Somewhere in a production office there is a person who owns chain of title: the paperwork asserting that every frame in a deliverable is cleared to be there. That person is the real buyer for a studio-grade video model, and no showreel answers their question. The Los Angeles Times reports that the studio talks have not closed partly because of the legal complexity around AI and sensitivities over how Hollywood talent and unions would react [2]. That is a clearance problem sitting on someone's desk right now, not a research backlog.
The insider quoted in the story makes the dependency explicit: for Google's tools to be usable at the scale of film and TV production or cinematic exhibition, the models have to be fine-tuned on highly dynamic content and metadata that only the studios hold [5]. That is a supply statement, plainly put. The corpus is the feature, and it comes only with a signature attached.
Weigh the sourcing before you weigh the strategy. Google, Warner Bros. Discovery and Disney declined to comment [3], and an NBCUniversal spokesperson said the company is not in active conversations with Google about licensing its intellectual property for AI models [4]. Three declines and one denial leave zero on-record confirmations from the four named companies [12]. The denial is also narrow. "Not in active conversations" sits comfortably alongside an approach that went nowhere, which is close to what the three people familiar with the talks described [1].
In practice, adoption is not gated on output fidelity, with licensing expected to catch up once the shots look good enough. The people actually rolling these tools out use them in the places where nothing has to be cleared, and stop at the boundary where a deliverable starts. Internal usage counts go up, but that growth says little about value on its own. The measure that would settle it is how many licensed seconds reached a released cut, and on the Times account there is no major-studio agreement in place to produce that number [2].
Google's disclosed money went to the shorter clearance chain: $75 million into A24 in June as part of a strategic partnership on AI tools [6], plus a partnership with Darren Aronofsky's venture Primordial Soup [7]. Rob Rosenberg of Moses & Singer, formerly general counsel at Showtime Networks, told the Times that A24 gives Google a credibility it has not yet earned in Hollywood [9]. The only dollar figure in the report attaches to an independent, not to any of the majors named in the licensing talks [13]. Meanwhile the majors are the ones holding libraries worth licensing, the ones who have sued over AI [8], and the ones under pressure to cut costs with the technology while stopping unauthorised use of their IP [10].
So the 2x2 for anyone running a pilot on Monday. Down one axis: does the output leave the building inside a deliverable. Across the other: is every element in the chain covered by a document someone with authority will sign. Deliverable and covered is production. Internal and uncovered is cheap and harmless, and it proves nothing about whether the tool works. Internal and covered means you bought paperwork you did not need. Deliverable and uncovered is where your own legal team halts the release, and it is where most pilots quietly sit. The forcing function is to name the signer before you buy the seats. If nobody in your organisation will put their name on the clearance memo for generated output, what you are running is a demo, a category that already sits outside any licence requirement.
