A film opens in cinemas across China. It has bronze trees, masked figures and a future city. It also has something no earlier Chinese feature film is publicly documented to have had: an official release permit for a movie made largely by machines. The first days were a study in contrasts — screenings across the country, and a box office that barely registered.
On October 23, Sanxingdui: Future Past (《三星堆:未来往事》) became the first AI-produced feature film in China to receive a national release permit. It was produced by Bona Film Group (博纳影业) with model support from Shengshu Technology's Vidu (生数科技 Vidu). The permit is a genuine institutional milestone. The ticket sales are the other half of the story.
Key takeaways
- The film is the first AI-produced feature to receive a public-release permit (公映许可证), colloquially a "dragon label" (龙标), from China's National Film Administration.
- It runs 100 minutes and took roughly two years to make, with a team of about 100 and more than 1.2 million base images generated during production.
- The producer says the project cost tens of millions of RMB, against an estimated RMB 800–900 million (≈ US$113–127M / HK$879M–989M) for a conventional shoot of the same scale. That is a producer's estimate, not an audited comparison.
- Box office in the first two days was just over RMB 500,000 (≈ US$70,000 / HK$549,000), according to Maoyan — a modest return for a heavily covered "first."
What the "dragon label" means
In China, a film cannot be shown in cinemas without a public-release permit from the National Film Administration. The permit is popularly called a "dragon label" (龙标) after the emblem printed on it. It is a regulatory and technical gate: the film has been submitted, reviewed and cleared for exhibition.
Plenty of films have used AI for individual shots or scenes. What is new here is that a full-length, AI-produced feature passed through the same system as a conventional film. The production team describes the result as a closed loop — character generation, scene construction, shot rendering, editing and final delivery all completed and then examined against cinema standards. That is a process achievement as much as an artistic one.
The story draws on Sanxingdui, a Bronze Age culture known for its unusual masks and ritual objects. The film rebuilds artifacts such as the bronze sacred tree and the upright bronze figure, then wraps them in a science-fiction plot about a future warning decoded from ancient symbols. A precedent exists on the small screen: a widely viewed 2024 AI science-fiction microdrama (短剧) on the same subject, released on Douyin, served as a trial run for the feature, according to Jiemian's account of the release.
How it was made
The numbers behind the production are more interesting than the plot. According to reporting citing Xinhua, the team worked for two years, produced more than 1.2 million base images and went through hundreds of full-film iterations. The film was assembled on a cloud production platform developed by Bona, with Vidu providing the underlying video models.
The division of labour is the part worth noting. Models generated and executed the visual material; creative decisions — story direction, shot language, performance detail and historical verification of the artifacts — stayed with the human team. "Tasks can be handed to AI, but judgment and responsibility cannot be outsourced along with them," said producer Qu Jixiaojiang (曲吉小江) at a preview event, according to Jiemian and other reports.
That sentence is the most useful thing to come out of the project. It describes a realistic workflow for machine-generated (机器生成) content: models produce volume, people decide what is correct and what is allowed to ship. The alternative — letting the model decide — is still not how a regulated feature film gets made.
The box-office reality check
The reception was softer than the coverage. Pre-sale figures before opening were low: roughly RMB 144,000 (≈ US$20,000 / HK$158,000) across platforms, with fewer than 6,000 "want to see" marks on Maoyan and about 17,600 on Taopiaopiao. The film opened with close to 11,000 screenings on its first day and more than 13,000 on the second, yet two-day box office just passed RMB 500,000. It was released in Malaysia at the same time, with further overseas deals under discussion.
None of that erases the achievement. A permit, a 100-minute runtime and a nationwide release are real thresholds. But the gap between "first of its kind" and "audiences want to watch it" is exactly what the early numbers reveal. A technological label does not automatically become a ticket sale.
The producer has said the project cost in the tens of millions of RMB and estimated that a comparable conventional production might have cost RMB 800–900 million. Both figures come from the production side and should be treated as claims. Even taken at face value, they describe a change in the cost of making images, not yet a change in the value of watching them.
Honest limitations
This article relies on reporting by Jiemian, Red Star News carried by Henan Television, China.com and other outlets describing the same October 23 release, with references to a Xinhua account of the production process. I have not seen the permit itself, the film's financial statements or any audited production budget. The cost comparison and the "first" claim are reported statements, and I have not searched every film database to test the scope of the latter.
The box-office figures are a snapshot from Maoyan at a particular moment, not a final total. I have deliberately left out coverage of the producer's share price and exchange filings; those belong to a different story and do not change what the film is or how it was made.
The plot is described differently across reports — the number of symbols and artifacts varies — so I have kept those details brief. The quote attributed to Qu Jixiaojiang is reported in Chinese and translated here; readers should check the original for exact wording.
What readers can do now
- If you work in film or media, separate the permit from the performance. A cleared production pipeline is a reusable asset. A strong opening weekend is not guaranteed by it.
- If you make AI-produced content, adopt the division of labour described here. Let models handle volume; keep verification, attribution and final judgment with people. It is both good practice and, in regulated media, a requirement.
- If you are simply watching the space, check the box office a week later. The first-day story is about novelty. The second-week story is about whether the film was any good.
The permit is a door that had not been opened before. What walks through it next depends less on the model than on the people deciding what deserves to be on the screen.
