SEOUL, August 03 (AJP) - A recent K-drama hit, Agent Kim: Reactivated, had all the ingredients of a blockbuster action sequence — North Korean agents, a former security-service commando, gangsters, explosions and a high-speed chase through a snow-covered tunnel ending in a plunge into icy water. None of it was filmed.
The roughly three-minute sequence that fooled audience as an expensive location shoot was generated entirely with artificial intelligence.
Described by its creators as the first full sequence in a Korean commercial drama produced with generative AI rather than live-action photography, the scene has become a test case for an industry confronting soaring production costs, shrinking advertising revenue and increasingly tighter production schedules.
The experiment also reached far beyond South Korea.
Agent Kim: Reactivated debuted at No. 1 on Netflix's global non-English TV chart with 10.5 million views and retained the top spot the following week with 9.1 million.
The figures do not show whether audiences embraced the AI-generated sequence itself, but they placed the experiment inside one of Korea's most widely watched television exports.
Its significance lies less in whether AI can produce spectacular images than in why it was used.
The sequence was initially planned as a conventional shoot but was deemed impractical because of the locations, action scenes and visual effects it required, according to Ryu Jae-hwan, deputy head of Morpheus Studio, the AI production company behind it.
Hong Sung-chang, chief executive of the drama's production company Studio S, said producing the sequence with AI reduced costs by more than 60 percent compared with filming it conventionally.
The savings mattered, Hong said, but so did preserving scenes that otherwise would have been abandoned because they were too dangerous, too expensive or simply impossible to shoot.
That distinction illustrates the role AI is beginning to occupy in Korean film and television. It is not replacing the industry's highest-end visual-effects work. Instead, it is filling the widening gap between what directors want to show and what producers can afford to film.
The Agent Kim sequence was not created at the push of a button.
Morpheus Studio spent about a month developing visual concepts and another two months producing the footage. The team generated roughly 1,000 still images and more than 2,000 video clips before selecting 289 shots for the final sequence.
Human judgment remained central throughout the process — from designing costumes and locations to selecting shots, editing the sequence and correcting color.
The production therefore challenges the assumption that generative AI simply removes labor from filmmaking. It may reduce the need for physical sets, location shoots, stunt work and some conventional visual effects, but it replaces them with a new workflow built around repeated generation, evaluation and refinement.
Traditional VFX, however, still offers greater control.
"With sufficient time and production resources, there is no visual element we cannot create," said Kim Gyu-min, a spokesperson for Dexter Studios, one of South Korea's leading visual-effects companies.
The amount of work varies dramatically depending on the shot, Kim said. Adding a single flying object to filmed footage bears little resemblance to constructing an entire computer-generated battle scene. The more complex and detailed the image, the more artists and production time are required.
Lighting illustrates the difference. VFX artists must reproduce the direction and intensity of light, along with shadows, reflections and atmospheric effects, so that computer-generated objects blend seamlessly into filmed environments.
Dexter treats lighting as a separate stage in its production pipeline, Kim said. Daylight scenes are particularly demanding because of the sharp contrast between illuminated and shaded surfaces.
A film director, who had not viewed the Agent Kim sequence but discussed generative AI more broadly, also questioned whether AI-generated footage could consistently blend into a longer dramatic production.
An AI clip may appear realistic on its own, the director said, but every sequence must match the color, camera movement, characters and physical environment of the scenes before and after it.
Viewers are especially sensitive to inconsistencies in brightness and shadow, making daylight CG and VFX particularly unforgiving. The technical capability of Korean VFX companies is rarely the limiting factor; production time and budgets often determine the final quality.
Yet VFX companies are not resisting the technology. They are increasingly incorporating it.
Dexter already uses AI for repetitive post-production tasks, including removing visible stunt wires. Work that artists once completed manually frame by frame can increasingly be automated, allowing them to devote more time to complex shots that still require human judgment.
AI is also opening possibilities for sequences that would otherwise require extensive sets, location shoots and large production crews, Kim said. It may not yet match the finish of a scene produced through conventional filming and post-production, but it can provide a practical alternative when traditional methods are no longer financially viable.
The emerging production model is therefore more likely to combine AI with conventional VFX than force a choice between them. AI accelerates visual experimentation and automates routine tasks, while traditional VFX provides the precision needed to control and integrate the final result.
The pressure to adopt those tools is particularly acute for broadcasters.
EBS President Kim Yu-yeol said total television advertising revenue has fallen from about 2.1 trillion won ($1.5 billion) two decades ago to roughly 550 billion won last year. During the same period, the production cost of a single drama episode rose from between 50 million won and 100 million won to between 2 billion won and 3 billion won.
In one AI animation experiment, Kim said, EBS reduced production costs to one-fortieth of those required under conventional methods.
For broadcasters facing shrinking advertising income and sharply rising production costs, AI is no longer merely an experimental technology. It is increasingly becoming a survival tool.
EBS has expanded AI-assisted programming, including the terrestrial drama Resurrection Class, while KBS has developed an AI television studio that automates camera tracking and plans to extend the system to news, interview and educational programming.
The economic and creative arguments are not mutually exclusive. Lower production costs can preserve scenes that otherwise would be cut, turning cost savings into greater creative ambition.
But the technology also introduces new risks.
Ryu said Morpheus Studio discovered that one AI model unintentionally generated the face of a real actor with no connection to the production. The team subsequently verified AI-generated characters through image searches to reduce the risk of using recognizable faces without consent.
The episode raises unresolved questions about how actors' likenesses enter AI training data and who bears responsibility when AI-generated characters resemble real people.
Audience acceptance also remains uncertain.
A December 2025 Hankook Research survey of 1,000 adults found that 49 percent felt uncomfortable with advertising or entertainment containing AI-generated images or video. The figure rose to 71 percent among respondents aged 18 to 29.
At the same time, 81 percent viewed the spread of AI-generated content as a natural consequence of technological progress.
The findings suggest audiences may regard AI's expansion as inevitable while remaining cautious about how it is used and disclosed.
The Netflix rankings and domestic television ratings for Agent Kim: Reactivated cannot determine whether viewers embraced its AI-generated sequence. It remains unclear how many even knew how it had been created.
What the production does establish is that generative AI has moved beyond demonstration videos and isolated experiments. It has entered mainstream television production, reached a global audience and delivered a sequence that producers say would have been prohibitively expensive to film conventionally.
For now, AI cannot offer the same precision and consistency as established visual-effects techniques across every type of scene. Nor does it eliminate the need for artists, editors or creative judgment.
Its first major role in Korean television may instead prove more pragmatic: not replacing images the industry already knows how to create, but making possible the ones it can no longer afford to shoot.
Copyright ⓒ Aju Press All rights reserved.