The release was not the problem. Bona Film Group secured what no AI production in China had secured before: a public screening permit for a 100-minute feature made with AI throughout the process, filed as animation, built from original digital characters rather than replicas of real actors. Cinemas gave it room: more than 10,000 screenings on opening day, October 23, across 6,901 cinemas. Then audiences were asked to pay between 20 and 90 yuan a ticket, and mostly did not.

By midday on Sunday, October 25, the film had taken 914,000 yuan, according to ticketing platform Lighthouse. Opening day accounted for 317,000 yuan of that, with seat occupancy of 0.9%. Occupancy improved slightly to 1.2% on the Saturday, while the number of prime-slot screenings was already being cut. Lighthouse, at that Sunday-midday reading, was projecting a final gross of 2.505 million yuan — a forecast, not an outcome. Bona's stated production cost is above 10 million yuan, and the producer does not receive the gross: of the 797,000 yuan in distributable revenue counted by Sunday midday, Lighthouse estimated 314,000 yuan flowing to the producer against 417,000 yuan to cinemas.

What 100 people and 1.2 million images bought

It would be wrong to read the result as proof that the production failed on its own terms. Sanxingdui: Future Past, written and directed by Yue Ming, was made by about 100 people over more than two years, working across 17 conventional film disciplines inside Bona's AIGMS division, with Vidu, developed by Shengshu Technology, as the principal technical support. The team generated more than 1.2 million source images and completed more than 100 full iterations of the film. AI produced the characters, voices and environments. People kept the concept, the story and the final calls, and worked around the tools in ways that will be familiar to anyone who has shipped with current video models: one planned embrace between a woman and a boy had to be restaged as an approach, a pause and a look, because the model blurred the characters when their arms overlapped.

That is a genuine production capability, demonstrated at feature length under a regulator's permit. The cost argument attached to it is weaker. Bona says the AI-produced film cost more than 10 million yuan. Its executives have offered substantially different estimates of what a conventional live-action production might cost: chief executive Jiang Defu put a comparable live-action version at at least 300 million yuan, while producer Qu Ji Xiaojiang put it at 800 to 900 million yuan. Those hypothetical comparisons are not measured savings. The verifiable economy is narrower: this team made a releasable feature for a low eight-figure yuan budget, and the market has now priced that feature at roughly 914,000 yuan by Sunday midday, against a projected final gross of about 2.5 million.

Free is a different business

The contrast that explains the weekend is on the phone. AI short dramas drew more than 200 billion views on Douyin in September alone, and Bona's own earlier online Sanxingdui series passed 200 million views. Large online viewing counts show that AI-produced video can circulate widely without a theatrical ticket purchase. They do not establish equivalent demand for paid, feature-length cinema. A cinema ticket is not a view with a price attached; it is a competing claim on an evening, judged against every other film on the same screen wall. Preview viewers quoted by AFP made the comparison themselves: the standard they named was not other AI work but Avatar.

Two caveats keep the result honest. First, Chinese cinemas are in a deep slump, with first-half box office down about 40.6% year on year and admissions down about 34.2%, so a weak opening is partly a weak market. Second, Bona treated the release as an experiment and said so before it opened, warning investors on October 12 that returns were uncertain. Investors had already acted on the production story, sending Bona's shares to their daily limit three times before release, while the company's AI unit was reporting a 96.83% gross margin on revenue of 1.9323 million yuan — 0.42% of the company's total. Despite extensive opening-day scheduling, the film attracted very few paying viewers relative to the seats available. The attendance data alone does not establish why: the production method, the story, the marketing, ticket prices and competing releases are all in play, and occupancy cannot separate them.

The next data point is already scheduled. The AI-generated feature Gods Don't Give Gifts is due to open in the United States on December 4, and Bona has three further AI projects in development. The production question — can AI hold a feature together for 100 minutes — has been answered well enough to stop being interesting. The open question is the older one: whether anyone involved can make a film people choose.

Sources

  1. [1] Asia Daily, "China's First AI Feature Faces Empty Seats as Bona…" — detailed account built on Lighthouse and Maoyan ticketing data: 914,000 yuan cumulative by midday Oct 25, opening-day 317,000 yuan, 0.9% opening-day occupancy and 1.2% on Oct 24, 6,901 cinemas, 2.9% screening share vs 1.2% revenue share, Lighthouse final-gross projection of 2.505 million yuan, producer/cinema split, production detail (1.2 million images, 100+ iterations, Vidu support, embrace restaging), both live-action cost estimates, and Bona's H1 2026 AI-unit financialsRead source
  2. [2] Aju Press, "Chinese AI Film 'Sanxingdui' Struggles at Box Office with Tepid Audience Response", October 26, 2026, citing Lianhe Zaobao — cumulative box office of about 900,000 yuan as of Oct 25, less than one-tenth of a production budget above 10 million yuanRead source
  3. [3] AFP via Abb Takk, "China's first AI-made feature opens to a subdued cinema audience", October 23, 2026 — ticket pricing context, preview-viewer reaction including the Avatar comparison, and Jiang Defu's "first to try, first to get a theatrical release permit, first to be tested by the audience" statement to YuliRead source
  4. [4] TechNode, "China's first AI-produced theatrical film sets October 23 release", September 28, 2026 — Bona's original announcement: 100-minute runtime, AI used throughout production, original digital characters with no replicas of real actors, creative concept and final decisions held by a team of more than 100 at Bona's AIGMS divisionRead source