5 months, a full-length AI drama, the established industry giant saw its market value skyrocket by 11.6 billion yuan in just 2 days.
The boundaries of the film and television industry are blurring, and all future film and television companies may evolve into technology companies.
On October 4, *Battle in the Heavenly Palace*, the second chapter of the first season of *Later Journey to the West*, China's first full-length AI drama, was officially launched on Mango TV and Hunan TV.
Although some audiences commented that the micro-expressions of the characters still have a "plastic feel", the public praise for this round is significantly better than that of the first chapter. In the drama, the "workplace" in the underworld has been frequently discussed by audiences. The judge evades responsibilities with ambiguous remarks, and the king of hell passes the buck. When facing the questioning of the protagonist Sun Xiaosheng, they can only reply "I don't know, I can't remember clearly, I'm just an employee here". That night, a topic popped up on social platforms: The screenwriter must have had real work experience. Some users wrote: "All the staff in the underworld pass the buck and know nothing about the case, which hits a strong chord with every ordinary office worker." Some regard it as a slice of real life: "Sun Xiaosheng has such a hard time investigating the case, it is exactly like looking for people to go through formalities in the workplace."
The positive public praise has sustained the popularity. According to Hunan TV's disclosure, the drama ranked first among provincial satellite TV stations in both Kuyun and Huanwang's simultaneous segment ratings on its premiere night, and the total number of full-episode views on Mango TV exceeded 320 million. Data from Meilande shows that the total cross-platform spread volume of the two rounds of broadcasts has exceeded 4 billion views, and the drama has been on the national hot search list 311 times. On October 9, Mango Super Media hit the daily limit again after the National Day holiday reopening, with a daily increase of 19.98%.
A month ago, the opening of the first chapter *Mountain of Flowers and Fruits* was even more explosive: On the premiere day of August 31, the share price of Mango Super Media hit the 20% daily limit only 16 minutes after the market opened, and the company's market value increased by about 11.65 billion yuan in two trading days. Mango TV revealed that the second chapter was originally scheduled for the winter vacation, but was advanced to the National Day holiday due to continuous audience requests for earlier release.
Shortly before the launch of the second chapter, *China Entrepreneur* met with chief director Li Dongshen in an industrial park near the East Fifth Ring Road. It was Saturday that day, and there were few people in the park, but as soon as we walked into the office, we could feel the intense, fast-paced atmosphere. For this phenomenal work, he admitted that under the broadcast-while-reviewing model, the team has been racing against time.
*Later Journey to the West* has extraordinary significance for him. After 20 years of making historical documentaries, he is the creator of *The Hexi Corridor* and *China*. However, in the traditional film and television industry system, it seemed unreachable for a documentary director without traffic resources to shoot a drama. The cost of a single episode of platform S-level drama often reaches tens of millions of yuan, "We can't gather the 'full plate' of resources, including funds, core creative teams, actors and other project resources, nor can we get the lineup of top-tier artists."
AIGC is a path that emerged unexpectedly. He put forward the idea in March, got the project approved in May, and launched the work on August 31 — Li Dongshen built a brand new "production line" that no one had ever walked on from scratch. In April this year, many people in the industry predicted that the first hit full-length drama produced purely by AI would be released as early as this summer, no later than this autumn, and he was determined to seize this time window.
He defined this season as a "limit test": to test the real capability of the model in long-form narrative, to test the market's acceptance of AI full-length dramas, and more importantly, to test the production limit of the team. "Producing 5 episodes within a month is an extremely challenging task," he said, but the "0 to 1" breakthrough is of great value, that is, to create the first AI full-length drama in the real sense.
But what made this goal achievable and profitable is a time-honored company that has to find a new growth pole. At the beginning of this year, Mango established the AIGC Innovative Content Center, as a first-level business center with independent project approval rights, production rights and budget rights, and all workflows run on the self-developed AIGC content creation platform "Mango Creative AI".
*Later Journey to the West* can be regarded as its first released work. Cai Huaijun, Chairman of Mango Super Media, said that the value of this drama "lies not in the single hit of one work, but in opening up the second growth curve for content supply for the company".
This is not an isolated attempt. On September 17, at the 2027 Mango Co-creation Ecological Conference, "Mango Creative AI Theater" was released as a high-quality AIGC content label, covering three tracks: revitalization of classic Chinese-style IP, high-quality mid-to-long dramas for female audiences, and high-quality children's animations. *Later Journey to the West* is the flagship work of the Chinese-style track; *Eternity: The Life of Xin Zhui*, which premiered on September 30, is based on the archaeological findings of the Mawangdui Han Tomb, and restores the life of Xin Zhui, the wife of Marquis Dai of the Western Han Dynasty in 40 minutes.
According to Ren Xu, Assistant President of Mango TV, Deputy Chief Editor of the Editorial Board and General Manager of AIGC Innovative Content Center, there are more than 20 reserved high-quality projects, which will be released successively from the fourth quarter of this year to the first quarter of next year. She said that what she wants to prove to the market is not just one single drama, but Mango's capability of continuous high-quality content supply.
The industry trend is gradually becoming clear, and platforms are accelerating their layout in the AI full-length drama track at the same time. On September 20, Tencent Video launched the first "AI fantasy drama with real actors" *The Escort*; even earlier on August 22, iQiyi's *Soul Ferry: Floating Life Dream* was exclusively launched, and its split-box office revenue exceeded 6 million yuan in 10 days. All major platforms are increasing their investment in the technical side, and have launched their own AI workflows and video models one after another.
"In my opinion, the boundaries of the film and television industry are blurring," Li Dongshen said, "All film and television companies in the future may become technology companies."
The Death of the Camera
In 2023, Li Dongshen was still an "opponent of AI", and generating videos with AI was even a fantasy to him.
That year, he was going to produce the third season of *China*. Before starting, a problem had been weighing on his mind: the first two seasons covered the history from the Spring and Autumn Period and Warring States to the 1911 Revolution, and the characters could be played by actors. The third season will focus on ancient myths and the Xia, Shang and Zhou dynasties, including the stories of Pangu creating the world and Nüwa mending the sky — who can play these roles?
On a youth painter program of Hunan TV, Li Dongshen came up with a new idea: to gather these young people who have received systematic art training and have contemporary creative expressions, to draw the historical scenes in imagination, and then make these paintings move. At that time, AI image generation had just emerged, and video generation was rarely mentioned by people.
He invited the painters to draw for nearly a year. After the film was broadcast, a friend commented on him: "You have subverted all the characteristics of traditional historical documentaries." He didn't realize at that time that this attempt had already stepped into another dimension. "Looking back today, what I did at that time was exactly the logic of generating videos from images."
The third season of *China* adopted the technology of painting plus CG (computer animation). After the production was completed, Li Dongshen felt that he had tried everything possible in historical documentary production. He described that period as "very satisfying but also very lost", not knowing where to go next.
The turning point came from a creative impulse: he really wanted to shoot a documentary about the history of Taiwan, which later became *Chronicle of Taiwan Island*. He is very familiar with the subject, but the biggest difficulty lies in the visuals: naval battles and large-scale historical scenes cannot be achieved by live shooting, and traditional CG is too expensive to afford. The Mango team proposed to try it with their own AIGC platform.
In March 2025, he started to use AIGC systematically for the first time. "The technical capability at that time was just like the Stone Age, everything was made completely manually." His solution was stylization: taking the visual texture of oil painting as the keynote, and expressing historical situations with a sense of grandeur and surrealism, which made up for the "lack of visual details" to a certain extent.
But it is also because of this project that he changed his judgment on AIGC: "Suddenly, I had an epiphany, and it felt like my soul was connected to the model."
Li Dongshen realized that the creative logic of AIGC is completely different from traditional live shooting, which he called "frame-set thinking". He said that the first thing to do is to "throw away": throw away camera movement, camera position design, and montage editing — these are the core capabilities of traditional director training. The alternative solution is to converge layer by layer: first grasp the overall composition, determine the scene list of the whole film and the position relationship of the characters; then set the scene map and scheduling of each scene; finally, lock the first and last frames of each single shot.
"I don't care much about how the camera moves between shots, what I care about is the overall composition. Once the key frames are set clearly, the movement in between will be completed by the model."
Until the beginning of 2026, video models represented by Seedance 2.0 completed a capability leap. His judgment is that "it is fully capable of producing full-length dramas". The day after the Lantern Festival, he called the leaders of Mango TV and said he wanted to make an AI full-length drama.
At that time, the mainstream impression of AIGC in the industry was still "low cost". It was the battlefield of micro short dramas, and most of the hit works were short films with a cost of several thousand yuan, and the whole industry was competing for lower costs. He wrote a long letter to the leaders of Mango TV, the core point of which is: the workload of making short dramas and full-length dramas with AIGC is not at the same level at all. To produce AI full-length dramas, nearly 100 people need to work together, and the degree of connection, the complexity of worldview and art design are completely incomparable.
From the time he made that call to the official project approval in May, only two months passed.
When choosing the script, he ruled out realistic themes — audiences may not accept it for the time being, and the model is not ready either. Among the targeted "Journey to the West" IP, the original work is too classic, and adaptation is very difficult and may not be well received. Finally, he chose the Ming Dynasty novel *Later Journey to the West*, which leaves enough adaptation space for its worldview, and readers are not familiar with its content.
The reason why he could get the resources so quickly is that Mango was also "anxiously" looking for the same thing — the full-length drama market is shrinking. According to Yunhe Data, the total effective views of the top 20 new full-length dramas launched in 2025 reached 29.6 billion times, a year-on-year decrease of 20%; in the same period, the average daily usage time per capita of micro short drama applications reached 129 minutes, exceeding that of long video platforms.
It is getting harder and harder for long video platforms to make money. The number of paid members has reached the ceiling, advertising investment is under pressure, but the cost of content production cannot be reduced, so the company needs a new increment.
AIGC is the answer at the current stage. According to Cai Huaijun, Mango's AI layout has been laid out in four levels: from the early infrastructure of computing power and cloud platform, to the launch of AI 1.0 since 2019 to make machines understand content, then to the self-developed Mango Creative AI platform of AI 2.0 to make machines participate in creation, until this year AI was promoted as the core strategy and the AIGC Innovative Content Center was established.
At this node, Li Dongshen and *Later Journey to the West* appeared exactly at the right time.
The Film Crew Without Cameras
The crew of *Later Journey to the West* barely overlaps with traditional film crews.
More than 100 people are divided into four or five teams: Li Dongshen himself is the chief director and core screenwriter; the production team has only two people; the art team has more than 30 people, responsible for character design drawings and scene design drawings; the largest group is more than 60 "card drawers", which he prefers to call "AIGC directors", who use the model to generate images; there is also a team of algorithm engineers following the whole production process. When the director puts forward demands, the team will develop corresponding tools, which can be launched in about a month. Less than 30% of the people in the whole team studied film-related majors.
Under the tight schedule, Li Dongshen's most important work is to empower these AIGC directors with the "frame-set logic". Traditional directors think about how the camera moves and where to place the camera position. Now they need to figure out every frame of the picture first: where the character is located, where the light comes from, and what state the emotion is in. The model is neither a camera nor a modeling software.
In the first few months, his main work every day was to review and modify pictures; a rival scene with only two characters was revised for 40 to 50 versions before finalization; the success rate of generating qualified images was only 1 out of 16 at the beginning, and later increased to 1 out of 5. The team produced more than ten episodes of content, overturned them repeatedly, and finally only kept 5 episodes for launch.
Details problems emerged in endlessly. At the end of the fourth episode, Sun Xiaosheng grew up, and the team found that the movements and expressions of the generated characters were very stiff — with the same prompt words and the same model, they could never reproduce the vividness of the character in his childhood. It took nearly ten days to find the reason: there was no tail in the initial character design drawing, so the model defaulted that this was a "human" playing a monkey, and the facial soft tissue could not move, resulting in stiff expressions and subsequent stiff movements. After adding the tail, the problem was solved immediately.
The pictures can still be adjusted. Li Dongshen said that the visual consistency problem has been basically solved now, and the difficult part is "emotional continuity": one shot can be extended to 30 seconds with fluctuating character emotions, but when it comes to the next 30 seconds, to continue the previous emotional logic, the whole setting needs to be rebuilt. He has put forward suggestions to the model team, hoping to develop a tool with "memory function".
The model also needs to be "trained". For example, when the monkey was talking, the mouth shape once could not match the voice, because the general tool defaults to human face, which does not fit the monkey's face at all. The team reported the problem to Mango Creative AI, and the technical side quickly adjusted the algorithm to make a solution that can