China's AI Feature Film: What It Proves About AI Casts

A fully AI cast just carried a 100-minute theatrical release

On October 23, 2026, Bona Film Group released Sanxingdui: Future Past in theaters across China, on more than 10,000 screens. It's being described as the first hyper-realistic, wholly AI-generated feature film to receive an official theatrical screening permit from China's National Film Administration. Five days later, Vidu-maker GensPI confirmed its AI video platform powered the production.

The film runs 100 minutes and blends the ancient Shu civilization with a futuristic sci-fi setting. Every character is an original digital creation. No real actors were scanned, cloned, or licensed for digital likeness. That detail matters more than the runtime or the box office, because it's the clearest large-scale proof yet that an entirely invented AI cast can anchor long-form narrative at the highest production stakes a creator can imagine: a nationwide theatrical release, not a demo reel or a festival curiosity.

The production wasn't push-button AI

It's worth being precise about what this film actually demonstrates, because the headlines oversimplify it. Bona Film Group spent two years on production with a team of over 100 people. The script and creative direction were controlled by humans throughout; AI mainly handled image generation. This was a human-led production that used AI as a visual production tool at scale, not an AI system that wrote and directed itself.

That distinction matters for anyone drawing lessons from this for their own work. The achievement isn't 'AI made a movie.' It's that a disciplined creative team used AI tools to generate a consistent, original cast and visual world across 100 minutes of story, something that used to require live actors, sets, and traditional VFX pipelines. The tooling did the heavy lifting on image generation; humans still did the storytelling.

Original characters sidestep a real problem: rights and consent

One of the most practical details in this story is also the easiest to overlook. Because every character in Sanxingdui: Future Past is an original digital creation, with no digital replicas of real actors, the production avoids issues commonly associated with digital likenesses, including image-rights authorization.

This is directly relevant to short-form creators experimenting with AI video. If you build a cast of original AI characters rather than generating content that mimics real public figures or licensing a real actor's digital likeness, you remove an entire category of legal and ethical risk before it starts. You don't need releases, likeness agreements, or platform takedown battles over someone else's face and voice. An original AI character is yours outright, and that's a structural advantage, not just a creative choice.

The box office is the part creators shouldn't ignore

Here's the part of the story that's easy to skip past if you're excited about the tech: advance ticket sales were weak for a landmark production reported to cost $1.5 million. The novelty of 'first fully AI-generated theatrical film' did not translate into audience demand on its own.

That's a useful, if uncomfortable, data point for short-form creators. 'Made entirely with AI' is a production fact, not a hook. Viewers don't watch something because of how it was made; they watch because of what happens, how fast it moves, and whether they care what happens next. The lesson from Sanxingdui isn't just 'AI casts can work at scale,' it's 'AI casts can work at scale, but they still live or die on story and pacing like everything else.' The production team itself says it hopes for international release, including Western theaters eventually, but that ambition doesn't change the ticket sales story happening right now in China.

What this validates for short-form creators

Scale down the headline and the takeaway for short-form video is direct: a consistent, original AI character or cast can carry narrative weight, and investing in that consistency is worth it compared to generating one-off, disconnected clips. GensPI's CEO Yihang Luo framed the project as proof that 'advanced video models shouldn't just look good in curated demos,' pointing to how far AI video technology has come while giving creators room to iterate toward the right shot. The same logic applies whether you're making a 100-minute feature or a 60-second episode: a recognizable cast with consistent appearance and personality is what lets an audience build a relationship with your content over time, instead of treating each video as a disposable generation.

The practical move for creators isn't to chase feature-length ambitions. It's to treat an AI-generated character the way Bona Film Group treated its cast: as a defined, reusable asset you develop deliberately, rather than something you regenerate from scratch every time. This is the exact problem Episodo is built around, helping creators build a consistent AI cast from a one-line premise so the character, not just the clip, becomes the thing people recognize and come back for.

• Build original AI characters instead of relying on real-person likenesses to avoid rights and consent headaches entirely
• Treat your AI cast as a reusable, consistent asset, not a one-off generation per video
• Don't rely on 'made with AI' as your hook, audiences reward story and pacing over production novelty
• Remember that even large-scale AI productions still need strong human-led creative direction and writing