Korean AI production pioneers have cautioned that artificial intelligence could eventually make feature filmmaking more expensive than the traditional 35mm film era. Speaking at the Taiwan Creative Content Fest in Taipei on November 8, industry leaders highlighted that while AI reduces costs for commercials, the mechanics of token usage and iterative revisions pose significant financial risks for feature films.
Kim Kwang-jib, founder and CEO of Studio Meta-K, explained that the token bill for a 100-minute film using premium commercial tools can exceed 500 million won, or approximately $366,000. He noted that AI companies are currently keeping prices low as a marketing strategy while operating at a loss. Kim warned that a tenfold increase in token costs could push expenses to 5 billion won, or $3.66 million, a figure that would challenge the efficiency of mid-budget Asian productions.
Unlike the digital era, where shooting and cutting freely incurred no extra material costs, AI requires payment for every re-render. Kim illustrated that if a client submits changes individually over ten days, the studio must pay for ten separate render passes. Conversely, batching changes into a single round reduces costs significantly. He argued that without meticulous preproduction, the final budget could surpass that of the celluloid days.
Lee Sang-wook, head of the AI Content Lab at MBC C&I, agreed that budgeting is complex due to rapidly evolving tools. His lab's 139-minute feature, Raphael, took more than a year to produce, during which new software updates arrived weekly. Lee noted that a single shot could require ten artists for a full day, and using outdated technology would make scenes appear dated to modern audiences.
The film, created by emerging directors with support from the Korean Ministry of Culture, Sports and Tourism, is set to open in Korean cinemas on November 25. Lee explained that the cinema release was chosen to demonstrate that AI studios can take commercial risks through IP development, rather than relying solely on outsourcing. This follows an earlier initiative in 2024 where the lab produced AI shorts that received a cold market reaction.
In the advertising sector, the cost savings are already evident. Kim cited a Pizza Hut commercial that originally cost 2 billion won to produce over eight months. Studio Meta-K recreated the spot in two weeks for approximately 30 million won, reducing the expense by more than 90%. However, for drama productions, Kim favors a hybrid model that combines live-action star performances with AI-generated background elements, such as large crowds or war scenes.
Studio Meta-K is also testing a reversed pipeline where entire films are generated in preproduction, with only specific emotional close-ups reshot in live action. The company, founded in 2022, has completed over 100 commercial projects and recently raised 10 billion won in funding. It is preparing for an IPO next year and is currently collaborating with Busan's Dongseo University to convert Im Kwon-taek's 1993 classic Seopyeonje into an animated feature for a theatrical release next summer.
The two experts diverged on the use of open-source tools. Kim praised the recent improvements in free models, while Lee expressed caution due to concerns over training data transparency and compliance with EU AI Act guidelines. Lee also noted that delivering AI content to streamers is technically challenging due to strict mechanical requirements for bitrate and color grading.
Regarding employment, Kim predicted that large on-set crews will become rarer, with roles like drone operators potentially disappearing. He believes that workers who master AI will replace those who do not. Lee compared the transition to the introduction of sound or digital cameras, predicting the rise of new roles such as previsualization artists and continuity supervisors. He added that actors have generally welcomed AI, particularly for handling dangerous scenes, though he maintains that human emotional detail remains irreplaceable.
The session was moderated by Archin Yen of WWWIND Studio. When asked how directors can maintain their unique vision, Kim emphasized that while AI tools can be learned quickly, film grammar takes years to master. He observed that most AI-generated content defaults to standard eye-level shots, and concluded that the human element ultimately determines the quality of the final product.
