A Fortress on Jeju, Rebuilt Without Ever Leaving a Living Room
The first time Young-Ha Choi watched the finished cut of his film, there was no premiere, no red carpet, and no crew to celebrate with. There was a laptop, a pair of headphones, and eight months of quiet, solitary work compressed into eleven minutes of footage. On screen, warriors in Goryeo era armor stood against a Jeju winter wind that never actually blew, at a fortress that no camera had ever pointed at, because no camera had been used at all.
 |
| A frame from Sambyeolcho, The Last Island: the AI rendered warriors of Hangpaduri fortress, holding their final winter position on Jeju. |
What Choi uploaded to his YouTube channel that week did not look like an experiment. It looked like a finished historical drama, complete with sweeping coastal shots, armored formations, and a mournful score. Only the credits gave it away: planning, screenplay, and prompt design by Claude, visualization by Higgsfield and Seedance, sound sourced from Adobe, Artlist, and Envato, final cut assembled in Premiere Pro. This is the shape South Korea's state driven AI push tends to take when it reaches an individual desk instead of a policy briefing. The government has been talking about AI as a national growth engine for months, but Choi's film is one of the clearer signs of what that talk becomes once it leaves the podium and lands in front of one working professional deciding whether to trust a new tool.
Twenty Years on Real Sets, Then None at All
Choi is not a newcomer chasing a trend. He has spent more than two decades on actual film sets, the kind with lighting rigs, blocking rehearsals, and the particular exhaustion of a location shoot running past sunset. That history matters, because it changes what his decision to drop the camera entirely actually signals. A twenty two year old creator picking up generative tools first is a story about a new generation finding its native medium. A twenty year veteran doing the same thing, after two decades of knowing exactly what a real production costs in time, crew, and compromise, is a story about the tools finally becoming good enough to matter to someone who never needed them before.
He has said plainly, in the notes accompanying his upload, that he watched paid courses promising to teach people how to make AI films selling for hundreds of thousands of won, and concluded that a well used Claude workspace alone could produce something in the same range. That is not a marketing line. It reads like the kind of judgment a person only reaches after actually trying to replace years of production habit with a chat window and a rendering queue, and finding that the replacement mostly held up.
The transition itself was not framed as a leap of faith. Claude carried the groundwork a production office would normally handle: breaking the Sambyeolcho story into scenes, drafting the shot by shot continuity, and writing out the prompt structures that would later feed Higgsfield and Seedance. Seedance 2.5 and 2.0 then carried the visual weight, turning those written scenes into the moving images an audience actually sees. What used to require a location scout, a costume department, and a call sheet became a sequence of iterations between a script document and a render preview.
Sambyeolcho, The Last Island: Directed by Young-Ha Choi 2026.8 [youtube iframe from @directorscut-aimovie]
Why a State Airport Operator Ended Up in the Credits
The film was produced with the backing of the Korea Airports Corporation, a public agency whose usual work involves runways and terminals, not historical drama. That pairing is worth sitting with rather than skimming past. A state operator choosing to support an AI generated history piece is a small but concrete data point about how public institutions in Korea are starting to treat generative content: not as a novelty to keep at arm's length, but as a legitimate channel for telling regional and historical stories that might otherwise never get funded through a traditional production budget.
South Korea's broader policy record backs up why an agency might make that call now rather than two years ago. The country's AI Basic Act took effect in January 2026, giving generative AI a defined legal footing for the first time, and the national AI Action Plan approved the following month set a public target of reaching the world's top three AI powers by 2028. Budget requests tied to that plan run into the hundreds of trillions of won, aimed at everything from GPU procurement to workforce training. A single short film funded by an airport authority looks modest against those numbers, but it is exactly the kind of small, visible use case that a national strategy needs in order to feel real to the public rather than abstract.
 |
| Two scales of the same shift: a 728 trillion won national AI push, and one director's five app workflow running on a single desk. |
A Forgotten Winter on Jeju, Told for a YouTube Audience
The story Choi chose was not an easy or obvious one for a first AI production. In 1270, when the Goryeo court surrendered to Mongol forces, a military unit called the Sambyeolcho refused to accept the terms. They retreated from Ganghwa Island to Jindo, then to Jeju, fighting a losing campaign that ended at Hangpaduri fortress in a final winter siege. It is a footnote in most general histories of Korea, familiar to specialists but largely invisible to casual audiences, and almost entirely unknown outside the country.
Rebuilding that footnote with generative tools instead of leaving it in a textbook changes who gets to encounter it. A regional history like this would rarely justify the budget for costumed reenactment or location shooting in a traditional production pipeline; the audience is too niche to cover the cost. Once the cost structure shifts toward a single creator's time and a subscription to a handful of AI tools, stories that used to be economically unviable become worth telling. The Sambyeolcho campaign gets a wider audience not because interest in Goryeo history suddenly grew, but because the price of visualizing it collapsed.
 |
| Where the film actually happened: not a set, but a desk with Claude open in one window and Seedance rendering in the other. |
 |
| Another frame from the film: the coastline retreat that carried the Sambyeolcho forces from Ganghwa to Jindo, and finally to Jeju. |
Distribution followed the same logic. Instead of a festival circuit or a broadcast slot, the film lives on Choi's channel, Director's Cut AI Cinema, alongside a stated plan for a wider series adapting Korean literature, including Kim Yu-jeong's Camellia Flowers. The channel's own description frames the project bluntly: a cinema built without a camera, run by a director with twenty years on real film sets, taking Korean history and literature and rendering all of it through generative AI from planning to final edit. Choi has also offered to publish his prompt structures and continuity methods publicly if enough viewers ask, turning what could have stayed a private workflow into a shared reference for other creators working the same way.
What Changes When the Crew Disappears
Set against a conventional production, the differences are not subtle.
- A historical short like this would typically require a location permit, a costume budget, and a multi day shoot; Choi's version required none of those, only iteration time on scripts and renders.
- Traditional casting limits a scene to the actors and extras a budget can afford; an AI pipeline lets a fortress full of soldiers appear without hiring a single one.
- A regional history documentary usually needs institutional backing before production even starts; here, the finished film came first, and institutional backing followed once the result already existed.
 |
| No set, no rig, no crew call sheet. Just a living room turned into a full post production room. |
None of this makes Choi unusual within his generation of Korean creators, even though his career length is. Younger video makers across the country have been running similar experiments quietly for the past year, but most lack the production instincts to know which shots actually need craft attention and which ones a generative model can carry unsupervised. Choi's twenty years on real sets function less like a barrier he overcame and more like a filter that let him spend his tool time efficiently, because he already knew what a convincing frame looks like before he ever typed a prompt for one. That combination, veteran judgment paired with a generative pipeline, is likely to define the next wave of Korean creator output more than raw technical novelty will.
The government's AI budget and legislative framework explain why this kind of production is now possible without legal ambiguity hanging over it. Choi's film explains why it is worth paying attention to. A national plan sets the conditions; a single desk in a living room is where those conditions actually get tested against a real audience, one upload at a time.
References
- South Korea's AI Basic Act, effective January 22, 2026 (Ministry of Science and ICT)
- Republic of Korea AI Action Plan 2026-2028, approved February 25, 2026
- Republic of Korea 2026 government budget request of KRW 728 trillion for AI era development, including GPU procurement and AI workforce training targets (Korea.net, Office of the President)
- K-content Fund increase of KRW 200 billion announced as part of the 2026 budget request (Korea.net, Office of the President)
Explore more Insights into Korean Lifestyle:
- culture / FutureKorea / GlobalInfluence / insight / pillar / SoftPowerMasteryJun 25, 2026
- culture / insight / Inspiration / Leadership / NationalStrategyJun 25, 2026
- CulturalLogic / culture / insight / KoreanHeritage / ktoday / pillarJun 25, 2026
.webp)
.webp)
.webp)
.webp)
.webp)
.webp)
.webp&description=Inside Korea's AI Video Boom: A Veteran Director's Camera Free Debut){kind=link}