China’s First AI Movie Flop Is a Warning for Synthetic Cinema

Novelty is not a business model

The most revealing thing about Sanxingdui: Future Past is not that a Chinese studio managed to put an AI-made feature into cinemas. It is that the public test was so unforgiving.

The film opened in China on October 23, 2026. It was promoted in Chinese and international coverage as the country’s first theatrical feature made through an AI production workflow and approved for cinema release. By October 26, multiple Chinese box-office reports, citing Maoyan Pro or Dengta, put its cumulative gross at roughly RMB1.18 million to RMB1.23 million. The Paper, citing Dengta Pro, reported RMB1.216 million by 4 p.m. on October 26, including previews; Beijing Daily, citing Maoyan Pro, reported RMB1.198 million by publication time; City News Service reported just over RMB1.18 million by 11 a.m. on October 26. (thepaper.cn)

That is a weak result for any wide theatrical release. It is sharper still because the reported production cost was above RMB10 million, and the opening was not hidden in a few experimental screenings. Reports say the film had more than 10,000 screenings on opening day, with occupancy around 1.1%, roughly one viewer per screening. (www1.citynewsservice.cn)

This article deliberately uses the word “flop” narrowly: based on the first four days of theatrical box office, the film performed far below its reported production cost and far below any plausible theatrical break-even. That conclusion could change only if later income — streaming, licensing, subsidies, overseas sales or non-theatrical contracts — materially alters the project’s economics. No public evidence found during this review proves such later recovery.

The broader lesson is not that AI has no place in cinema. The lesson is more uncomfortable for investors: lowering some production costs does not automatically create audience demand. A cinema ticket is not a technology grant. It is a purchase of time, trust, emotion and story.

What is established — and what is not

Because AI cinema is now surrounded by marketing claims, the evidence needs sorting.

One correction matters. Some English-language reports converted RMB900,000 into roughly US$180,000. That is not a sound conversion. At an exchange rate around RMB7 per US dollar, RMB900,000 is closer to US$125,000–130,000. Even RMB1.2 million is roughly US$165,000–170,000, not automatically US$181,000 unless a different exchange rate or higher gross is used. The economic conclusion does not change, but careless currency conversion should not be repeated as fact.

There is another limitation. I found secondary reports citing Maoyan Pro and Dengta Pro, but I did not find a stable public archive or official screenshot from those platforms for each date between October 23 and October 27. The box-office numbers therefore remain reported industry data, not independently downloaded raw data.

The film and the promise

Sanxingdui: Future Past is a roughly 100-minute science-fiction animation built around the Sanxingdui archaeological culture in Sichuan. Pre-release and release coverage described a story in which future humanity, trapped by overreliance on algorithms, searches ancient Shu civilization for clues to survival. Xinhua reported that Boyue Xingji Lantu companies produced the film and that Vidu, the AI video platform from Shengshu Technology, provided chief technical support. (xinhuanet.com)

The premise is almost too neat: a future civilization damaged by algorithmic dependence turns to ancient culture for rescue. But the market response suggests viewers did not buy the package merely because the production method was new.

Bona’s own public filings show why the project attracted attention. Its 2025 annual report listed Sanxingdui: Future Past as an AI film project that had passed content review and obtained the “dragon label” — the public screening permit — while warning that AI film projects, application-platform plans and strategic outlooks were forward-looking and subject to technical feasibility, market demand, competition and regulatory uncertainty. (static.cninfo.com.cn)

That warning now looks less like boilerplate and more like the central issue.

The economics: a small budget can still lose money

The reported RMB10 million-plus production cost is modest by commercial film standards. But “cheap” does not mean “profitable.” Under common Chinese theatrical revenue sharing, gross box office is not what returns to producers.

Chinese film-industry explanations commonly describe the first deductions as 5% for the national film development fund and 3.3% VAT/tax, leaving roughly 91.7% as shareable box office. A Beijing International Film Festival distribution masterclass summarized a common split as about 43% for the producer/distributor side and 57% for cinemas and cinema chains, after the deductions; China’s film regulator previously stated that cinema share for first-run films should in principle not exceed 50%, showing the policy background behind such negotiations. (bjiff.com)

A rough calculation illustrates the problem. If four-day gross was RMB1.2 million, then after the 8.3% deductions the shareable pool would be about RMB1.10 million. At 43%, the producer/distributor side would receive roughly RMB473,000 before distributor fees, promotion costs, minimum guarantees, settlement adjustments and any internal allocation among investors. That is less than 5% of a RMB10 million production cost. Even if Dengta’s AI forecast cited by The Paper — RMB2.609 million final gross — proved correct, the producer/distributor pool would be only about RMB1.03 million before further costs. (thepaper.cn)

This is not a full profit-and-loss statement. We do not know whether the reported budget includes marketing and distribution. We do not know whether Bona capitalized the project as inventory, production cost, intangible asset or expensed item in a way visible at project level. Bona’s public reports disclose AI film and television as a segment, but they do not provide a project-level cost schedule for Sanxingdui. Its 2025 annual report showed AI film and television revenue of RMB2.3868 million, only 0.19% of total 2025 revenue, and stated that the AI business remained early-stage with technology path, business model and market acceptance still needing verification. (static.cninfo.com.cn)

The 2026 half-year report showed total revenue of RMB458.3 million, AI film and television revenue of RMB1.932 million, and AI film and television cost of RMB61,260 in the reported segment table. That high apparent margin may look attractive, but it relates to the disclosed segment for the first half, not necessarily the theatrical economics of Sanxingdui. Treating it as proof that AI features are profitable would be a category error. (money.finance.sina.com.cn)

Why China was an important test market

This did not happen in a marginal cinema market. China’s National Film Administration reported that the country’s 2025 box office reached RMB51.832 billion, with 1.238 billion urban cinema admissions and 93,187 screens at year-end. Domestic films took RMB41.293 billion, or 79.67% of the total. (chinafilm.gov.cn)

So the film did not fail because Chinese audiences are uninterested in domestic cinema or spectacle. The same official report celebrated Chinese animation and advanced film technologies, including virtual production, AI generation, cloud computing and new immersive formats. (chinafilm.gov.cn)

Nor was this an anti-AI policy environment. China’s 2023 interim generative-AI rules explicitly combined support for innovative applications with obligations around public interest, security and personal information. Its 2025 AI-generated synthetic-content labeling measures, effective September 1, 2025, require providers and platforms to identify AI-generated content in specified ways. (english.www.gov.cn)

That makes Sanxingdui more revealing. The film launched in a market where domestic cinema, cultural heritage and AI development are all strategically useful narratives. The missing element was audience appetite.

The quality problem: feature films are not clips

Generative video works best when the unit of attention is short. A striking few seconds can go viral even if continuity is unstable. A theatrical feature is different. It needs characters who remain visually and emotionally coherent across scenes. It needs clear geography, rhythm, escalation, voice performance, lips that match speech, and a story that rewards staying for 100 minutes.

City News Service reported that some Douban users praised the visuals and music but criticized inconsistent character appearances, mismatched lip movements and disjointed storytelling. This is not a systematic audience study; Douban comments are a signal, not a scientific sample. But the reported complaints go to the core weakness of current AI video: continuity over time. (www1.citynewsservice.cn)

This is where the phrase “fully AI-made” can mislead. Futurism, citing company claims, reported that nearly 100 people worked on the film for more than two years. Other reports said the team generated massive numbers of base images and repeatedly revised the film. (futurism.com)

If that is accurate, the real workflow was not “press a button, get a movie.” It was a human-heavy attempt to control machine output. That is not a scandal — animation has always used tools — but it weakens the investor fantasy that AI will immediately turn feature production into near-zero-cost content manufacturing.

Follow the incentives

Bona had rational reasons to try. Its 2025 annual report showed total revenue of RMB1.253 billion, down 14.28% year on year, and a net loss attributable to shareholders of RMB1.464 billion. In October 2026, after investor excitement around Sanxingdui and another AI project, Bona issued abnormal stock-movement announcements. On September 28, it said the film had obtained the public screening permit and was scheduled, but box-office income remained uncertain; on October 12, it repeated that warning and disclosed that the company remained loss-making in the first half of 2026, with net loss attributable to shareholders of RMB169 million. (static.cninfo.com.cn)

The incentives are obvious. A studio under financial pressure wants a cheaper production pipeline. AI vendors want professional proof that their tools can move beyond social-media clips. Cinemas need content. Investors want a growth story. Local cultural institutions benefit when ancient heritage becomes exportable screen IP.

But all those incentives sit upstream of the viewer. A cheaper supply chain does not create demand if the product feels derivative, unstable or emotionally empty.

The global comparison: Hollywood is not immune

China’s failed test lands in the middle of a global AI-cinema gold rush. In September 2024, Runway and Lionsgate announced a partnership to train a custom AI model on Lionsgate’s proprietary catalogue, with Lionsgate Vice Chair Michael Burns describing AI as a tool for “capital efficient content creation” and Runway presenting the model as a way to augment creative workflows. (runway.com)

A year later, TheWrap reported that the partnership had run into complications: limitations of using a single model and one studio catalogue, copyright concerns, and possible ancillary rights of actors. Importantly, TheWrap also reported Lionsgate’s response: the studio said it was still pursuing AI initiatives on several fronts and that the Runway deal was not exclusive; Runway did not respond to TheWrap’s request for comment. (thewrap.com)

This comparison matters because the sober use case is different from the hype. AI may be valuable for backgrounds, previs, concept art, localization tests, cleanup, marketing versions or controlled visual effects. That is not the same as replacing directors, writers, performers, editors and animators in a finished feature that audiences pay to see.

The labor issue is not theoretical. SAG-AFTRA’s 2023 TV/Theatrical agreement says producers must obtain informed consent and compensation for creating and using digital replicas of performers; it also requires notice and bargaining opportunities around synthetic performers. (sagaftra.org)

Europe is moving through transparency law. The EU AI Act is a risk-based framework that includes obligations for certain AI systems and transparency duties around AI-generated content; its broader implementation is staged. (digital-strategy.ec.europa.eu)

China’s path is state-supervised development with licensing, content review and labeling. The common theme across regions is clear: AI film is not entering a legal vacuum. It is entering disputes over copyright, labor, likeness, disclosure, censorship, market concentration and consumer trust.

What remains unanswered

The public record still leaves important gaps.

Bona has not, in the accessible materials reviewed here, provided a technical breakdown of what “fully AI-made” means: which shots were generated, how much human animation or compositing was used, which models were used, how training data was licensed, how voices were created, whether metadata labeling rules were applied, and how many people worked in each production function.

The company also has not publicly disclosed a full Sanxingdui P&A budget or a project-level accounting schedule. Without that, no one outside the company can calculate the final loss precisely.

Maoyan and Dengta box-office readings should ideally be archived directly for each date, not only quoted in media. Cinema operators should also explain why so many opening-day screenings were scheduled despite weak presales, whether this was driven by novelty, distributor pressure, low competition in the release window, or platform forecasting.

Finally, workers deserve a clearer voice. The debate often jumps from studio executives to AI vendors, while VFX artists, animators, voice actors, writers and junior production workers carry the economic risk of “efficiency.”

The real warning

Sanxingdui: Future Past does not prove AI cinema is dead. It proves that the audience is not obligated to reward a workflow.

AI may become a serious film tool. It may help small teams visualize worlds they could not otherwise afford. It may reduce selected costs and open new forms of animation, restoration, localization and experimental storytelling. But that is a different claim from the fantasy that synthetic cinema will automatically flood theatres with profitable features.

The first Chinese AI theatrical feature — as described by its producers and the press — had policy relevance, cultural IP, a large domestic market, a wide opening and investor attention. What it lacked was paying demand.

That is the lesson the AI-cinema gold rush should not ignore. Cinema is not generated. Cinema is judged.