Bona Film's AI movie "Sanxingdui: Future Past" enters theaters, tens-of-millions cost, first-day box office 317,000 yuan

At the premiere in late October at the CINITY giant-screen auditorium of the China Film Cinema in the Museum of the Communist Party of China in Beijing, a scene never before seen in the history of Chinese cinema unfolded: the creative team stood on stage, and seated in the audience were viewers, media, and film critics—but notably, no "actors."

"This is my first time attending a premiere with no actors; it's a bit awkward," said well-known film critic Tan Fei half-jokingly from the audience.

Standing on stage were Jiang Defu, CEO of Bona Film Group, and Qu Ji Xiaojiang, a renowned producer and head of Bona Film Group's AIGMS Production Center. Behind them, on the large screen, was the 100-minute theatrical feature titled "Sanxingdui: Future Past."

On October 23, "Sanxingdui: Future Past" opened in theaters nationwide. It had a 2.9% share of first-day screenings and grossed 317,000 yuan. By the evening of October 25, cumulative box office was about 1.18 million yuan. Its production cost was in the tens of millions.

Jiang Defu experienced a similar scene more than twenty years ago, when Chinese cinema was transitioning from film to the digital era. This time, Jiang Defu was equally candid: "When we weren't yet doing that well, audiences were already coming into the theaters. As long as someone watches, that is encouragement and affirmation for industry innovation."

To understand "Sanxingdui: Future Past," one must first understand the circumstances in which this film came about.

In 2022, relevant parties in Sichuan extended an invitation to Bona, hoping to spread "Sanxingdui" culture through film—after all, too many people nationwide and worldwide cannot travel to Guanghan or visit the Sanxingdui Museum.

This was a subject so grand that it made one "deeply apprehensive." Jiang Defu did the math: to make such a film recreating ancient Shu culture and history with live-action shooting, "you can't do it without 300 million."

For a film company, as long as the money is in place, everything can proceed step by step; the hard part is never "what to shoot," but whether anyone dares to invest. During this period, China's film industry was already in a cycle where "no one dares to invest money easily anymore." After the capital bubble receded, the industry was relatively passive.

In early 2024, when OpenAI released the Sora demo, the visuals had a direct cinematic quality. She predicted, "This thing should happen very quickly," and so "without any hesitation" began attempts to use AI to produce film and television works.

Jiang Defu's account seems closer to the real starting point of this work.

Because to realize a Sanxingdui-themed work using live-action shooting, you would need at least location shoots, and you couldn't rely entirely on VFX shots. Locations, actors, a large amount of CG… all told, it was a massive project, and after filming, would audiences buy it? They weren't sure.

They first decided to try making a short drama with this subject. In 2024, the AI short drama "Sanxingdui: Future Apocalypse" got 200 million views on Douyin, and the film project was put on the agenda.

This decision itself conforms to the laws of the film industry. Validate a subject with low cost, then use the validated subject to leverage greater investment. But the problem is that being watchable as a short drama and being watchable in theaters are separated first by a chasm in production standards.

"Originally we wanted to run along with the technology and see, but it turned into us leading the race." Jiang Defu described the situation this way: "There's no one ahead of me. I've taken the first step, and I can't stop, so I just keep running."

In Jiang Defu's view, this journey was especially grueling.

The first layer of difficulty came from the unpredictability of large-model technology itself. Model progress is not a linear process; the producers never know whether tomorrow a more stunning model will emerge and overshadow months of hard work. For a film in its production cycle, AI's "progress" is like a curse—stuck in a technological upward channel of "it can always get better."

Bona's AI team later set the tone for this: there must be a deadline, "leave as few regrets as possible, but it must meet the review standards of the Film Bureau." Otherwise, you could only foolishly wait for large models to update every two months, never finish, always feel it could be better, and the film would never be released.

Another difficulty is the consistency problem of video models, the core issue that all AI video today cannot avoid. The same character may look one way in the first five minutes and change when the shot changes; in the same scene, facial texture under the light source may be inconsistent from one moment to the next. It may be passable on a phone, but on the big screen the flaws are infinitely magnified.

Image source: Official still from "Sanxingdui: Future Past"

The numbers can illustrate the extent of this "hand-crafting." "Sanxingdui: Future Past" is 100 minutes, with about 2,000 shots, and cumulatively produced more than 1.2 million base images, with a usable-to-discarded ratio of nearly 600:1.

For one usable shot, the team often had to generate and filter out hundreds or thousands. At the most exhausting times, they completed only one shot a day—no different from traditional filmmaking.

The team numbered nearly a hundred people, a standard film crew configuration. They also had to recruit and train new people, and figure out some new positions that simply don't exist in traditional crews, including "consistency verification" (specifically watching to ensure characters don't change faces from shot to shot) and digital character casting (which is not the same as live-action casting).

When model companies start entering the set

On the cinema poster, Shengshu Technology's Vidu is listed as "Chief Technical Support." From the experiences of Jiang Defu and Qu Ji Xiaojiang, one can sense that the role model companies played this time went far beyond simply providing API interfaces.

Jiang Defu said: "Shengshu Technology gave us a lot of support, not only in computing power but also in video generation, providing a lot of technical guidance."

He added one more line: "In the future, when you make a film, there should be a large-model company and a video-generation company behind it."

Jiang Defu, CEO of Bona Film Group

The crew's early workflow used images to control the model's hallucinations and consistency issues—for example, building training sets with ComfyUI—and at that stage, as long as there were enough GPUs, token consumption was not that high. But as generation duration and complexity rose, the structure of compute consumption changed. In this process, Shengshu Technology provided not only compute but also specialized support for film-grade requirements.

General video models are designed to generate "good-looking short videos," whereas film requires shot language that can be connected, controlled, and reproduced. The gap between the two cannot be filled by the film team unilaterally "drawing cards."

According to Shengshu Technology, they set up a service team for the crew built around the underlying technical engine, dedicated to solving technical problems at specific points. The two sides worked together to complete production from storyboards and frames to video generation, building the underlying architecture for industrialized AI filmmaking.

Technology can only handle image generation and execution, while story ideas, scripts, shot design, and aesthetic judgment are still completed by the human creative team. More critically, character consistency, shot continuity, and image precision for large-screen projection still require extensive manual adjustment. Shengshu Technology's presence gives the model layer an interface that can respond to film-grade requirements.

Model companies seem to have changed from mere tool suppliers into "co-producers," which not only changes the production workflow of the film industry but also changes the producer's role and cost structure.

Previously, producers did not need to worry about technical issues; even if a hard drive broke, there was no need for excessive concern. Now, GPUs have become an important means of production. GPUs are not only a core factor of production—they hold the film's most core data—but their prices have kept rising, making them both expensive and hard to buy.

In Jiang Defu's view, traditional crews are "easy to manage." Even for the largest projects, the director manages the key creatives and the production manager manages everyone else: "As long as the money is in place, he can shoot every day." AI crews are highly unstable: "What you want, it can't give you; what it gives you, you may not think is good enough."

Interestingly, they unanimously take a negative view of "super individuals" and OPC's future in theatrical films.

Short dramas can be hand-made by individuals, but on the big screen they are basically unwatchable. AI production, by contrast, demands more of each individual: everyone in an AI crew has to be a director, because each person is responsible for a certain block of AI content, and without a director's mindset, the handoffs within the team won't connect.

"This is not something a model company can train. A director's feel for film—what he wants may be an outpouring of emotion, a micro-expression, or the hope that the audience can discover the story behind the eyes here," Jiang Defu said.

Cost and box office: narrative problems cannot be cured by technology

Qu Ji Xiaojiang, head of Bona Film Group's AIGMS Production Center

A script can show a certain appearance, but once it actually becomes a film, more problems emerge: whether the pacing works, whether it can make audiences buy tickets for it.

The technical team put its effort into the hardest places and indeed made visible progress. But a film is not a tech demo show; most audiences buy tickets and go to the cinema ultimately to see story, characters, and emotion—the narrative grammar that has been running for more than a hundred years.

Jiang Defu repeatedly said, "This is a test of the waters." The key significance is not how much box office it earns, but rather that once this new technology pipeline is connected and proven, they will know how to empower the film industry with new technology going forward.

"Getting the production process to run through—that is the most important thing. As for the box office, we don't calculate that much. We don't fight the market, and we don't fight the audience. If you force audiences to watch, if you force the market to give you screenings, I think that's meaningless. But after all, we have taken the first step," he said.

The term "run through" is favored by tech companies, and here its exact meaning is that from project approval, production, review, scheduling, distribution to release, the whole process was walked through completely for the first time.

The film was submitted for review last September, and Jiang Defu said that at first they did not even think it would pass review smoothly.

Someone told Jiang Defu, "You're showing it on such a big CINITY screen—you've got real nerve," and his response was: "You have to honestly show everyone your cards; you can't deceive the audience, right? You must find the venue with the best projection quality to show your film and expose its flaws."

Jiang Defu is well aware that AI production is completely different in quality from live-action and real-location shooting, and its market appeal is different too. In the early stage, they did not spend excessively on marketing, and on pricing, he especially hoped cinemas would not set prices too high; in the revenue-sharing ratio with theater chains, they also significantly lowered their own share.

The film's settlement ticket price is 20 yuan, while Beijing screenings generally range from 30 to 50 yuan, with some auditoriums even exceeding 100 yuan.

Ticket prices are ultimately decided by cinemas, and the production side cannot control them, because this involves individual cinemas' rent, utilities, labor, and other operating costs. But Jiang Defu called on theater chains to lower ticket prices a bit so that more audiences can enter the cinema, which will bring more market feedback.

As for audience feedback... an interesting review seen so far is that if it were watched for free on a streaming platform, it might become a legend, but at this stage they would not accept paying to watch it. Some viewers also offered a compromise judgment: "If you can get a ticket under 30 yuan, it's still worth going to see."

This puts "Sanxingdui: Future Past" in a somewhat awkward position: it is not good enough for people to overlook the ticket price, nor bad enough to have no viewing value at all. The first work has the significance of a technological milestone, but most audiences are currently not paying for milestones.

Conclusion: This Has Happened Before

Jiang Defu often recalls that more than 20 years ago, when he was promoting digital cinema at China Film, he personally witnessed China's first digital film "Heavenly Grassland" (2002) enter theaters. Judging by its more than 800 reviews on Douban, probably not many people have seen this film.

"At that time, we had to promote digital equipment and install machines in cinemas. Many cinemas refused to install them, saying, 'I won't install it, I won't do it, I'll just show film.'" Theater chain managers repeatedly said, talk about when digital arrives later; anyway, I'm not showing it now.

At the time, a digital projector cost 2 million yuan and was bulky, and many cinemas found it troublesome. Jiang Defu's response was not to fight the trend, but also not to pretend that the trend would solve all problems. If this cinema won't show it, go talk to the next one. Then they talked with people about technology and about the future.

When "Heavenly Grassland" officially entered theatrical release, what he said to the audience entering the theater was: "This is the first digital film. If anyone is willing to buy a ticket and come in, I thank you."

Behind this posture is a judgment that only practitioners who have come all the way from film to digital and from digital to AI possess: awkwardness during a technological transition is normal; what matters is not pretending during the transition that things are already perfect. A new technology will not be accepted just because it is new. At least for a century-old film industry, it must first prove that it can truly tell a good story.

No matter where this film's final box office lands, it has already moved the question of "whether AI can actually make a film that enters theaters and sells tickets" from PowerPoint slides and short videos onto China's big screens for the first time, handing it to real audiences and real box office for testing.

Behind this report card is a group of people who spent two years relentlessly tackling one thing. Their choices, struggles, judgments, and misjudgments may themselves be more worth recording than the box office. (This article was first published on TMTPost APP, text by LccBeta)