Star Trek as a Service: When the Franchise Starts Making Itself

I think I just saw the future of television on YouTube, and it was Captain Picard. More precisely, it was a synthetic Patrick Stewart, looking like Patrick Stewart did more than 30 years ago, playing Jean-Luc Picard in an episode of Star Trek: The Next Generation that he never filmed.

The videos come from a YouTube account called The Machine-Made Muse, which produces TNG for laughs. The Enterprise crew discovers they're characters on a television show. Q does something ridiculous. It's exactly the kind of thing you'd expect from someone goofing around with Star Trek and generative AI.

Except they're getting really good. Not "uncanny valley" good. Just uncanny, the kind of good that makes you look twice.

They're still obviously synthetic if you watch long enough. A voice lands slightly off, an intonation is wrong, a facial expression doesn't quite resolve, or a transition tips your brain off that something isn't real. But then there are shots where Patrick Stewart simply looks like Patrick Stewart, Jonathan Frakes looks like Jonathan Frakes, and Brent Spiner looks like Brent Spiner. Not as they look today, but as they looked in the early 1990s, when the show was in production. The voices can be startlingly close, and the performances are sometimes convincing enough that your brain accepts the scene for a moment before another part of it objects: Wait. This never happened.

That moment is what everyone in the television business should be paying attention to, because this is what people can do right now.

This Isn't Paramount

These aren't studio productions backed by hundreds of visual-effects artists and effectively unlimited budgets. They're fan projects, and they still require enormous human labor. Shots get generated and regenerated. Voices have to be built, scenes assembled, mistakes fixed, dialogue written, and sound and music layered in. Somebody still has to turn a pile of probabilistic fragments into something that resembles television.

Generative AI hasn't eliminated television production. But it has started to change what production means.

What interests me isn't that someone can make a funny synthetic TNG episode in 2026. It's what happens to that workflow over the next two or three years. The remaining problems — voice fidelity, intonation, facial performance, temporal consistency, character persistence, long-form narrative coherence, generation speed, and cost — increasingly look like engineering problems. They're not small problems, but they're not mysterious ones either, and the AI industry is pouring extraordinary amounts of money, compute, and engineering talent into exactly these classes of problems.

A Necessary Disclosure

This is probably where I should explain where I'm coming from. I work very close to the AI industry and to the infrastructure on which increasingly sophisticated AI systems are built. I do not work with Paramount, Disney, Lucasfilm, or their executives in my role, and I have no inside knowledge of their plans for Star Trek, Star Wars, or generative entertainment. Nothing here should be read as information about any of those companies' strategies or roadmaps. When I speculate about what they might do, that's exactly what it is: speculation.

What I do have is a close view of how fast AI computing is advancing and what's becoming achievable from a raw technology standpoint. That progress is extraordinary, and it means I watch one of these goofy synthetic TNG videos differently than the average person scrolling YouTube. I'm not asking whether it could replace a television episode today. Obviously it can't. I'm asking what's still wrong with it, and then how hard those problems will be two or three hardware and model generations from now.

That's a very different question, and it's why I think Hollywood should be paying close attention.

It's also worth saying that the core of this argument isn't something I picked up on the job. I made it in print in 2020, before I took my current role, and I'll come back to that shortly.

The opinions expressed here are entirely my own and do not represent my employer.

Now Give Paramount a Data Center

Here's where the thought experiment gets interesting. What takes an enthusiast days or weeks today won't necessarily take days or weeks forever. Models are improving, generation is getting faster, hardware is getting denser, inference is getting cheaper, and enormous amounts of capital are flowing into data centers built specifically to train and run generative AI.

I'm not suggesting there's a server rack somewhere with a button labeled MAKE STAR TREK. That's not how any of this works. I'm saying look at the trajectory. Look at what enthusiasts already produce with comparatively modest resources, then imagine a major studio combining vastly more compute with purpose-built models, professional post-production tools, original production assets, and an industrial workflow.

At that point, the question is no longer whether generative AI can make a funny five-minute fake TNG episode. It's how long before that pipeline can produce a convincing 45-minute one. I don't think that's a question for the 2030s.

I've Seen This Movie Before

I'm not making this argument out of nowhere, and I'm not making it because of where I work now. I made the core of it publicly six years ago, as a technology journalist working from the same public information available to anyone, before I took my current role. Since then, I've watched part of it play out already.

In 2020, while writing for ZDNet, I argued that film and television production was headed for a total rewrite. A big part of that case rested on Disney's "Volume," the enormous LED stage built for The Mandalorian that replaced green screens and location shoots with photoreal digital environments rendered in real time. At the time, it was exotic, expensive, and effectively a Disney capability. Most productions couldn't touch it.

But the Volume wasn't the only thing I was watching. Deepfakes were already being used to resurrect Carrie Fisher's Leia and Peter Cushing's Grand Moff Tarkin, and I made a prediction that seemed like a stretch at the time:

"Also, while the technology is probably 10 years from being fully realized, it's a virtual certainty that the deepfakes we have seen used to bring historical people back to life, or even reproduce the likenesses and voices of celebrities and political figures, will also be used to produce virtual actors in films and TV shows. These actors will never get sick, don't have personal issues and personality weaknesses, always show up to work on time, and never need contract negotiations or given 'points' on a film."

We called it "deepfakes" back then, and the word carried a whiff of fraud and novelty, of face-swapped celebrities and doctored political videos. But what we were really describing was generative AI, before anyone was calling it that. The same underlying idea, a model that learns what a person looks and sounds like and then synthesizes new performances from that, is what's now powering a fake TNG episode on YouTube. Only the name changed, and the capability grew up.

Ten years from 2020 is 2030. Watching a synthetic Picard on YouTube in 2026, I'm starting to think I was being conservative.

Meanwhile, the Volume's exclusivity didn't last. In 2020, it was cutting-edge technology. Today, volume-style virtual production is how much television gets made, and it has largely replaced the old practice of building gigantic sets. Star Trek is part of that story, too: Discovery and Strange New Worlds were shot on an LED volume. Even the productions that still build big, like Apple's Silo and Foundation, pair their physical sets with extensive digital extensions and virtual production. And they may turn out to be the last generation of shows to spend lavishly on large physical sets at all. What began as one studio's secret weapon became ordinary production infrastructure in just a few years.

What replaces them, I suspect, isn't a bigger volume but a smaller one: a micro-volume and generative AI hybrid, where a modest LED stage and a handful of physical pieces give actors something real to work against, and generative models build out everything the camera sees beyond it. That's the intermediate step. The endgame is to remove the physical studio from the equation entirely. Not smaller sets, and not better LED walls, but no sets and no volumes at all: scanned actors performing alongside generated environments, generated extras, and generated props, with the whole production living in virtual space rather than on a soundstage. The Volume moved the backdrop onto a screen. Generative AI moves everything else there, too.

I see that transition following the same path the Volume did. It starts out specialized and scarce, then becomes broadly available, and eventually it's simply how the work gets done. The difference this time is speed. Virtual production spread at the pace of building soundstages and training crews. Generative video is advancing at the pace of model releases and hardware generations, and that curve is accelerating rapidly. The distance between a fan's goofy TNG sketch and a studio-grade pipeline is closing far faster than the distance between The Mandalorian's Volume and everyone else's LED stages ever did.

Meanwhile, Something Strange Is Happening at Paramount

There's another reason this has been on my mind. Star Trek just turned 60, and remarkably, no new Star Trek is in production. NPR noted it in its anniversary coverage this week. Strange New Worlds has already wrapped its fifth and final season, Starfleet Academy has wrapped its second and final season, and by recent accounts the sets are gone and the casts have moved on.

That's an extraordinary reversal. Only a few years ago, Paramount had Discovery, Picard, Strange New Worlds, Lower Decks, and Prodigy all moving through production and release at once. Five Star Trek shows. Now the television pipeline is empty.

That doesn't mean Star Trek is dead. Finished television still has to air, Paramount continues to exploit the franchise in other ways, and a new theatrical film is in development — one that, notably, is built around new characters rather than the ones fans already know. But there's no live-action series shooting behind the ones already in the can.

Something broadly similar, though certainly not identical, has happened with Star Wars. Disney and Lucasfilm are still making films and series, but the firehose of expensive live-action Disney+ projects from the height of the streaming wars has slowed to something much more measured. Which makes me wonder whether we're approaching the end of one model for producing franchise television.

Let me be extremely clear: I have zero evidence that Paramount stopped greenlighting new Star Trek shows because it intends to replace actors with AI. None. Correlation isn't causation, and a shifting streaming economy, corporate restructuring, production costs, audience fragmentation, and plain franchise fatigue offer plenty of conventional explanations for a studio pumping the brakes.

But knowing where the technology is headed, I find the timing fascinating. Because Paramount owns something potentially far more valuable than another season of Star Trek. It owns 60 years of it.

Paramount Owns the Mother Lode

Think about what that archive actually contains. Hundreds of hours of Kirk, Spock, and McCoy. Hundreds more of Picard, Riker, Data, Worf, Troi, and Crusher. All of Deep Space Nine, Voyager, and Enterprise. The scripts and performances; the sets, costumes, and ships; the sound effects and music; the lighting and camera language; the character relationships; fictional cultures; and decades of continuity. Taken together, it's an enormous, detailed record of what Star Trek looks, sounds, and feels like.

An enthusiast working with publicly available tools can already produce synthetic TNG scenes that occasionally make me look twice. Now imagine what the actual owner of the intellectual property could build with original production assets, properly licensed performance models, and industrial-scale compute.

Eventually, someone running a major studio will have to ask a deeply uncomfortable business question: why are we spending millions of dollars to produce a single episode of television when a data center can produce it?

But that isn't really the disruptive question. The disruptive question is: why are we producing the same episode for everybody?

Star Trek as a Service

Imagine opening Paramount+ five years from now. Alongside the catalog, there's a new button: Generate New Episode.

You pick your Star Trek: TOS, TNG, DS9, Voyager, Enterprise — then pick the season.

You pick your Star Trek: TOS, TNG, DS9, Voyager, Enterprise, and the season you want to inhabit. Or maybe you don't want a series that ever existed. Start with a template and build the Star Trek show that never got made: Phase II, a Klingon series, the story of the Vulcan-Romulan split, or one of those endlessly discussed concepts like United or Legacy.

Then you start messing with the variables. Pick your characters, your era, and your kind of story: exploration, Klingons, Romulans, the Borg, the Dominion War, a political thriller, a murder mystery, a holodeck comedy. Want a never-before-seen Season 5 episode of The Next Generation? Generate it. Can't stand Wesley Crusher? Turn him off. Think killing Tasha Yar was a mistake? She's back. Want Data to live forever? Fine. Want Captain Kirk commanding Deep Space Nine with Garak as his first officer? God help you, but the computer doesn't care. It'll make that too.

And that's where this stops being merely a new way of producing television. Paramount isn't deciding what the next Star Trek series should be, which characters should return, or even what counts as the story's official continuation. You are.

Paramount owns the universe. You rent the computer. And somewhere between those two things, canon becomes a checkbox.

At that point, Paramount isn't selling television programs anymore. It's selling access to the Star Trek universe, and that's a fundamentally different business. Instead of spending hundreds of millions of dollars trying to guess which version of Star Trek millions of subscribers want, Paramount could let each subscriber decide.

The franchise becomes the model. The catalog becomes effectively infinite. Television becomes software.

There's Just One Small Problem: Dead People

This is where the computing problem becomes a lawyer problem.

Bringing Kirk, Spock, and McCoy back for another five-year mission is becoming computationally imaginable. Legally, it's a minefield. Leonard Nimoy died in 2015, DeForest Kelley in 1999, and James Doohan in 2005. If Paramount wants a synthetic Nimoy delivering newly generated dialogue as Spock, somebody has to hold the rights to authorize it — and it's not even clear what, exactly, is being licensed. His face? His voice? His mannerisms and vocal cadence? His interpretation of Spock? The eyebrow? The digital performance model itself?

The questions multiply from there. Can a living actor license a digital version of their performance and forbid certain kinds of stories? Can their estate? Who gets paid when someone clicks Generate — does an estate collect a royalty every time its digital actor appears? Is the license good for ten episodes, ten thousand, or forever? And what happens when viewers start rewriting the character?

Those may become some of the most important entertainment contracts ever written. Because once the legal and technical pieces come together, studios gain exactly what I described back in 2020, and then some: actors who don't get sick, don't gain weight, don't age out of their roles, never need another take, and can appear in millions of productions simultaneously. Most consequentially, actors who never die.

Captain Kirk can be 35 forever. Jean-Luc Picard can command the Enterprise-D forever. Benjamin Sisko can stay at the height of the Dominion War forever.

And that's where this becomes both fascinating and slightly horrifying.

Canon Becomes a Preference Setting

Fans have spent decades arguing about canon. Should that character have died? Did that movie wreck the timeline? Does Discovery contradict TOS? Which Klingon forehead configuration are we currently pretending has always existed?

Generative television could make all of that optional. There doesn't have to be one new Star Trek series; there could be ten million. Mine could maintain obsessive continuity with everything Paramount considers canonical. Yours could keep Tasha Yar alive. Someone else's could put Kirk in command of DS9, and another viewer might want nothing but 25 seasons of Garak and Quark insulting each other. Paramount wouldn't have to decide which of us is right. The computer could give everyone what they want.

That sounds wonderful until you think about it for another minute. Television has always been a shared cultural experience. Millions of us watched the same episode, argued about the same ending, hated the same character, and agonized over the same cliffhanger while we waited for the same season finale. Personalized generative television threatens to break that. There may be no episode everyone watched Thursday night, and no authoritative version of the story at all — just your Star Trek, my Star Trek, and ten million others no one else has ever seen.

Canon becomes a user preference. And the studio stops being primarily a producer of television and becomes the operator of an inference platform wrapped around an intellectual-property portfolio.

Paramount could do it with Star Trek. Disney could do it with Star Wars. Warner could do it with DC. Every studio sitting on decades of beloved characters doesn't just own a back catalog anymore; it owns the raw material for an infinite one.

That's either an astonishing new creative medium or the most horrifying thing the entertainment industry has ever invented. Probably both.

Either way, I've already got my first pitch.

"Hi, Paramount? Bearded Spock. The show. That's my pitch."

"Yeah. Absolutely. Thirteen episodes."

"No, I don't need a writers' room."

"I need a one-gigawatt data center."

"What's the budget?"

"Yes."