On June 22, 2026, Google DeepMind announced a research partnership with A24 that puts engineers inside the production process of the studio behind "Everything Everywhere All at Once" and "The Brutalist." The deal comes with a $75 million investment from Google, reportedly in the same range Thrive Capital put in during A24's last funding round, and it marks DeepMind's first equity stake in a film studio. The premise floating around before the announcement — that this was a wholesale AI-video deal — undersells what's actually on the table, and overselling it is exactly the wrong read for anyone trying to understand where this goes.
What a prestige studio actually asked for
A24 didn't option its back catalog and it didn't sell training data. The agreement is explicitly non-exclusive and multiyear, and Google does not get access to A24's existing film library. What A24 gets is access to DeepMind's research infrastructure, including Veo, the video-generation model, with DeepMind researchers embedded alongside working filmmakers to build tools those filmmakers say they'd use. The first product off that line isn't a text-to-video engine or a synthetic-actor pipeline — it's an AI storyboard generator, the comic-book-panel sketch a scene gets before a camera rolls. That's a telling choice. Storyboarding is pre-production scaffolding, not the shot that ends up in the theater, and it's a place where a prestige studio can absorb the tool without touching anything an audience sees or a guild contract governs.
A24 hired Scott Belsky, the Behance founder and former Adobe chief product officer, back in January 2025 specifically to run technology and innovation. That hire predates this deal by a year and a half. It tells you the studio was shopping for infrastructure on its own terms rather than getting cornered into a platform partnership. I'd argue that sequencing — hire the product person first, then negotiate from a position of already knowing what you want — is the difference between a studio absorbing a tool and a studio getting absorbed by one.
What Veo can and can't do for a real film, right now
Veo 3.1 generates 4K clips from text or image prompts on a latent diffusion architecture, produces synchronized native audio, and holds a reference character consistent across a shot. That's a genuine jump from where generative video sat two years ago. But the model caps out at eight-second clips. A24's films run ninety minutes to two hours of continuous narrative logic — blocking, continuity, a director's eyeline match from one setup to the next. Closing the gap between an eight-second generation and a feature-length shoot is, per DeepMind's own framing, one of the open research problems this partnership exists to work on. It is not solved. Anyone describing this deal as "AI making movies now" is describing a future research goal as a shipped capability.
The central technology is Veo, DeepMind's video generation model — but the first application is an AI storyboard generator, not a text-to-video engine and not a synthetic actor pipeline.
What generative video is actually good for at this stage is pre-visualization: storyboards, animatics, mood boards, a director showing a cinematographer what a lighting setup might look like before the crew is on the clock. That's real production value — it compresses the expensive, iterative front end of filmmaking, the part that used to run through pricier concept artists and previs houses. It is not a replacement for the eight-second clip becoming the ninety-minute film.
The labor context nobody gets to skip
This partnership arrives the same month SAG-AFTRA ratified a new TV/Theatrical Agreement, in June 2026, that requires producers to show "significant additional value" before an AI performer replaces a live actor or that actor's own digital avatar — and the union can't strike over AI again until 2030. The WGA's deal, ratified in April, is weaker: studios agreed to notify the guild if they license writers' scripts for AI training, but there's no requirement to pay writers when that training happens. Meanwhile Los Angeles County has lost more than a quarter of its film and TV jobs since 2022, with production employment down over 35% between October 2022 and February 2026. Disney, the MPA, SAG-AFTRA, and Paramount have all threatened legal action against ByteDance's Seedance 2.0 over unauthorized use of performers' likenesses. That's the room this deal was announced into.
- SAG-AFTRA: "significant additional value" bar for AI performers, no AI strike authority until 2030
- WGA: notify-only on AI training use of scripts, no compensation requirement
- LA County film/TV production jobs: down over 35%, Oct 2022–Feb 2026
Craft-enhancing, not taste-diluting — with a condition
I'd take the position that this specific deal is craft-enhancing, not taste-diluting, because A24 scoped it correctly: a pre-production tool, no content-library access, no exclusivity, a technologist running point instead of a studio executive chasing a headline. That's what a prestige indie studio should want from a lab partnership — leverage on the boring, expensive parts of making a film, with creative control staying exactly where it was.
But the industry-wide trend this deal gets folded into is a different story, and it's the one worth watching. Every one of these announcements arrives while guild protections lag the technology and while production jobs in the traditional filmmaking hub keep shrinking. A24's storyboard tool doesn't dilute anyone's taste. The pattern of studios treating "research partnership" as cover to normalize generative tools faster than labor agreements can track them — that's the part that should make anyone paying attention nervous, regardless of how carefully any single deal is worded.
