The Invisible Cast

AI is already populating your favorite shows. You just don't know it yet.

There is a moment in the sixth episode of The Eternaut — a post-apocalyptic Argentine science fiction series that arrived on Netflix on April 30, 2025 — where a high-rise building in Buenos Aires collapses in the aftermath of a catastrophic train derailment. It lasts seconds. It is dark, deliberate, and photorealistic enough that almost no one watching noticed anything unusual.

What they didn't know was that the sequence had been generated, not built. No miniature. No on-location demolition. No months of modeling and simulation from a VFX team. Netflix's internal visual effects division, Eyeline Studios, had produced it using generative AI tools — and the whole thing was delivered ten times faster than traditional methods would have allowed.

Netflix co-CEO Ted Sarandos disclosed this on the company's Q2 2025 earnings call, framing it as a milestone: the first time the streamer had used AI-generated visuals in an original scripted production. "The cost of it just wouldn't have been feasible for a show on that budget," he told investors. The Eternaut was produced for approximately $15 million — modest by the standards of English-language prestige television — and the AI sequence made an otherwise unaffordable shot possible.

That framing matters. Not "we replaced our VFX artists with a machine." Rather: "we made something possible that wouldn't have existed otherwise." Whether you find that reassuring or alarming depends on what you think the real question is.

The real question is not whether generative AI belongs in film and television. It is already there. The question is who gets to decide how it is used, under what conditions, and at whose expense.

The background has always been constructed

To understand what is actually changing, it helps to understand how manufactured the background of any given scene has always been.

The soldiers in the climactic battle of Avengers: Endgame — tens of thousands of them — were not human beings in costume. They were generated by Massive, a crowd simulation software developed by Weta Digital, the same tool used to populate the armies of The Lord of the Rings more than two decades earlier. Weta's VFX supervisors spent days in Atlanta working with stunt teams, capturing the specific combat styles of Wakandans, Asgardians, Chitauri, and Ravagers — then fed that motion-capture data into Massive to build a virtual army capable of fighting with authentic, differentiated behavior. The result populated what was effectively an entirely CG environment for the majority of the film's final sequence.

This is not new technology. It is not even controversial. Digital crowd simulation has been standard industry practice for well over twenty years. What is new is the layer that is being added on top of it: generative AI that does not just simulate, but synthesizes. Not armies that behave convincingly, but environments and visual elements that look like they were filmed when they were not.

The distinction is significant. Traditional crowd simulation still required human motion-capture artists, trained animators, and a team of technical directors to manage each individual agent's behavior. The new generation of tools — text-to-video models, AI-assisted compositing, generative environment synthesis — dramatically compresses that pipeline. You describe a result in language, and the system produces something that moves toward it.

The moment the seam showed

The first high-profile moment of friction in this shift did not come from a background crowd or a distant cityscape. It came from a title sequence.

When Secret Invasion premiered on Disney+ on June 21, 2023, something felt wrong to viewers within the first sixty seconds. The opening credits — morphing, green-tinted imagery of Skrull faces, city skylines, and deformed human figures — had a quality that was immediately legible as AI-generated. It was the wrongness that is hard to describe precisely but impossible to unsee: shapes that almost resolve but don't, motion that follows no physical law, faces that suggest identity without committing to it.

Series director Ali Selim confirmed to Polygon that the sequence had been produced by Method Studios using AI. "We would talk to them about ideas and themes and words, and then the computer would go off and do something," he said, adding, somewhat disarmingly, that he didn't "really understand" how it works.

The backlash was immediate and sharp. A former Marvel concept artist who had worked on the production, Jeff Simpson, posted that he was "devastated" — he had spent nearly half a year on the show and found the AI credits a betrayal of the broader creative team. Storyboard artist Jon Lam called it "salt in the wounds" of artists and writers then on strike with the WGA. The timing was brutal: the guild strikes were already underway, and AI's role in the industry's future was the central anxiety underneath both labor disputes.

What made the Secret Invasion controversy particularly revealing was not that AI had been used in a high-profile production. It was that the use was visible, and in being visible, it forced a confrontation that invisible use would never have triggered. A building collapse in Buenos Aires, seamless and dark, slides past. A title sequence that looks like a fever dream does not.

The body scan problem

While Secret Invasion ignited a cultural debate, a quieter and more legally complex version of the same confrontation was playing out on physical sets across Hollywood.

During the 2023 SAG-AFTRA strike, reports surfaced that background actors — extras — were being directed to trailers equipped with hundreds of cameras designed to create full 3D body scans. The digital replicas created from these scans could theoretically be inserted into scenes indefinitely, in productions the actor had never worked on, without additional compensation. In some reported cases, background actors who declined to be scanned were sent home without pay.

The implications were not subtle. A studio that scanned a hundred background actors on a single day of filming had, in principle, acquired a permanent cast of digital people who could populate any crowd, any street, any battlefield, for the life of the production and potentially beyond.

The 2023 contract ratified in December of that year — approved by 78 percent of the union's membership — established new guardrails. Under the agreement, an employer legally owns the materials created from an actor's work, but cannot use or authorize use of those materials without the actor's explicit consent and, in most cases, further payment. Digital replicas cannot be used to circumvent the hiring of background actors entirely.

SAG-AFTRA's official FAQ for background actors puts it plainly: "The employer cannot use or authorize use of those materials without your consent and, in most cases, further payments." The phrase "in most cases" does a lot of work in that sentence.

These protections exist on paper. Enforcement is another matter, and the technology continues to move faster than the contracts designed to contain it.

The economics driving the shift

None of this is happening in a vacuum. The global VFX market was valued at approximately $19.5 billion in 2024 and is projected to surpass $28 billion by 2028. But raw growth obscures what is actually happening inside the industry: compression. AI is compressing timelines, costs, and the skill differential that once separated major studio productions from independent films.

Streaming platforms are one driver. Netflix alone greenlit hundreds of originals with significant VFX requirements in 2025, and the episodic cadence of television demands that shots turn around in days, not months. When Sarandos noted that The Eternaut's building collapse "wouldn't have been feasible" at the show's budget without AI, he was describing a real constraint. The show involved approximately 2,000 VFX shots — an enormous number for a $15 million production. The gap between what a story requires visually and what a budget can realistically support has always been a problem. Generative AI is being positioned as its solution.

The human cost is harder to calculate precisely, but it is not invisible. A survey by the International Alliance of Theatrical Stage Employees found that 70 percent of VFX workers already perform unpaid overtime, and only 12 percent have health insurance. The industry was under significant pressure before generative AI arrived. The tools that studios frame as efficiency gains are tools that directly reduce the demand for the labor of people already working in precarious conditions.

AI-related job cuts across the broader US economy accounted for over 55,000 positions in 2025, according to data from the outplacement firm Challenger, Gray & Christmas. Visual effects and creative production, while not always separated in such figures, are among the sectors where the displacement is most legible — and most contested.

What audiences actually see

The phenomenology of this shift is strange. Audiences are increasingly watching content that contains AI-generated elements they cannot identify, produced by processes they have never been told about, for reasons that have more to do with budgetary arithmetic than creative necessity.

That is not entirely new — the "reality" of any given film frame has always been manufactured to a degree most viewers don't fully register. The question is whether the scale and nature of the manufacturing matters, and whether audiences have a right to know about it.

The Eternaut moment is instructive here in an unexpected way. Netflix's decision to disclose the AI use — voluntarily, on an earnings call, with the CEO's name attached — was unusual. Most productions that use AI tools at this stage do not announce it. There is no legal requirement that they do so. The disclosure came not from any principle of artistic transparency but because Sarandos wanted to tell shareholders that Netflix had found a new efficiency tool.

"The audience was thrilled with the result," he said. Which may be true. They were also, at least until the earnings call, unaware of it.

The line that keeps moving

There is a version of this story that is genuinely optimistic: AI as a democratizing force, making ambitious visual storytelling accessible to filmmakers who couldn't previously afford it, extending what is possible at every budget level. The Eternaut is, in some ways, an example of that version. It is an Argentine production, adapted from a beloved Latin American comic series, with a story too cinematically demanding for its budget to support through traditional means. The AI sequence arguably made the show better than it could have been otherwise.

There is another version in which what is being optimized out of existence is not just cost, but labor, craft, and the human decisions that accumulate into what we recognize as a visual aesthetic — the particular way a crowd of real people in a real space creates a texture that simulation has never quite replicated, the contingency of a real environment that no generated background fully captures.

Both versions are true simultaneously, which is what makes the conversation so difficult.

What is clear is that the background — the populated world behind your protagonists, the street they walk down, the crowd that surrounds them, the skyline that burns in the distance — is no longer necessarily a record of something that existed. It may be a synthesis of descriptions, trained on images of things that existed, rendered into something that looks enough like existence to pass.

You have almost certainly already watched it without knowing. The building fell convincingly. The crowd moved. The city looked like a city.

Whether any of that is a problem is a question the industry has not yet agreed to ask out loud.