavior as a shield.

There’s also a direct implication for how developers use AI tools. The studios have been using Midjourney themselves, and Midjourney wants full logs of those prompts and outputs. This raises the question: when you use a generative AI service, should you assume your prompts could become evidence in litigation? Enterprises using such tools should consider that query logs might later be subject to discovery, especially if they are involved in any dispute with the tool provider or with rights holders.

Industry Custom vs. Fair Use Doctrine

Fair use hinges on four factors: purpose and character of use, nature of the copyrighted work, amount used, and effect on the market. Midjourney is arguing that if the studios themselves are training models on unlicensed content for internal creative processes, that shows the third and fourth factors—amount and market effect—are not as harmful as the studios claim. If internal storyboarding with AI is itself a form of copying, then the studios are simultaneously accusing Midjourney of doing something they themselves do.

This line of reasoning is not entirely novel—defendants in copyright cases have often tried to point out plaintiff hypocrisy—but it takes on new weight in AI because training fundamentally involves copying. If a judge accepts that internal use of generative AI for ideation is an accepted practice even among the most aggressive copyright enforcers, then training on copyrighted data for non-public purposes could be viewed as less egregious.

The studios, of course, will argue that internal use is different: it’s for their own creative process, not for distributing derivatives to the public. Midjourney will counter that training a model inherently creates copies that are then used to generate new content—and that the studios’ internal models do the same thing. The judge’s decision on discovery will essentially determine whether this argument gets a fair hearing.

Broader Implications for the AI Tooling Landscape

This case is unfolding alongside many similar lawsuits against AI companies by creators, publishers, and rights holders. The outcome could influence settlement trends. If discovery reveals that studios rely on similar AI tools internally, they may be less motivated to pursue aggressive litigation that could backfire. That, in turn, could reduce legal uncertainty for developers building on platforms like Midjourney.

If you are evaluating which image generation model to adopt for your projects, legal risk is now a factor alongside output quality. The best tools for the job may shift depending on how courts rule on fair use. For a current overview of leading options and their legal postures, see our comparison of the best AI image generators for 2026.

Similarly, the pricing and accessibility of AI models are affected by these legal battles. Companies may raise prices or restrict features to limit liability. Developers should keep a close eye on how API costs and usage terms evolve as these cases progress. Our LLM API pricing reference provides a useful benchmark, though Midjourney’s case focuses on image generation, the same principles apply to text models when training data is contested.

What to Watch Next

The court is expected to rule on Midjourney’s motion to expand discovery in the coming months. If the judge grants the request, the studios will be forced to produce internal documents about their AI usage. That could lead to a wave of revelations about how Hollywood really deploys generative AI. It could also accelerate settlement talks if both sides prefer to avoid full public disclosure.

For now, developers can only monitor the case and prepare for either outcome. If you are building products that rely on models trained on web-scraped data, consider implementing more rigorous provenance tracking now. Regardless of how this specific case resolves, the trend is clear: transparency around training data will become a competitive and legal necessity. The days of “train first, ask permission later” are numbered, and this filing is one more sign that the courts are beginning to ask hard questions of everyone involved—including the plaintiffs.

Midjourney’s push for studio transparency may ultimately be remembered as the moment the AI industry forced copyright holders to practice what they preach. Whether that is enough to win in court remains to be seen, but it is a savvy strategic move that every builder should understand.

Source: TechCrunch. Details as reported; verify specifics at the source.