Hengdian and ByteDance's Volcano Engine jointly build the "Hengdian π" AI platform; the rumor that 150,000 extras' data was fed to AI is refuted by Doubao

Hengdian has once again been pushed into the eye of the public-opinion storm. On October 28, a rumor began to circulate on the internet and quickly fermented. The rumor claimed that Hengdian would cooperate with ByteDance to use the likenesses, voices, and even acting skills of the local 150,000 background actors as corpus data to "feed" artificial intelligence large models. On the same day, Liu Xing, public relations head of ByteDance's Doubao AI large model, posted that the rumor was "seriously inaccurate." The source of the rumor is not hard to infer: at the 2026 Hengdian Film and Television Culture Industry Development Conference, which opened on October 23, Hengdian announced that it had reached a cooperation agreement with ByteDance's Volcano Engine to jointly launch the "Hengdian π" AI service platform. At the conference that day, Xiao Ran, general manager of solutions for Volcano Engine's digital intelligence platform, said when introducing the platform: "Hengdian has more than 50 real-scene bases, more than 2,300 scene locations, and more than 150,000 registered background actors. More importantly, film and television production and creation are genuinely happening here every day, which will provide sustained, high-quality material for the development of film and television large models." The remarks by the relevant head of Volcano Engine were interpreted as "refining" the data of 150,000 background actors to "feed" AI large models. However, behind this quickly debunked rumor lies a reflection of the anxiety about "AI replacing people." What is more, in a certain sense, "AI replacing people" is already playing out for real in the continuously booming short-drama industry. According to statistics, from January to May this year, the market size of domestic AI dramas and comic dramas reached 22 billion yuan, and for the full year it is expected to exceed 40 billion yuan, a year-on-year increase of about 138%. Data released by the National Radio and Television Administration show that in the first eight months of this year, 430,000 micro-short dramas went online nationwide, 13 times the total for all of last year, of which AI dramas accounted for more than 90%. If the pace of short-drama development in recent years was already astonishing enough, then this round of AI drama fever after this year's Spring Festival can almost be described as "crazy." AI dramas are advancing triumphantly, while live-action short dramas have been forced into a corner. As "the winds of change" increasingly become a consensus, the small city of Hengdian, known as the "Oriental Hollywood," has become the most concentrated sample for observing an industry undergoing drastic change. Three years ago, the explosion of vertical-screen short dramas turned Hengdian into "Verticaldian" overnight. Facing the even more ferocious momentum of AI dramas, people cannot help but wonder whether "Verticaldian" is about to change course once again and transform into "AI-dian." Hengdian. Photo by Yu Liang. Hengdian's answer. "Before our crew came in, this set had not been used for more than half a month." On October 29, the reporter met producer Li Kangsheng at a filming location in Hengdian. That day, the vertical-screen short drama he was working on, a period costume family-strife drama, was about to wrap. The entire shoot lasted seven days, with a total investment of about 1.1 million yuan. "If a production of this specification had been made last year, the investment would generally have been around 1.3 million yuan, and filming would have required at least eight days. The impact of AI short dramas is simply too great. This year everyone is tightening their belts, budgets are limited, and we save wherever we can," Li Kangsheng paused and said, "Or we simply do not shoot at all." Live-action short dramas are experiencing a cliff-like collapse. A few months ago, such a judgment might still have drawn skepticism in Hengdian, but right now it is a cold reality that is visible to the naked eye and impossible to dispute. According to the crew updates published weekly by the official Hengdian World Studios, compared with the same period last year, the number of vertical-screen crews starting production in Hengdian in the third quarter of this year plunged 82.5%. Even in August, when the volume of starts was relatively high, the peak number of crews filming in a single week was only 24, whereas last year that number was 106. This is clearly not an ordinary fluctuation in the industry cycle. The explosive growth of AI short-drama production capacity has dealt a devastating squeeze to the living space of live-action short dramas, and industry capital has accordingly "yanked the steering wheel." At this time last year, live-action short dramas were still constantly creating miracles of traffic and wealth; now they have seemingly become "outdated goods." With the industry in upheaval, Hengdian, as a long-established real-scene film and television shooting base and a major film and television industry town in the Yangtze River Delta, must take action. The first question that needs answering is: when the production of audiovisual products itself no longer necessarily has to take place on a physical set, what else can Hengdian offer? The "Hengdian π" AI service platform, which triggered the rumor about "refining 150,000 background actors to feed AI," is one of the answers. At Zhejiang Hengdian World Studios Co., Ltd., the reporter chatted with a relevant person in charge about the rumor that completed the entire process of "appearance—spread—debunking" in less than a day. The person immediately put a hand to their forehead, face full of helplessness: "Let's not talk about anything else first. The total permanent population of all of Hengdian is only just over 220,000. Where would 150,000 background actors come from?" The figure of 150,000 is the cumulative total number of actors registered with the Hengdian Actors Guild since its establishment in 2003. According to this person in charge, over the past more than 20 years, "Hengpiao" have changed batch after batch, and the background actors who remain active in Hengdian over the long term have consistently stayed at around 10,000. The vast majority of these 150,000 people were merely passing through in a hurry, and the "image data" they left behind may be nothing more than a registration photo, or a glimpse measured in frames in some film or television drama. Setting aside the feasibility of data collection, even if one truly considered it from the perspective of "refining" background actors, such data would not actually be of much value. However, building a digital background actor dataset is indeed one of the goals of "Hengdian π." The project tender document explicitly states in the digital background actor dataset section that all data "must be collected with lawful authorization, with a clear authorization chain and clearly defined boundaries of use." At the same time, the scale of the dataset "must be able to support vertical model training or intelligent agent service needs." In other words, "refining" background actors is indeed in the plan, but the targets of "refining" are not all "Hengpiao" indiscriminately, nor are there as many as 150,000, and the refined data will not be applied to ByteDance's general-purpose large models. "The project only completed bidding on October 13, and delivery is planned in phases over the next three years. Everything being discussed about this project right now is just speculation." That was the summary from this person in charge. However, based on the project tender document, one can try to infer Hengdian's intentions. "Hengdian π" plans to build no fewer than 9 high-quality industry datasets, with no fewer than 50 billion tokens of pre-training data, covering core directions such as real film-base locations, digital background actors, costume and prop supply chains, and period makeup and styling. This means Hengdian is trying to systematically digitize a series of assets that in the past depended on physical space and were reset to zero as film crews came and went, turning them into model capabilities that can be rights-confirmed, invoked, and repeatedly traded. And these film and television assets accumulated over the years are also the biggest trump card in Hengdian's hand. The answer is almost obvious: in the face of the surging AI wave, Hengdian no longer intends to be merely a "landlord" renting out soundstages and sets, but is trying to become a supplier of data and models in the AI film and television production chain. Short-drama shooting site. Photo by Yu Liang. Coasting on inertia. While betting on AI, Hengdian does not intend to abandon the fundamental base of live-action shooting. In July this year, Hengdian World Studios publicly released a series of preferential policies for film crew shooting. Under the policies, all rent is waived for the more than 130 soundstages under Hengdian World Studios with a total area of more than 500,000 square meters, as well as modern-drama sets with a total area of more than 200,000 square meters, and the price of hotel rooms for film crews has also been lowered accordingly. At the 2026 Hengdian Film and Television Culture Industry Development Conference held on October 23, Huang Leiguang, deputy general manager of Zhejiang Hengdian World Studios Co., Ltd., once again emphasized this policy in his speech, saying that the move is intended to activate the "fundamental base" and enable traditional live-action crews to "spend less and produce more." Huang Leiguang's remarks carried a somewhat tragic tone: "During a period of industry adjustment, whoever is willing to bear the sunk costs of infrastructure is safeguarding the elasticity of production capacity for the entire industry. Hengdian chooses to do something the industry cannot do without, so that creators can still afford to shoot at the bottom of the cycle." Amid the "period of industry adjustment," Hengdian is also trying to graft the fundamental base onto AI. Huang Leiguang said that in the AI era, the iterative upgrade of film and television content toward hybrid production will become an inevitable trend. To this end, Hengdian will also empower live-action real-scene + AI hybrid shooting, with specific measures including that crews' live-action shooting spending can partially offset computing power and token costs, and conversely AI spending can also partially offset live-action shooting costs. "Our goal is very clear: let pure AI creators settle in Hengdian with peace of mind, let hybrid-shooting crews complete their shoots in Hengdian more efficiently, and let small crews enjoy digital-intelligent production technical services that are reliable in price, technology, and artistry," Huang Leiguang said. However, judging from the current situation alone, Hengdian's efforts are not enough to reverse the precipitous decline of live-action short dramas. Crew updates released by Hengdian officially on October 28 showed that only 18 vertical-screen crews were shooting that week. The people inside the industry can best feel the piercing chill. Xiao Wan, a producer based in Hengdian who works for a film and television production company in Hangzhou, told reporters that she handled as many as more than 120 short-drama projects last year, but this year that number has plummeted to fewer than 40, and a considerable portion of them are low-cost productions with investments of less than 200,000 yuan. Speaking of the roller-coaster experience of the past two years, this 28-year-old young producer smiled bitterly: "Last year I still thought the good days had just begun. I never expected that this year the good days would already be over." Compared with the obvious drop in income, Xiao Wan has also gradually begun to doubt whether she can keep her job. Starting in May, Xiao Wan began hearing news of resignations from acquaintances in the industry, and the company she works for has also cut some positions. It was also around this time that Xiao Wan noticed that the AI short dramas she occasionally came across had "a noticeably smaller sense of incongruity" compared with a few months earlier. What also happened at the same time was a shift in company strategy. Xiao Wan said that her company has now put nearly 70 percent of its manpower into the development and production of AI dramas: "Our company began testing the waters with AI dramas last year. At that time, we used money earned from live-action dramas to subsidize AI dramas. Now, instead, we have become the 'money-losing goods.'" Li Kangsheng's "good days" also came to an abrupt end this year. In March last year, after finishing several long-drama projects on hand, Li Kangsheng plunged into the booming short-drama track. In nine months, he handled no fewer than 70 short dramas, "and more than a dozen became hits in one go. To be honest, I made quite a lot of money." In Li Kangsheng's view, live-action short dramas are now in the "long-tail period" after an explosion. Even though output has fallen sharply, relying on the huge momentum accumulated last year, the industry is still continuing to "coast" under the effect of inertia. How long it can continue to "coast" in the future depends entirely on the face of its rival, AI dramas. Once large-model capabilities improve further and AI drama production capacity expands further, the position of live-action short dramas may be further lost. "Investors are the most realistic. Wherever there is money to be made and wherever there is heat, that is where they go. And what is hottest right now is, without question, AI dramas." "People cannot change the environment." As for the future, Li Kangsheng has no clear plan and cannot see the exact direction. He intends to follow the industry and continue to "coast," waiting until the inertia driving that "coasting" is exhausted before making plans. People cannot change the environment, and neither can Hengdian. In a sense, preserving the fundamental base is also a necessary move for Hengdian as it shifts from a "vertical-screen shop" to an "AI shop." Indeed, as a relevant person in charge of Volcano Engine said at the Hengdian Film and Television Industry Conference, what is "more important" is the second half of the sentence—film and television production and creation are genuinely happening here every day, and this will provide sustained, high-quality material for the development of film and television large models. AI, too, needs Hengdian to have productions shooting. Hengdian Digital-Intelligent Film and Television Innovation Incubation Center. Photo by Yu Liang. Yesterday Once More. In August this year, the Hengdian Digital-Intelligent Film and Television Innovation Incubation Center was officially inaugurated. Fully funded and independently operated by Hengdian World Studios, the center focuses on new AI-driven film and television business formats, providing incubation services to startups and creative teams working in areas such as AI short dramas, AI comic dramas, digital assets, virtual production, and post-production. As one of the first companies recruited to the center, Sang Li and her team arrived in Hengdian last month. After years of grinding it out in the traditional film and television industry, Sang Li resolutely chose this year to pull out of the live-action short drama track. After arriving in Hengdian, her goal was clear: go all in on AI dramas. At present, Sang Li's company has reached a monthly production capacity of 1,000 minutes, focusing mainly on overseas markets. "Four people can turn out a drama in three days." Sang Li's partner, director He Zhenyu, is not satisfied with this result: "This is just the beginning. Production capacity needs to be pulled up to at least 10,000 minutes a month before we even qualify to 'sit at the table.'" He Zhenyu said that the AI short drama market is still in chaos, and the audience's aesthetic system has not yet taken shape. Against such a backdrop, for a single AI short drama to become a hit, the only option is to rely on economies of scale—"gambling on probability" by "exploding production capacity." "The cost of a single live-action drama now starts at at least 200,000 yuan, while the cost of an AI drama can be kept to under 10,000 yuan at the lowest. If you're rolling dice either way, rolling one die at a time versus rolling 20 dice at a time—how do you think you should choose?" This is how Xiaowan explained the logic of capital abandoning live-action dramas and turning to AI dramas—an expression that sounds familiar—in 2023, the boss of a film and television production company in Hengdian made similar remarks when facing reporters. At that time, it was precisely the beginning of the wild growth of vertical-screen short dramas, and a large number of low-cost short dramas quickly filled the market in a short period. "The path that live-action dramas have taken in recent years, AI dramas will definitely walk again," Li Kangsheng said. "It feels a bit like yesterday once more, except the interval is a bit short." Indeed, it is a bit short. A person in the film and television industry lamented to reporters that the film and television industry has always been known for stability, and the dramatic ups and downs of the industry in recent years are unprecedented. However, after experiencing the frenzy brought by short dramas, everyone seems to have adapted to this high-speed, ferocious rhythm, so much so that they are not surprised by the current round of drastic changes. "The only thing that may surprise everyone is probably that they didn't expect the one slamming on the accelerator would be AI," the industry person said. Three years ago, regarding short dramas, this "new species" in the audiovisual field, people in Hengdian did not have a unified attitude. Some embraced it wholeheartedly, some were full of doubts, and some stood by watching. Now, facing AI dramas, this "new new species," people remain the same. Only, with the experience of short dramas, more people tend to believe that this is the general trend. He Zhenyu is firmly optimistic about AI dramas. As a director, switching the object of his directing from live actors to AI did not cause him any discomfort. In his words, it was "completely painless switching," "like replacing a fuel car with a new energy vehicle." He Zhenyu believes that given the "semi-automatic" production method of AI drama industrialization, its supply will further expand in the future and will ultimately become the mainstream of online audiovisual products. Sang Li, on the other hand, values the low cost of AI dramas more. She feels that this characteristic can allow film and television companies to operate more nimbly, thereby gaining more room to maneuver in the market. Xiaowan said she has recently been seriously considering a possible job transfer. The focus of her company and the industry has already shifted comprehensively to AI dramas, and if she wants to keep her job, this is very likely a choice she must make: "It's not that I couldn't do it if I were really asked to. But compared with sitting in front of a computer all day, I'd still rather deal with one living person after another on set." Li Kangsheng, however, firmly believes that AI dramas are the future. According to him, "AI replacing people" is happening right now. Not long ago, Li Kangsheng bought a motorcycle he had long coveted. That day, he rode the motorcycle "running the mountains." After reaching the mountaintop, looking at the scenery before him, Li Kangsheng naturally took out his phone to take pictures, planning to post them on Moments. Then, he just as naturally opened Doubao on his phone and asked Doubao to design a Moments caption based on the photos he had taken. "At that moment I suddenly realized that AI has already thoroughly entered our lives," Li Kangsheng said.