Movies
Moviesfera Newsroom
Movies

An AI movie bombed in China, and the audience was right to shrug

Sanxingdui: Future Past drew 10,000 opening-day screenings and almost no one. Novelty sells a press cycle, not a ticket, and viewers are now the real regulators of generative film.

Published
October 28, 2026
Reading time
3 min
Categories
Movies

Photo: N509FZ / Wikimedia Commons (CC BY-SA 4.0)

What happens when you give a movie every technological advantage except a reason to exist? China just ran the experiment, and the answer was a shrug the size of a nationwide release.

The film is Sanxingdui: Future Past, billed as China's first fully AI-generated feature to reach cinemas, a sci-fi thriller that opened nationwide on October 23. It was no stealth release: PetaPixel notes it got 10,000 screenings on opening day. IndieWire reports it earned just $181,000 over the weekend. Other outlets cite a lower number, and a budget of about $1.5 million has been reported, so treat the exact math with caution. The direction of travel is not in doubt. Advance sales were already lagging before opening day, per VnExpress.

Curiosity gets you one weekend, not a run

The industry's recurring fantasy is that "first" is a genre. It isn't. "The first fully AI-generated feature in theaters" is a headline, and headlines are free. A ticket costs money and two hours, and that is a very different ask. In our view, the opening-day screening count is the most revealing detail: exhibitors can be persuaded to give a film space, but nobody can be persuaded to fill the seats.

Notice what the pitch actually offered. The selling point was how the film was made, not what it was about. Audiences have never paid for a production method on its own. Nobody bought tickets to see sound, color or CGI as such; they bought the stories those tools made possible. When the method is the whole marketing hook, the film is asking viewers to do the studio's PR for it.

The public isn't hostile to the tech, just unimpressed by the output

It would be lazy to read this as a mob against machines. The Guardian's review of Prime Video's Castle Walls, a Turkish AI-made series from Ay Yapım, offers a sharper diagnosis. The critic concedes the technology is "amazing" and will alter motion pictures, then points to a gap: early cinema offered carnival-like wonder, while these AI works tend to provoke a more cynical response. Viewers, the review argues, seem more interested in what the tech threatens to take away than in what it promises. The writing, it adds, is banal, even though humans wrote it.

That last point matters. Castle Walls used human screenwriters and voice actors, and still read as "make it cinematic" repeated at a prompt box. We suspect the same trap applies to Sanxingdui. Spectacle is now cheap to generate, which makes spectacle worthless as a differentiator. What remains scarce is the thing no model hands you by default: a point of view, a character you worry about, a reason the story had to be told.

There is a caution for the other camp too. Box office is not a moral scoreboard. Laika's stop-motion Wildwood holds a 99% critics' score on Rotten Tomatoes and, per IBTimes, opened to about $12 million domestically against a reported $100 million budget. Handmade does not guarantee a crowd, and machine-made does not guarantee a flop. The lesson isn't "human good, AI bad." It's that nothing, not craft, not novelty, not acclaim, substitutes for a film people actually want to see.

Audiences will draw the line, and it will move

Studios and tech firms love to frame the future of generative film as a policy or labor fight. Those fights are real, but the market is quietly running its own referendum, one weekend at a time. A film that costs a fraction of a traditional budget can survive a small audience. It still needs one. And that audience, not a boardroom, decides whether "AI-made" reads as a curiosity, a cost-saving tool or a red flag.

Our prediction, clearly only that: generative tools will keep seeping into the pipeline, mostly invisibly, in the hands of filmmakers with something to say. The all-AI feature sold as an event will keep struggling until it earns the event. Financiers should read Sanxingdui as a warning against confusing enthusiasm for a technology with demand for its products.

The verdict: if you're a studio, stop buying the "first" and start buying the story. If you're a viewer, keep doing what you just did. Your ticket is the only vote that has ever really counted.

Debate topics

No topics yet: start the first one.

More stories