Four days, a green screen, and AI: Making a Korean horror film with a full cast but no set or effects

Apartment has all the hallmarks of a classic Korean horror film: Creepy atmospheric settings, twisted and monstrous creatures, and clever acting all overlaid with a good dose of social commentary.

Yet the biggest surprise hiding around every corner is how the movie was made.

CJ ENM shot the principal photography for this 60-minute occult horror film in just four days. The actors performed on a green-screen stage, but every environment around them — including the hallways, the rooms, and the views out the apartment windows — was generated using Google Cloud's AI tools.

When Apartment was released on April 30, the project became yet the latest demonstration of the potential for an AI-native production pipeline. We’ve seen lots of examples already of AI pushing the boundaries of filmmaking, and here was yet another, this time delivering a complete narrative film, and at a fraction of what a comparable production would normally cost.

Four days on stage, the rest generated

Apartment follows Yumi, a woman who can see ghosts, as she moves into a new apartment and starts to encounter what's already living there. The genre is a classic but the production approach is new.

Every performance scene was shot on an indoor green-screen stage at CJ ENM. There was no location work and no exterior set construction, both of which are typically among the most expensive line items in a production budget. The apartment building, its corridors, and the world outside its windows were all virtual environments.

The team began the project using Imagen for environment stills, integrated Nano Banana once it launched, and went back to upgrade earlier shots after Nano Banana Pro became available. Veo 3 then brought those still environments into motion, turning the generated apartment interiors and exteriors into the moving footage the actors' performances were composited into.

The hardest part was cultural specificity. A lot of AI models default to Western standards, since those tend to dominate the culture, and that includes cinematic conventions. But the team needed an apartment building an audience in Seoul would recognize, not one that looked more at home in New York or Paris.

"Most generative AIs have been centered on Western artwork, often generating images that are too realistic or, conversely, too dramatic," said Chang-ik Jung, AI studio lead at CJ ENM. "This is why many movies based on generative AI are often set in space, deserts, or future dystopias."

Nano Banana's multimodal input made Korean-specific direction possible. Jung's team could hand the model storyboard sketches and reference images in addition to describing shots in text prompts, giving the director direct control over camera angles and composition.

Maintaining that level of consistency across a 60-minute runtime has been one of the field's stubborn technical problems, since most generative models lose coherence well before feature length. Holding that fidelity for an hour, across every character, every room, and every shot, was the threshold the production needed to clear.

"During the production period, the development of Nano Banana gave us the conviction that we can do more," Jung said.

Lower budgets, retained rights

Hyun-jung Baek, chief AI officer and head of content AI-tech at CJ ENM, framed the strategic question more directly. Production costs for film and television have climbed so steeply that studios are sometimes forced to trade away rights to their own work in exchange for production financing. For a company whose long-term value lives in its catalog, that trade-off cuts at the heart of the business.

"We believe that content production using AI will be of great help in protecting IP sovereignty as a production company, beyond just making better videos," Baek said. "AI could eventually open up opportunities for production companies that own IP to create high-quality content themselves within limited production budgets."

Lower production costs also widen the field of who gets to make films in the first place. When a high-quality feature no longer requires studio-scale financing, new creators can compete on the strength of their ideas rather than their access to capital.

Baek described this as moving beyond an era in which "the conditions for a good work were always linked to large-scale capital," toward one that draws in new creators who succeed through creative ideas. A deeper pool of working creators is good news for the studios that eventually partner with them.

Replacing CGI, not performances

One question deserves a direct answer: Apartment didn't replace its actors or its director.

What generative AI took over were the parts of production that used to demand expensive CGI, location work, or built sets, including the ambitious sequences directors most want to shoot but rarely get to. Large environments, complex atmospherics, the kinds of shots the industry calls "money shots" are typically cut in early budget conversations, long before a crew is hired or a camera rolls.

Generative AI makes those scenes affordable enough to actually produce, which means audiences will see work that would never have made it onto a shooting schedule under the old economics.

The day-to-day workflow on Apartment shows how the human-AI division of labor played out in practice. When Jung needed a specific camera angle or framing, he drew it as a storyboard, fed the sketch to Nano Banana, and got back a generated image that matched his direction. The creative decisions stayed with him. The tool handled execution.

An industry set for transformation

Apartment is one film, but the model behind it is portable. Live performances on a green-screen stage combine with generated environments and a short principal photography window to produce a finished film that holds together visually from start to finish. The same approach scales to longer formats, to other genres, and to the kind of ambitious work that has spent the last decade sitting on the shelf because no one could afford to make it.

For studios, that means a path to producing more of their own catalog. For directors, it means putting more of what they imagined on screen.

To discover how Google Cloud's generative AI tools can support your next production, explore our media and entertainment industry solutions.