For years, the relationship between Hollywood and technology platforms was framed as a conflict: catalogues, characters, scripts and performances on one side, models trained on enormous quantities of material whose origin was often impossible to reconstruct on the other. Google’s approaches to Disney, Universal, Warner Bros. Discovery and other studios suggest a possible new phase. There is no broad agreement yet, but the question is no longer only how to prevent unauthorized use. It is how much an archive might be worth when it becomes operational material for artificial intelligence.
A film library is not simply a collection of files. It contains characters, costumes, sets, voices, movement, narrative worlds and an enormous quantity of metadata. To build tools that can function inside production, a model must preserve continuity across shots, distinguish versions of the same character and understand the rules of a world. Studios own precisely what indiscriminately collected web datasets lack: organized, dynamic material linked to identifiable rights.
Licensing could turn conflict into a market. An authorized character might be used in promotional content, interactive experiences, translation, games or new serial forms. But granting a name and appearance is not the same as surrendering every possibility. A contract would need to specify who can generate, which actions are prohibited, how outputs are approved, how long they remain available and whether new images can be used to train the system again.
People are the most delicate issue. A catalogue belongs to a studio, but a face, voice or performance may involve performer rights, union agreements and estates. A character is often inseparable from the work of the person who embodied it. The technical ability to recreate that character does not eliminate consent. It makes it necessary to distinguish among ownership of the film, personal identity and a new synthetic performance. Without that separation, archives risk converting past labour into permanent availability.
Authorship also needs protection. A model trained on a studio’s films can learn aesthetic patterns, but style is not identical to a trademark. Directors, writers, production designers, costume designers, editors and cinematographers created what the system recognizes. A licence negotiated only between platform and studio might be legally possible in some settings yet culturally incomplete. The archive’s value comes from many kinds of work that an agreement should make visible.
For Google, a deal would provide quality data and cultural legitimacy. For studios, it could open revenue and production tools while creating a competitor capable of generating unlimited variations of their assets. It resembles the earlier dilemma of YouTube: attempt to control circulation, or build a system that identifies and monetizes it. This time, however, the platform does not merely distribute copies. It can produce new images from rules learned from earlier ones.
GENERATIVE ARCHIVES names a profound transformation. Preserving a film may no longer mean only restoring it and making it available. It may mean deciding which elements of its world can act again, under what conditions and for whose benefit. If Hollywood truly opens its archives to AI, the decisive question will not be how many new scenes can be created, but whether every scene retains a legible trace of the rights, people and memory from which it came.