Ben Affleck’s $587M AI sale and why he thinks movies are safe from AI

Ben Affleck isn’t just starring in movies—he’s been quietly building AI for them. After selling his AI company to Netflix for a reported $587 million, he’s offering a rare insider view on how generative video actually works, why he built his own dataset from scratch, and why he’s convinced AI will change filmmaking tools but not replace filmmakers.

From VFX tinkerer to AI founder

Affleck explains that his path into AI started years ago with a genuine interest in computers and the shift from analog film to digital workflows. While working with visual effects teams, he was exposed early to machine learning tools used for tasks like cleaning up green screens and tracking edges in footage.

He picked up enough Python to write simple scripts and learned the basics of how neural networks process images. That hands-on exposure with GPUs and VFX pipelines eventually led him to recognize how powerful the new wave of transformer-based models could be for video.

How AI sees video: tensors, edges, and patterns

To make sense of what these models do, Affleck breaks down how AI “sees” images and video. A frame of video can be translated into numbers: frame index, batch number, and the red, green, and blue value of every pixel. That numeric structure is called a tensor.

Convolutional neural networks (CNNs), which were the precursors to today’s transformer models, scan those tensors to find patterns. They can perform tasks like edge detection and feature extraction—figuring out where a window ledge is, where a face begins, or where the background ends. That’s how visual effects teams can more easily remove a green screen and replace it with a digital environment.

Transformers take this further, running far more computation in parallel and enabling today’s generative models that can create realistic images and video from scratch.

The moment he realized how powerful generative video had become

Affleck describes visiting top AI labs after being tipped off by VFX colleagues that Google and others were doing “really interesting stuff” with transformers in video. Thanks to his Hollywood name, he could simply call and ask to see what they were building—and they said yes.

What he saw shocked him: with just a few keystrokes, these systems could generate very realistic-looking video. For a day or two, he admits he panicked, wondering if they needed to make as many movies as possible before the technology made traditional filmmaking obsolete.

Why early AI video models disappointed him

That initial fear didn’t last long. Once he sat down with engineering teams and dug into how the models worked and how the data was labeled, he realized their limitations.

He points out several key problems:

• Lack of control: The models could generate striking visuals, but you couldn’t direct them with the precision filmmakers need. Ask to “move the camera,” and instead of simulating a camera move, the model might just draw a camera into the image.

• Inconsistent results: Outputs weren’t reliable or repeatable enough for professional production.

• Data issues: Many models were trained on scraped internet video (like H.265 clips) with loose or noisy captions. That’s not how real film production data is structured.

Because he understood both filmmaking and the basics of machine learning, he realized there was a gap: brilliant AI researchers were building powerful models, but they didn’t understand the “domain expertise” of film—how shots are actually captured, organized, and described on a real set.

The hidden advantage of real film data

Affleck argues that native filmmaking comes with a built-in, highly structured dataset. Every shot has metadata: camera position, lens choice, timecode, scene numbers, takes, and more. Unlike scraped web video, this data is ordered and meaningful from day one.

That means you don’t always need to manually caption everything. The raw format of professional production already encodes a map of how images are created in the real world—if you know how to use it. This is where his Hollywood background became a technical advantage.

Why he refused to let big AI companies own the value

Once he started talking to major AI players, Affleck says they made offers along the lines of: “Work with us, show us what you’re talking about, and we’ll do the AI for you.”

He turned that down for two reasons:

• They would capture all the value: If he just handed over domain knowledge, the big AI company would own the core technology and economics.

• They didn’t care about rights and likeness: These weren’t Hollywood-native companies. He worried they wouldn’t prioritize issues like performers’ rights, likeness, and consent.

He didn’t want a world where powerful generative video tools were built far outside the film industry and then pushed back onto filmmakers without their input or control.

Building his own dataset from scratch

Instead of selling out early, Affleck decided to build his own company and dataset. He used his access to capital to:

• Set up extensive multi-camera shoots over months

• Use LIDAR and location-specific devices to precisely map environments

• Incorporate Unreal Engine and LED volume stages into the capture process

• File patents around how this data was captured and structured

The goal was to create a robust, clean dataset that could train models without relying on scraping other filmmakers’ work. He didn’t want to have to tell a peer, “By the way, part of the intelligence in this model came from your movie.”

In his view, you don’t have to train on copyrighted work—it just takes more effort and intention to build your own data from the ground up.

AI as a tool for filmmakers, not a replacement

Affleck is blunt about his company’s mission: capture this technology for filmmakers and artists, not against them.

He wanted AI tools that:

• Are controlled by creators and studios

• Respect rights, likeness, and ownership

• Help people make the movies they want to make, faster and cheaper

• Don’t pretend to “replace” directors, writers, or actors

He’s adamant that the idea AI will simply let you type out “make my movie” and instantly get a finished, studio-quality film is fantasy. In his view, that’s not how the technology works, and it’s not where it’s headed.

Why he believes AI won’t replace movies or filmmakers

Affleck calls feature filmmaking “the last thing that gets disintermediated by AI.” He gives several reasons:

• Compute is brutally expensive: Generating high-quality video at scale requires far more compute than text. Training and inference for large video models are enormously costly, even compared to today’s biggest language models.

• Story is not just pattern-matching: While models can mimic style or generate a shot, they don’t have lived experience, taste, or intent. The human choices behind a story are still central.

• Production is more than pixels: Movies involve collaboration, performances, direction, editing, and countless creative decisions. AI can assist, but it doesn’t replace that ecosystem.

His prediction lines up with a broader trend in AI commentary: a shift away from early “job apocalypse” rhetoric toward a more nuanced view of augmentation over replacement. For a related perspective from another industry leader, see how Sam Altman has softened his own warnings about an AI-driven job apocalypse.

Why creatives are scared—and why that fear is rational

Affleck is sympathetic to the fear in the creative community. If someone tells you, “This is going to replace you and your livelihood,” of course you’re going to be angry and terrified.

He points out that much of the public narrative around AI has been driven by people with strong incentives—either to hype their product or to gain power by framing themselves as gatekeepers of an existential threat. Saying “this will wipe everything out” grabs attention, drives clicks, and can justify more control or funding.

He also notes that humans are wired to focus on threats. Evolution tuned us to remember danger more vividly than safety; it’s why we’re drawn to doom headlines and why traumatic memories stick so hard.

What he sees as the real AI risks

While he rejects sci-fi scenarios like “Skynet” taking over the world, Affleck does see serious risks—especially when AI is given control over systems with life-or-death consequences.

He draws a clear line between traditional computing and AI:

• Deterministic computing: Like a calculator, it gives the same correct answer every time.

• Generative/stochastic models: They don’t produce exact, repeatable answers. They generate likely outputs by “aiming” based on patterns in data.

Because of this, he warns against handing full control of critical systems to AI:

• Autonomous targeting in warfare

• Fully automated security systems that can lock, detain, or harm people

In these contexts, mistakes are inevitable. The technology is inherently probabilistic, not perfectly reliable.

Why accountability still matters more than sci-fi fear

Affleck argues for a measured, responsible approach to AI—grounded less in abstract existential fear and more in concrete accountability.

One of the strongest tools society already has is liability. If a company deploys AI in a way that harms people, it can be held responsible. That legal and financial pressure, he believes, is a powerful counterweight to the rush to move fast and break things.

Rather than letting fear of “are we doomed?” dominate the conversation, he suggests focusing on:

• Clear rules around where AI can and can’t be trusted

• Strong protections for rights, likeness, and consent

• Tools that empower workers and creators instead of displacing them

We’re already seeing similar debates across industries, from software to media. For another example of how leaders are quietly reshaping their businesses around AI while keeping humans in the loop, see this story of a CEO who rebuilt his company with AI.

Affleck’s big prediction: AI will reshape tools, not erase storytellers

Underneath the technical talk and the Hollywood details, Affleck’s core prediction is simple: AI will dramatically change the tools of filmmaking, but not the existence of filmmakers.

He sees a future where:

• AI speeds up visual effects and post-production

• Directors can iterate on shots and scenes far more quickly

• Smaller teams can achieve visuals that once required massive budgets

• Creators retain control over their style, likeness, and stories

What he doesn’t see is a world where you press a button and a fully formed, emotionally resonant, culturally meaningful film just appears. That, in his view, is not how either technology or storytelling works.

For now, his bet has paid off: by building a filmmaker-first AI company with its own ethically grounded dataset, he not only landed a major exit to Netflix but also helped nudge the industry toward a version of AI that works with creators instead of against them.