Hollywood’s newest set has no skyline, only servers
Culver City still looks like Hollywood if you squint in the right direction. Promise’s base sits there, close enough to Sony’s historic lot that the old studio world feels like it’s just next door, only with a different rent bill and a very different sound. Inside, the new horror project, Touch Grass, is being made in a way that would have sounded absurd in a backlot meeting not long ago: a human actor, a small crew, and an AI-built environment that appears almost as soon as the camera rolls.
The setup is spare enough to feel almost improvised. White sheets hang where a finished set might normally be. A camera follows the performer. On a handheld monitor, the director watches the scene come together instantly, instead of waiting for a VFX team to sort it out weeks later. The actor isn’t imagining the background and hoping for the best either. The environment changes live in front of him, shifting from a grassy field to a cavern while he stays in the shot. That would make any traditional soundstage executive reach for a coffee and a spreadsheet.
The unsettling part is how ordinary it looks once the machinery starts working.
For the people building it, that ordinariness is the selling point. The production doesn’t need the kind of sprawling infrastructure that has long defined studio filmmaking. No cavernous set dressing department. No long wait for a first-pass render. No elaborate handoff from the shoot to the postproduction crew before anyone can see what the scene actually feels like. The actor performs, the model responds, and the frame fills itself in while everyone is still in the room.
That speed changes the mood on set. Scenes can be tested, altered and repeated with a kind of ease that conventional production rarely allows. If the director wants the grassy field gone and a cavern in its place, the swap happens fast enough to feel casual. That control is part of the appeal, and not a small one. In tech news terms, this is the sort of demo that makes investors lean forward. In practical terms, it means a smaller team can do work that used to require a lot more people, a lot more space and a lot more time.
Which is where the unease creeps in. The same setup that looks efficient also invites a blunt question: efficient for whom? Actors, crew members and technicians can all see the arithmetic. If an AI tool can generate the world around the performer live, some jobs may shrink, disappear, or get pushed further into the background. That’s the jobs-and-power story hiding inside the glossy language of digital culture, and it’s already shaping the talk around ai policy and power and politics as much as it is shaping the filmmaking itself.
The project also lands at a moment when studios are getting more comfortable saying the quiet part out loud. They want faster production. They want more control over costs. They want fewer bottlenecks. But when the set starts behaving like a software environment, the old balance of who does what gets rewritten in real time. A horror film about being trapped in a changing world may be a fitting place to test that idea. The trick, of course, is that everybody on set can see the walls moving.
Why the money is chasing these micro-studios
The money is following a very plain idea: if a film can be built with fewer rented stages, fewer standing sets and less time spent waiting on everything from weather to warehouse space, the math starts to look less brutal. That’s the pitch behind Promise and a growing crop of AI film studio startups, and it explains why the room suddenly includes names that rarely sit together in the same sentence. Promise has drawn backing tied to Google, venture capital and Disney, a mix that would have sounded odd in the old Hollywood order but feels almost practical in the new one.
The spreadsheet is doing a lot of the persuasion here.
For investors, the appeal is not mystery. It’s overhead. Traditional production burns cash on big soundstages, long location holds, heavy postproduction schedules and the long chain of approvals that comes with studio-era financing. Micro-studios built around generative AI in film promise a cleaner setup. A smaller crew can do more of the work on the spot. A director can see changes earlier. A producer can cut weeks off a schedule instead of paying for them. When you’re trying to make a movie for a fraction of what a conventional production might cost, that matters more than the marketing copy ever will.
The pitch gets even sharper when the numbers come up. Hybrid productions that combine human actors with AI tools are being sold as roughly a quarter to a half cheaper than standard productions, depending on how much of the pipeline shifts into software. That kind of savings can be the difference between a script that gets financed and one that spends another year wandering through email threads. Obsidian has pushed that argument hard, saying its AI workflow can trim costs by about a third. That figure has helped bring established producers to the door, because a third off the bill is not a side note. It’s a very nice reason to pick up the phone.
There’s also a broader pressure point in the middle of the market. The film world has long been stuck in an awkward gap where blockbusters get the money and tiny indies get the passion, while mid-budget projects sit around asking who wants to pay for them. AI-heavy studios are now selling themselves as a way to make those films viable again. The sweet spot, at least on paper, sits in the tens of millions, not the hundreds. That range is big enough for recognizable actors, polished visuals and a real launch plan, but small enough that savings from AI tools might actually change the greenlight conversation. A thriller that once needed a giant infrastructure package might now get made with a leaner setup and a less terrifying balance sheet.
That’s why the interest isn’t coming only from startups trying to sound clever at a pitch meeting. Established companies are watching too, because the old model is expensive before a camera even rolls. A studio lot, a series of rented departments, a long VFX queue and months of financial exposure all add friction. If a new workflow can remove some of that friction, legacy players can test more projects without betting the farm on each one. It’s easy to see why the suits are sniffing around. They don’t need to love Hollywood AI to notice when it might shave real money off a production slate.
The move is also about control. Giant soundstages and old-school financing structures tend to concentrate power in the hands of the people who already own the pipes. AI-first studios sell a different arrangement. They can work with less physical footprint, fewer fixed assets and a more flexible production process. In practice, that means fewer gatekeepers between a script and a finished scene. For some filmmakers, that sounds liberating. For investors, it sounds efficient. For the people who currently make money from the old bottlenecks, it sounds less charming.
Google’s own recent AI film activity helps explain why this suddenly feels less like a fringe bet and more like a serious experiment. Its DeepMind partnership with A24 points to an obvious fact: major tech companies are no longer treating film as a novelty use case. They’re putting research money and product development next to it. The same goes for Flow updates, which show how fast generative video tools are being tuned for production workflows rather than casual demos. Once those tools start looking useful to real crews, the business case gets easier to sell.
Even the policy side matters here, if only because the companies making these tools need to tell studios what they’re signing up for. Runway’s film terms sit in that awkward but necessary corner of the market where creativity meets contracts. Nobody is pretending the legal side is simple. Rights, consent, reuse and credit all need sorting out before these systems can be trusted at scale. But the existence of those terms tells you the same thing the financing does: this is not just a flashy demo. It’s a product category being dressed for actual production money.
For now, the pitch is blunt. If AI can help keep a mid-budget film alive, reduce dependence on giant stages and cut a meaningful slice off production costs, investors will keep showing up. Legacy studios will keep comparing notes. And the smaller outfits, the ones trying to make movies without dragging an entire old system behind them, will keep calling that a business model instead of a moonshot.
How the shot actually works: actors in front, AI in the frame
What makes this setup feel less like a demo reel and more like an actual working pipeline is how ordinary the physical side looks. A performer stands on set. A camera tracks them. A small crew watches a monitor. Then the screen fills in the world around them before the take is even over.
For the Promise shoot, the studio used Seedance 2.5, a Chinese video model, to generate the background in real time while the actor performed. The actor’s movement was composited into the AI-built environment immediately, so the finished frame was visible on set instead of waiting for a VFX team to clean everything up later. That’s the part that changes the rhythm of the day. A director can see blocking, camera movement, and scene mood at the same time, while the performer reacts to a world that seems to exist already. A grassy field can appear in the monitor one moment, then swap out for an underground cavern almost instantly the next.
The set no longer waits for postproduction to tell everyone what the scene was supposed to be.
That speed changes what a shoot feels like for everyone in the room. Actors can respond to the environment instead of pretending it’s there and hoping the edit team will make sense of the blanks. Directors can make calls in real time, which sounds obvious until you remember how many productions have spent months fixing the fact that the “obvious” shot in the room looked weird once it left the stage. Here, the frame is built as the performance happens. That means the crew gets to judge light, motion, and composition while the scene is still alive.
The useful part is flexibility. A meadow is not a meadow for long. If the script wants a cave, the background can be replaced without rebuilding a set or waiting for a hard drive full of renders. If the shot needs fog, debris, or a strange glow from somewhere off-screen, AI-generated effects can be layered in with the same general workflow. And if the scene calls for synthetic performers, the system can slot those in too, whether they’re background figures, digital stand-ins, or other fabricated bodies that would once have required a separate pipeline. In other words, background replacement is only the easiest thing this kind of film production technology can do.
That broader toolkit matters because studios are not testing one narrow trick. They’re pushing on several parts of the pipeline at once. Google’s Flow and Veo AI filmmaking tool points in the same direction, giving creators a way to shape AI video with more control over scenes and continuity. Runway has been moving too, with Project Luxo and its work on building Runway characters, both of which treat motion, identity, and scene construction as software problems rather than separate crafts that only meet in post. The details differ, but the direction is hard to miss: the frame is being assembled earlier, faster, and with fewer handoffs.
That does not mean the old tools disappear overnight. Far from it. A camera still has to be aimed well. An actor still has to sell the scene. Someone still has to decide whether the cave should feel damp or sinister or oddly clean, because audiences notice those things even when they can’t explain why. The novelty here is that a live performance and a generated world are no longer split across separate stages of production. They happen together, in the same room, under the same clock.
Joel Hynek, a veteran visual-effects artist, reads the moment that way. He has seen enough shifts in the business to recognize one when it walks in wearing a headset and a laptop grin. For him, this feels like another jump of the sort that once separated optical effects from digital compositing. That earlier change didn’t erase filmmaking. It changed where the work lived and who had to know what. This one appears to be doing something similar.
The catch, of course, is that once the scene can be built live, the control center moves too. The person on set gets more power over the final image, but so does the software that generates it, and so do the companies that own the system. That’s the part people keep circling back to, even when the demo is impressive and the cavern looks pretty good. The next fight is no longer about whether the shot can be done. It already can. The real question is who gets to decide what gets made when the tools become this fast, and how many jobs sit on the other side of that speed.
Backlash, big-name bets and the fight over what comes next
For all the excitement inside rooms like Promise’s, the public mood is still pretty sour. A lot of filmmakers hear the pitch and immediately ask who gets replaced when the software gets better. That skepticism hasn’t been softened by the speed of the rollout. The tools are landing in productions, demos and pitch decks before many crews have worked out what the contracts should look like.
Major directors have been vocal about that. Some have argued that generative systems are being pushed into film and TV faster than the industry can judge their effects, and they’re not wrong to notice the audience reaction. Viewers can spot when a face looks a little too smooth, when a performance feels stitched together, or when a studio starts treating a human actor as a bundle of data points. The complaint isn’t just aesthetic snobbery. It’s a trust issue. If the audience thinks a movie was assembled by software first and people second, the bargain changes.
The fight isn’t really about whether AI can make a scene. It’s about who gets paid, who gets replaced, and who gets to say yes.
That question lands hard in labor talks. Actors worry about digital replicas that can be rented forever after a one-time scan. Writers worry about studios training models on their work and then asking those same models to produce something “close enough.” Union leaders have pushed for clearer rules around consent, compensation and reuse, which is where AI policy has started sounding less like a tech memo and more like a workplace fight. Nobody wants to sign away a face, a voice or a style and then find it popping up in a sequel, a commercial or some half-finished spin-off they never touched.
The backlash has only sharpened because the examples keep getting stranger. When dead performers are brought back through digital tricks, the reaction is rarely “cool trick, nice job.” It is usually some mix of discomfort and suspicion. The same goes for AI-made “stars,” synthetic personalities built to look marketable enough to front a campaign or fill a frame. Those experiments may be technically neat. They also make a lot of people ask whether the business is drifting toward a future where likeness becomes inventory and performance turns into a licensing line item.
AI advocates see the same scene and come away with a different read. To them, these tools could open a stretch of filmmaking that small teams have not had in decades. A director with a laptop, a lean crew and a weird idea might get farther than they ever could in the old system of huge soundstages, long approval chains and expensive postproduction. That argument has real pull for independent producers who have spent years trying to make mid-budget films pencil out. A horror movie, a sci-fi short or a very odd drama might finally be affordable without begging three layers of gatekeepers for permission.
There’s some romance in that case, but also a hard question of ownership. If AI makes production cheaper, does the savings go to the filmmaker, the audience and the crew, or does it mostly get captured by the companies that own the models, the cloud spend and the distribution pipes? That’s the part nobody can quite smooth over with a slick demo. A future with more voices on screen sounds lovely. A future where a few tech-heavy firms collect the fees while everyone else gets a smaller slice sounds a lot more familiar. And that, more than the headline-grabbing stunts, is where the real fight is heading.
