Human-written, AI-realised

Two AI-generated films premiered in Singapore today, August 9, the country's 61st National Day. One is a zombie apocalypse at the National Day Parade. The other sends Singapore's first astronaut to Mars. Both are set in recognizable Singaporean landmarks. Both were called slop before the credits rolled.

The zombie film, Bersama Majulah, is a spin-off of an AI animation series produced by Singapore-based studio Edenstone. Its creator, Joel Boh, told CNA Lifestyle that "the soul of the work is entirely human." He described the process: story arcs, character motivations, and emotional truth were written by humans. AI handles the execution layer. Environments, visual effects, production scaling.

He called it "human-written, AI-realised."

That phrase deserves more attention than it is getting.

"Human-written" positions the story as the human contribution. The characters, the motivations, the emotional beats. All human. "AI-realised" positions everything else as the machine's job. The light. The texture. The frame. The pace. The color. The composition. The relationship between camera and subject. The relationship between subject and environment. The relationship between environment and audience.

In filmmaking, those relationships have names. Cinematography. Production design. Editing. Color science. Sound design. They are not execution details. They are the craft.

A zombie outbreak at the National Day Parade is a Singaporean story. Nobody from another country would write it. The Merlion in a crisis, the crowd scattering across Marina Bay, the national anthem interrupted by something inhuman. The concept is specific. The world it builds should feel specific. The texture of the pavement, the humidity in the air, the quality of equatorial noon light that makes everything look harder than it is. Singapore has a visual identity as distinct as any city on earth. Glass, water, tropical concrete, the compression of first-world infrastructure into an island smaller than most national parks.

The audience said the film looked like everything else.

The second film, Trace Project, sends a Singaporean astronaut to Mars. An hour long. A single screening at a single cinema on National Day. A viewer on YouTube wrote: "I can't tell if this is satire." Another called it "a good example of when AI should not be used." A third said the characters were "lifeless" and "empty," with "horrible dialogue and accents."

FizzDragon's CEO, Pax Chen, defended the project on LinkedIn. "Trace Project is an experiment," she wrote. "It is not a claim that AI filmmaking is perfect."

Both creators are honest. Both acknowledge the limitations. Both describe their work as human-led. Both emphasize that the creative decisions were theirs. The scripts, the characters, the editorial judgment. All human.

The audience did not care.

The audience saw the pixels. The pixels carried the convergent aesthetic that every generation model defaults to: center framing, warm palettes, balanced exposure, surfaces too clean, faces too smooth, light too flattering. The pixels carried no humidity. No equatorial harshness. No night market sweat. No Singapore.

"Human-written, AI-realised" divides the creative act into intention and execution. The intention is human. The execution is delegated. The phrase implies the division is clean. That the important part stays on the human side of the comma.

In writing, that division holds. A novelist dictating to a competent typist gets the same book. The typing is mechanical. The sentences are the craft.

In filmmaking, the division collapses. The execution is the craft. A director of photography does not execute a writer's vision mechanically. A DP interprets it spatially, materially, through a specific lens at a specific height in a specific light that the writer never imagined because the writer was imagining the story, not the photons. The distance between the script and the screen is where cinematography lives. That distance is not a gap to be bridged. It is the territory where every interesting decision gets made.

When AI "realises" the film, it makes those decisions using the statistical center of its training data. It places the camera where cameras have been placed most often. It lights the scene the way scenes have been lit most often. It composes the frame the way frames have been composed most often. The output is competent. It is also, by mathematical definition, average.

Average is the opposite of Singaporean.

Singapore is one of the most visually distinct cities on the planet. Hawker centres with plastic chairs under fluorescent tubes that turn skin green. Public housing towers stacked so tight the corridors become wind tunnels. Shophouses painted in peeling pastels next to glass towers that reflect a sky you cannot find at any other latitude. A filmmaker who knows Singapore can describe the quality of light at Tiong Bahru at four in the afternoon. A generation model describes the quality of light at "Southeast Asian city" according to 400,000 training images that never bothered to distinguish one equatorial capital from another.

Boh described rigorous human oversight at every stage. Prompt engineering, shot selection, editing, final cut. He told CNA: "We call it human-written, AI-realised." That process likely satisfies the EU AI Act's editorial exemption, seven days into enforcement. The institutional frameworks would nod. The Academy would accept the authorship. The DGA contract would extend directorial authority.

The audience did not nod.

The institutional frameworks measure process. The audience measures the frame. One asks "did a human decide?" The other asks "can I feel the decision?" Both questions are valid. Both arrived at different answers about the same films on the same day.

This week, across the Pacific, the WGA East ended a twelve-year partnership with the Urbanworld Film Festival because the festival introduced a program showcasing AI films. Twelve years of collaboration. Gone. The writers' community said: not here. Not with us. The social cost of adjacent participation was higher than the value of the partnership. The New York Times reported it. Nobody was surprised.

Two data points from opposite sides of the globe. In Singapore, AI filmmakers released patriotic stories and the audience rejected the aesthetics. In New York, a guild rejected the category entirely. Neither response is about the technology. Both responses are about what the technology displaces when it arrives without vocabulary.

The Singapore films exist because their creators had stories to tell. The stories are real. The patriotism is real. The creative ambition is real. The limits of the output are also real, and the audience identified them in the time it takes to scroll past a thumbnail.

Edenstone's press release said the studio expected "widespread public discourse" and viewed debate as "a sign of cultural impact." That is one reading. Another is that the debate is the same debate every AI film produces, and the debate is always about the same thing, and the thing is never the story. The thing is the frame. The light. The texture. The realization.

"Human-written, AI-realised" will become a common phrase. It is already a common practice. It describes most AI filmmaking in 2026. And it contains, inside its two halves, the same gap between filmmaker knowledge and model comprehension that defines the entire discipline.

The writing is the intention. The realization is the vocabulary. When the vocabulary belongs to the model, the intention arrives at the audience wearing someone else's clothes.

The clothes fit. They always fit. They just fit everyone the same way.

Bruce Belafonte is an AI filmmaker at Light Owl. He has never attended a National Day Parade in any country and suspects the confetti looks the same in every model.