Tens of millions of yuan in funding and two years of production were not enough to help Sanxingdui: Future Past conquer the box office. The film’s initial failure has raised questions about whether AI can genuinely reduce costs for the film industry or simply shift the burden from the film set to post-production and final editing.
On October 23, Sanxingdui: Future Past premiered in theaters as China’s first feature-length AI-generated film. However, according to The Paper, the movie earned only around 1.2 million yuan after four days in theaters. Box office data also indicated that audience attendance remained low.
More Than Just a Single Click
According to Beifang Wang, nearly 100 people participated in the production of Sanxingdui: Future Past. The team generated more than 1.2 million background images and went through over 100 rounds of revisions.
“Even after the film was completed, the major AI models continued to be upgraded and updated. We made more than 100 minor revisions and five major revisions. Otherwise, the film’s visual effects would have become outdated.
“After receiving the film’s release permit, we noticed that the newer models could generate better-looking images, so we continued to take advantage of model upgrades to improve the film’s visual quality,” producer Qu Ji Xiao Jiang explained.
These figures show that AI filmmaking does not work by simply entering a prompt and receiving a finished movie. The technology can generate characters, settings, and shots, but turning separate images into a 100-minute production still requires human oversight at every stage.
“Some viewers believe the film was created with a single click, but that is not the case. We still follow the filmmaking process used by the film industry, producing the work frame by frame,” emphasized Jiang Defu, CEO of Bona Film Group, the company behind the production.
The biggest challenge facing AI-generated films is maintaining visual consistency throughout the entire production. Characters must retain the same appearance and clothing, while settings and movements between scenes need to remain coherent.
According to the Economic Daily, current technology still has limitations in visual consistency, physical laws, and image refinement, particularly when applied to a 100-minute film. As a result, production teams must continue to intervene technically and make manual adjustments.
AI can generate an attractive image in seconds, but connecting thousands of shots into a complete film is an entirely different challenge. In the case of Sanxingdui: Future Past, the creation of more than 1.2 million background images illustrates the extensive amount of experimentation and revision involved. Generating images quickly does not necessarily mean finishing a film quickly, as the team must still select suitable images, discard unsatisfactory results, and regenerate scenes when necessary.
AI therefore reduces the need for certain traditional production processes while introducing new tasks, including prompt adjustments, quality control, error correction, and efforts to ensure consistency between scenes. The costs and workload do not necessarily disappear; some are simply transferred to these stages.
Can AI Films Really Save Money?
If production costs alone are considered, Sanxingdui: Future Past still demonstrates AI’s considerable cost-saving potential.
Producer Qu Ji Xiao Jiang said the film received an investment of tens of millions of yuan. By comparison, producing a live-action version on a similar scale could require at least 800 million yuan.
AI films can create fantastical settings, characters, and visuals that would be difficult to achieve through conventional filmmaking at a significantly lower investment. They also offer opportunities for filmmakers who lack the budget to pursue ideas that would otherwise require large sets or expensive visual effects.
However, lower production costs should not be confused with an equivalent reduction in total business expenses. A film still requires a screenplay, creative personnel, editing, sound design, post-production, censorship, marketing, and distribution. For AI-generated productions, expenses associated with tools, computing power, quality-control personnel, and regenerating shots must also be fully accounted for.
More importantly, the amount saved depends on the type of film, its quality requirements, and the production team’s level of expertise. Short projects with few characters and limited complex movement may benefit more from AI than feature-length films with multiple storylines and demanding performance requirements.
Therefore, it is impossible to conclude that every AI film will deliver the same level of cost savings.
One Film Cannot Determine the Future of AI Filmmaking
Despite its limitations in quality, feature-length AI filmmaking remains worth exploring. Sanxingdui: Future Past is one of the first projects to incorporate this technology into the production of a 100-minute feature film and release it commercially.
The film is therefore not merely a test of visual quality. It also helps the production team identify which tasks AI can handle and which problems still need to be addressed.
In an interview with The Paper, Jiang said the project aimed to find a viable workflow for applying AI in filmmaking that could offer new solutions for the industry.
He acknowledged that current technology does not yet meet every creative requirement. According to the producer, the value of the experiment extends beyond the box office revenue of a single film, as it allows the film industry to explore new possibilities.
Historically, filmmaking technology has always needed time to mature. When the industry transitioned from film stock to digital production, it also went through a period of adaptation before fully realizing the benefits in cost and production efficiency.
With AI, current limitations in visual consistency, movement, and emotional expression are likewise challenges that must be addressed through successive projects.
This does not mean audiences should lower their expectations for AI-generated films. Movies released in theaters must still meet standards for storytelling and the cinematic experience. However, an imperfect film should not be used to conclude that the technology has no future. Even traditional films backed by experienced crews and substantial budgets are not automatically guaranteed to succeed.
Sources: kenh14