The first AI theatrical film "Sanxingdui: Future Past" has a cold box office, yet ByteDance and Kuaishou still push forward film-grade models

English text is being prepared.

On October 23, "Sanxingdui: Future Past" officially landed in theaters nationwide. As the first AI hyper-realistic film in China to obtain a theatrical release permit, it has carried a lot of industry expectations since its release date was announced. But until its release, those expectations never translated into much enthusiasm for ticket purchases.

As of before its release, the number of people who wanted to see it on Maoyan was fewer than 6,000, and about 17,600 on Taopiaopiao; total pre-sale box office across the two platforms was about 144,000 yuan, and Beacon Professional Edition predicted a first-day box office of only 311,000 yuan and a total box office of 2.169 million yuan. For a film with a production cost in the tens of millions of yuan, such a forecast is clearly not optimistic.

The film's market performance stands in sharp contrast to the enthusiasm it received in the capital market. After its release date was officially announced at the end of September, Bona Film Group's stock price hit the daily limit at one point; after the National Day holiday, the company's stock price rose continuously again. The industrial imagination brought by AI films has already received a round of market response, but whether audiences are willing to pay for it remains an unresolved question.

Although the first AI film to land in theaters has met with a cold market, video model companies such as ByteDance and Kuaishou have not given up on the big screen. Through higher-resolution generation capabilities, professional studios and film and television IP cooperation, they are striving to enter the film production process and comprehensively revolutionize the production methods of the film industry.

"Sanxingdui: Future Past" is produced by Bona Film Group, has a runtime of 100 minutes, and is registered as an animated film. In April this year, the film announced that it had obtained a film release permit issued by the China Film Administration.

Previously, AIGC had already entered the production of short dramas, comic dramas and long dramas, but an AI-generated feature-length film formally entering theaters nationwide is still a groundbreaking move.

To complete this film, Bona invested nearly two years of technological exploration. According to media reports, the production team generated more than 1.2 million base images, carried out hundreds of full-film iterations, and set up 17 professional divisions including cinematography, lighting and art in reference to traditional live-action crews, with nearly 100 people participating.

Qu Ji Xiaojiang, head of Bona's AIGMS production center, once told the media that this is a film with a cost in the tens of millions of yuan. He also said that if "Sanxingdui" were shot using traditional live-action methods, the cost could reach 800 million to 900 million yuan. The cost gap between the two production methods is an important starting point for Bona's attempt at AI films.

However, AI generation technology did not make the production process simple. After the team obtained the dragon mark, the model had a new version, so the film's visuals were upgraded again; from submission for review in September 2025 to subsequent revisions, there were more than 100 minor adjustments.

The model can quickly generate shots, but the consistency of character appearances, shot continuity and the overall audiovisual effect still require the team to repeatedly screen and adjust. Faster generation does not mean that the production cycle and manpower input can be greatly reduced.

Although the film has just been released, Bona Film Group's stock price has already staged a rally.

On September 27, Bona Film Group announced the release date of "Sanxingdui: Future Past," and the next day the company's stock price hit the daily limit. After the National Day holiday, Bona Film Group's stock price rose twice more on October 9 and 12, with gains of 9.94% and 10.04% respectively on the two trading days. The label of the first AI theatrical film brought new market imagination space to a film company under operating pressure.

But this round of gains did not continue. As pre-sale results showed no improvement for a long time, Bona Film Group's stock price began to pull back. On October 15, the company's stock price fell 8.24%; on October 22 it nearly hit the daily limit down, having fallen more than 20% from its October 12 high.

At the same time, Bona has not slowed down its AI film layout. In October, the company announced that its next AI film, "Off Plan: Guanghan Action," had started production and was scheduled for release in the first half of 2027. The first AI theatrical film has not yet formally delivered a complete box office answer, and new projects have already begun to advance; the industry's exploration of this technological route continues.

From the production process to commercialization, AI films still need to cross several thresholds: whether improved generation efficiency can bring stable finished-film quality, whether lower production costs can translate into actual cost advantages, and, after the novelty of the technology fades, what works can rely on to attract audiences to pay.

"Sanxingdui: Future Past" makes AI feature films entering theaters a reality, but this is only a starting point for industrial development. For the entire industry, the key next step is not only to continue launching new works, but also to find a balance point where technological capability, content quality and commercial returns can support one another.

According to a relevant report by the DataEye Research Institute, the domestic market size of AI dramas and comic dramas is expected to exceed 40 billion yuan in 2026, a year-on-year increase of 138%. The scale of the AI film and television industry is expanding rapidly, which is already an indisputable industry trend.

A survey conducted by the Social Survey Center of China Youth Daily in conjunction with Wenjuan.com shows that among 1,334 respondents, 75.6% had watched AI film and television dramas; among those who had watched them, 65.1% thought the overall viewing experience was average—watchable but not impressive, and occasionally immersion-breaking. Hollow expressions, unnatural appearances and stiff dialogue were the problems mentioned more often by respondents.

The big screen in theaters will further magnify shortcomings in production.

Director Li Tian'ang once gave an example in an exclusive interview: AI-generated characters may show changes in facial details as soon as the camera moves. On small mobile screens, such problems are not necessarily easy to notice; projected onto the big screen, flaws in characters' faces, movements and shot continuity will be more obvious.

The viewing scenarios for short dramas and animated dramas are relatively relaxed, and viewers can swipe away at any time; films, by contrast, require viewers to arrange their time in advance, pay for tickets, and maintain attention for a longer period. Works need to continuously provide visual and narrative appeal, and viewers' demands for character performance, visual continuity and plot completeness will naturally rise as well.

Looking back at the AI film and television projects that currently have revenue, most rely on mature IP and existing broadcast channels, reaching viewers through platform revenue-sharing or free coverage. Their commercial performance is inseparable from the works' existing audience base and distribution channels. This is a different business model from asking viewers to pay separately in cinemas for AI works.

iQIYI and Changxin Media's co-produced "Soul Ferry" "Floating Life Dream" series is a fairly representative case. The series includes three AIGC online feature films—"Butterfly Dream," "The Dream of the Heavenly Maiden" and "Princess Dream." According to data released by iQIYI, as of October 3, cumulative revenue-sharing box office exceeded 10 million yuan, and the first two films recouped their production costs in their first week online.

This result is not merely the product of AI production technology. The series is backed by the existing "Soul Ferry" IP, with Guo Jingyu serving as artistic director, Xiao Jixiangtian responsible for screenwriting, and the original series' lead actors also participating in image licensing and voice acting. It both retains the audience base accumulated by the original work and reaches users through iQIYI's existing content distribution channels.

More importantly, it adopts the revenue-sharing model of online feature films, so viewers do not need to first decide, as they would for a new theatrical release, whether to pay for an unfamiliar AI film. A familiar IP and a lower threshold for first viewing provide the work with a more favorable starting point.

"Journey to the West: The Later Years" provides another frame of reference.

This AI long-form drama premiered on Hunan Satellite TV's prime time slot and Mango TV on August 31, 2026. On the night of its premiere, the peak real-time rating on Kuyun reached 0.3384%, ranking first among provincial satellite TV channels in the same time slot, making it an important attempt for AI content to enter traditional television screens.

However, television ratings, full-network dissemination and users actively following a drama are several different indicators. A September report by Beijing Youth Daily, citing Yunhe Data, showed that in early September the online broadcast market share of "Journey to the West: The Later Years" was less than 1%. Although it gained widespread attention under the label of "the first AI long-form drama to air on satellite TV," this volume of attention did not translate synchronously into an equivalent level of online binge-watching performance.

After the drama premiered, Mango Super Media hit the daily limit up for two consecutive trading days, adding more than 10 billion yuan in market value. What the market is watching is the technical capability and industry prospects represented by the first AI long-form drama to air on satellite TV; whether viewers are willing to keep following it still must be answered by subsequent playback performance and word of mouth.

Putting these two cases together with "Sanxingdui: Future Past," it can be seen that the consumption scenarios they face are not the same. "Soul Ferry" has a mature IP and platform revenue-sharing channels, while "Journey to the West: The Later Years" gained coverage through satellite TV and a video platform; "Sanxingdui: Future Past," by contrast, needs to attract viewers into cinemas before its official release through an original story, a technology label and marketing.

These differences do not mean that AI films have no market, nor do they prove that free content will inevitably squeeze out theatrical demand. But for a new film, "the first AI theatrical film" can create attention, yet it may not be enough to become a reason to buy a ticket. How to turn a technology showcase into a paid viewing experience worth the money is still a question it needs to answer.

Model makers set their sights on the big screen, competing over film-grade production capability

AI films' theatrical box office remains to be tested, while model makers are looking for another path into the film and television industry. Rather than directly betting on whether an AI film can be a box-office hit, it is better to first embed generation capabilities into existing production workflows, provide directors and professional studios with more controllable creative tools, and then verify the technology's usability through actual projects.

Just one day before "Sanxingdui: Future Past" was released, Volcano Engine launched the first batch of enterprise trials for the Seedance 2.5 4K model, supplied in a targeted way to 20 studios and directors worldwide and tied to specific works, exploring film and television production needs for the big screen.

According to media reports, trial studios can first generate sample footage at 480P, then regenerate satisfactory shots at 4K to control production costs. In addition to improving image detail, this version also focuses on improving the naturalness of character performance and the consistency of reference-based generation and editing.

Volcano Engine had also previously advanced AI film-related cooperation. During the Pingyao International Film Festival on September 27, the company released the Seedance film and television cooperation plan, providing up to one million yuan in Token subsidies per project to support feature films, series, animation and creative short films, among other projects.

In June this year, Volcano Engine also entered into an AI video copyright cooperation with Stephen Chow's Bingo Group. By establishing ties with film and television projects and classic film IP, model makers are also looking for opportunities to apply generation capabilities to actual creation.

Kuaishou Kling is likewise improving the model's support for film-grade images. Kling 4.0 launched on October 15, focusing on improving the coherence of large-scale movements and complex camera work, supporting generation of up to 30 seconds at a time, and allowing creators to combine multiple reference materials and input keyframes to control image development; 4K and 21:9 ultra-wide aspect ratios are directions for subsequent support.

Model progress is also beginning to show in specific AI film creation. During the Busan International Film Festival, Kim Seong-su, director of "12.12: The Day," demonstrated the production process of the AI short film "Worm Train." In the film, the scene of giant insects having a complex collision with train passengers was once a difficult point in generative production, and the team completed the relevant shots with the iteration of the Kling model.

From Seedance 2.5 4K to Kling 4.0, competition among model vendors is extending from generating a single stunning shot to more specific production requirements such as character consistency, complex camera movement, and image precision. These capabilities determine whether generated footage can support longer narratives and whether the final film can meet the viewing standards of the big screen.

However, whether technological progress can ultimately translate into box office performance remains an open question. Higher resolution, more natural character performances, and more coherent shots can reduce the sense of disbelief during viewing, but they cannot replace story, character development, and narrative pacing. Lower production costs do not mean audiences will be willing to pay for AI works.

For the industry as a whole, whether generative capabilities can be translated into stable final-film quality and further into content that audiences are willing to pay for is the key to whether this path can go far.