Midjourney forces discovery on studio AI use in copyright defense

Midjourney has moved to compel discovery in its ongoing copyright litigation with three Hollywood studios, requiring the plaintiffs to disclose the scope and details of their own AI usage. The filing represents a tactical shift in generative-AI copyright defense: rather than litigate fair use in isolation, Midjourney is making the studios' internal AI practices a material fact in the case.

The studios sued Midjourney over alleged training-data infringement, claiming the service generates images derived from copyrighted film stills and promotional materials without permission. Midjourney's response now hinges on demonstrating that the plaintiffs themselves deploy similar technologies in production pipelines, visual-effects workflows, and marketing operations. If the studios cannot draw a clean line between their own AI adoption and Midjourney's, the infringement claim weakens on both statutory and equitable grounds.

This is not a tu quoque argument. It is a request for evidence that the industry has already determined AI-assisted generation falls within the scope of transformative use. Studios routinely license large image corpora for training proprietary models used in pre-visualization, texture synthesis, and background generation. If those practices survive internal legal review and vendor contracts, Midjourney argues, they constitute commercial acknowledgment that training on copyrighted works does not require per-image licensing.

The discovery motion also forces studios to quantify their AI spend and disclose vendor relationships, potentially surfacing partnerships with Runway, Stability AI, or in-house research teams. Any evidence that the plaintiffs trained models on third-party datasets—licensed or otherwise—will complicate their standing. Courts have shown increasing willingness to dismiss copyright claims when plaintiffs engage in functionally identical conduct, particularly in technology disputes where industry practice defines the contours of fair use.

The case is narrow but the precedent is broad. If Midjourney prevails in discovery, every generative-AI defendant will replicate the tactic. Studios, publishers, and record labels that sue over training data will face mandatory disclosure of their own AI pipelines, turning IP litigation into a mutual audit. The result is not clarity but symmetry: if everyone is infringing, no one has clean hands, and the statutory framework collapses into a negotiation over licensing terms rather than a binary question of legality.