Why A24 Took Google's Money (And What It Actually Means for Filmmakers)

The first thing worth knowing about the A24-Google deal is what Google is not getting.

No footage. No screenplay archive. No access to the library that includes "Everything Everywhere All at Once," "Midsommar," or "The Whale." When Lionsgate signed with Runway, the studio handed over its film catalog as training data. When Netflix acquired AI startup Elephant in 2024 for roughly $600 million, it folded the technology directly into its infrastructure. A24's arrangement with Google DeepMind is structured differently from both of those deals, and that structural difference is exactly what makes it worth paying attention to.

Google is putting $75 million into A24. What Google gets for that is not a content library. What Google gets is a seat at the table inside active A24 productions, watching how working filmmakers actually encounter tools under real production conditions. That is a fundamentally different value exchange, and understanding why it's different tells you a lot about how AI tools for creative work are, or should be, built.

What the Other Deals Actually Did

To understand why the A24 structure is notable, it helps to be clear about what the comparison deals actually involved.

The Lionsgate-Runway partnership was essentially a training data transaction. Lionsgate provided Runway with access to its film and television library so Runway could train generative video models on it. The studio got a stake in Runway and access to the resulting tools. The artistic community's concern was straightforward: filmmakers' work, their visual language, their performances, their cinematographic choices were being used as raw material to train a model that would then be sold commercially, without meaningful consultation with the people who made that work.

Netflix's acquisition of Elephant was a vertical integration move. Netflix decided it wanted AI capabilities in-house rather than licensing them, so it bought the company. That approach gives Netflix control but is essentially a corporate infrastructure decision, not a collaboration with working artists.

A24's deal structure is different in a specific way: the explicit exclusion of content data. According to reporting from Variety, the partnership terms keep A24's library out of Google's training pipeline entirely. What Google's DeepMind researchers gain is observational access to productions in progress. They get to see where friction actually exists in the workflow, what problems filmmakers are trying to solve, what workarounds crews have developed on their own, where time gets lost that doesn't need to get lost.

That is genuinely valuable to a tools team. It is also a very different kind of valuable than a terabyte of labeled video footage.

The Scott Belsky Framing

Scott Belsky is running A24 Labs, the internal unit that will develop these tools. His background is relevant context: he founded Behance, sold it to Adobe, and spent years at Adobe working on creative software before joining A24. He has thought seriously about how tools interact with creative process, and his framing of what A24 is building is worth quoting directly.

Belsky described the tools as something that "won't look anything like the prompted generation type of AI that people feel uncomfortable with." That's a pointed statement. He's explicitly distancing the A24 approach from the paradigm that has created the most friction between the AI industry and the creative community, the text-to-image or text-to-video models where you describe what you want and the system synthesizes something from its training data.

The first specific tool coming out of A24 Labs is an AI storyboard generator. Not a video generator. Not a script analysis tool. Not an AI-generated performance system. A storyboard generator.

Storyboards are pre-production. They're how directors communicate visual intent to the rest of the crew before a single frame of production footage exists. A good storyboard artist is expensive, hard to schedule, and often a bottleneck. The process is also primarily mechanical in a specific way: translating what a director has already conceived in their head into a visual sequence that others can work from. The creative decision, what happens in each shot, how the camera moves, what the emotional register is, has already been made by the time the storyboard artist starts drawing. What the storyboard artist does is give that decision a physical form that can be shared and refined.

That is exactly the kind of step where AI can reduce friction without touching the creative decision itself. If a director can rough out a sequence in an hour using a tool rather than waiting two weeks for a storyboard artist's availability, the direction of the film doesn't change. The scaffolding around the decision gets faster.

Why This Structure Produces Better Tools

There is a practical reason why the "embed researchers in active productions" model produces better tools than the "train on existing content" model, and it has nothing to do with optics or PR positioning.

Training on existing content teaches a model what finished work looks like. It captures the output of creative decisions, but not the decisions themselves, and certainly not the problems that arose during production that shaped those decisions. A model trained on A24's library might learn something about A24's visual aesthetic, but it would have no idea why a particular scene was shot the way it was, what went wrong in the first take, what constraint forced a creative choice that ended up defining the film.

Production observation is different. Watching a world-class production designer figure out how to communicate a visual concept to a director in thirty seconds, under time pressure, tells you something real about where the workflow breaks down and what a useful tool would actually do. That information doesn't exist in the finished film.

The tools that come out of embedded production research should, in theory, solve problems that actual filmmakers actually have, rather than problems that seem like they should exist when you look at the industry from the outside. That's not guaranteed, but the structural conditions for it are better.

For comparison, think about the difference between a video editing tool built by developers who interviewed editors and a tool built by a team that sat in editing suites for six months. The latter team knows that editors don't need fifty more transitions. They know that the real friction is in the organizational layer, managing proxies, finding the right take across forty hours of footage, keeping project files from corrupting on deadline day. Those are not glamorous problems. They don't make for impressive demos. But they are the problems that eat professional time.

The Kane Parsons Tension

A24 built its brand specifically on the idea of filmmaker autonomy. The studio's founding principle, which it has maintained unusually consistently for a mid-sized studio, is that directors get to make their films. The business model is structured around that. You go to A24 because you don't want a committee rewriting your third act.

Kane Parsons made "Backroads" (titled "Backrooms" in some coverage), a film that became A24's biggest theatrical release in the studio's history. He has said publicly, and without apparent ambiguity, that he would erase generative AI from existence if he could.

A24 now has DeepMind researchers embedded in its active productions.

Whether those two things coexist without friction over a multi-year partnership is genuinely unknown. The honest answer is that no one knows yet, including A24. What we can say is that the deal as structured at least attempts to draw a line: the tools are being designed to assist workflow, not to generate the work itself, and content data is explicitly off the table. Whether that line holds as the partnership evolves, as the tools become more capable, as Google's investment thesis requires a return, is a real question.

What the Parsons situation illustrates is that "AI in the workflow" is not a position that resolves artist concerns, it just moves them to a different place. An artist who objects to generative AI on principle isn't going to feel better about a storyboard generator because it doesn't touch script or performance. The line between "scaffolding tool" and "creative tool" is not as clear as tools developers tend to suggest, and the people making the tools should sit with that honestly rather than papering over it with reassuring framing.

That said, for creators who are not categorically opposed to AI in their workflow, the distinction between these categories of tools matters practically. A tool that speeds up pre-production logistics is not the same thing as a tool that generates visual content from a text prompt. Both are "AI tools." They raise different questions and serve different functions.

What This Means for Creators Evaluating AI Tools

The A24-Google deal is, in concrete terms, a big-budget version of a question that individual creators face all the time: which AI tools are worth building into your workflow, and how do you evaluate them?

The structural indicators from this deal are worth carrying into that evaluation.

First, look at where the tool sits in your process. The storyboard generator is pre-production, meaning it operates before the creative work that defines the project exists on camera. Tools that live in that layer, project organization, shot planning, rough sequence visualization, can remove friction without touching the final output. Tools that generate the actual artifact you're selling as your creative work raise different questions about attribution, consistency, and what you're actually contributing.

Second, pay attention to what data built the tool. The distinction A24 drew, no training on their content library, reflects a real concern about what models internalize and who benefits from that internalization. If a tool was built by training on other people's work without those people's knowledge or compensation, that's worth knowing when you decide whether to use it. This is not a legal argument, it's a practical one: tools built on that model are also built on fragile ground, both legally and in terms of creator community trust.

Third, notice whether the tool was built with working practitioners involved or built and then handed to them. The embedded-production approach A24 and Google are using is not unique to large studios. It's the same principle that makes some smaller AI editing tools feel purpose-built for editors while others feel like they were designed by someone who watched a few YouTube tutorials about editing. Products built in proximity to actual workflows tend to solve actual workflow problems.

The Bigger Picture

The A24 deal will not resolve the broader tension between AI development and creative communities. It's one partnership, with one studio, covering one specific slice of pre-production tooling. The industry conversations about training data rights, about AI-generated performances, about what "human authorship" means in a legal and credit context, those conversations are ongoing and will likely run for years.

What the A24 deal represents, structurally, is a different set of assumptions about how to build tools for creative work. The assumption that the most valuable thing an AI lab can get from a creative company is not its content archive but its production knowledge, the assumption that useful tools for artists look like accelerants on manual scaffolding rather than replacements for creative judgment, the assumption that artists involved in tool development from the start produce better tools than artists brought in for post-hoc endorsement.

Those assumptions may or may not prove correct over the life of this partnership. But they are the right assumptions to start from, and the fact that a studio with A24's track record on filmmaker autonomy chose to structure the deal this way, rather than taking the easier path of a licensing arrangement, suggests someone on both sides thought carefully about what they were actually building.

For creators watching from outside the studio system, the question worth sitting with is not whether A24 made the right call. It's whether the tools that come out of this process look like tools built by people who understand your workflow, or tools built by people who understand your workflow as a data problem. Those are different products, and you can usually tell which one you're holding within the first week of using it.