Google offers Disney millions to train Gemini on characters

Google has approached major entertainment companies, including Disney, Warner Bros. Discovery, and Universal, seeking licensing agreements to train its artificial intelligence models on their copyrighted libraries. To secure these rights, the tech giant is willing to spend hundreds of millions of dollars. For instance, Google could pay a studio like Disney or Pixar approximately $40 million just to generate outputs of a single copyrighted character, such as Darth Vader or Snow White. These payouts could eventually scale into billions of dollars, and studios could even receive a share of advertising revenue generated from AI content featuring their intellectual property on platforms like YouTube.

This aggressive push follows other major entertainment AI deals. Earlier this summer, Google's DeepMind division finalized a $75 million investment agreement with A24. Meanwhile, smaller studios are already experimenting with the technology; Lionsgate partnered with Runway in 2024 to train a model on its catalog for upcoming AI-generated shorts. However, other high-profile collaborations have faltered, such as a planned partnership between Disney and OpenAI that collapsed before producing results.

For AI practitioners and developers, securing high-quality, licensed studio data would represent a massive shift away from controversial models trained on uncompensated web scraping. It would allow tools like Gemini to legally generate highly recognizable, polished media. Yet, Hollywood executives remain cautious. While shareholders are attracted to AI's potential to lower production costs and accelerate timelines, the threat of job losses for creative workers looms large. Furthermore, studios fear public backlash. Audiences recently expressed dismay over AI-generated art in the official artbook for Spider-Man: Brand New Day, a film that nevertheless grossed $2 billion. Studios are wary of being the first to release a major project heavily branded with Google's AI tools.