What is the AI Backlot at the Shanghai International Film Festival?

The AI Backlot is a production initiative at the 28th Shanghai International Film Festival, developed in partnership with MiniMax's Hailuo AI. It pairs four traditional film directors with four AI creators to collaboratively produce short films inside the festival itself — not as a sidebar competition, but as an active production facility where cinematographic expertise meets hands-on fluency with AI generation tools.

Why are AI video generation tools difficult for filmmakers to use precisely?

AI generation tools struggle to execute highly specific aesthetic instructions in a single step. As Nina Zheng of ASUS China described, filmmakers want precise emotional quality, subtle lighting changes, and deeply personal aesthetic judgments — but AI models often require frequent adjustments to approach that intent. The gap between what a filmmaker envisions and what an AI model produces by default is the fundamental challenge the field is still working through.

How does MiniMax Hub help filmmakers maintain creative control over AI-generated video?

MiniMax Hub consolidates image creation, video generation, voiceover, music, and editing into a single platform driven by an AI agent that reads natural-language briefs or PDF proposals. Crucially, the agent pauses at key decision points and waits for human confirmation rather than running autonomously to completion. It also includes Memory and Skill functions so filmmakers can load their own workflows and aesthetic standards, allowing the system to apply their personal creative logic to future projects.

The backlot -- CinePrompt Field Notes

The 28th Shanghai International Film Festival opened last week with something no Western festival has tried. An AI Backlot, developed with MiniMax's Hailuo AI, pairs four traditional film directors with four AI creators to produce short films together. Not an AI sidebar. Not a separate competition. Not a protest banner in the lobby. A production facility inside the festival where the person who knows what a shot should look like sits next to the person who knows which buttons to press.

The pairing is the interesting part. Not the technology. Not the government backing. Not the five service platforms or the High-Tech Films and Televisions City in Songjiang or the G60 Science and Technology Innovation Corridor. The pairing.

Because the pairing is an admission. The person who carries decades of cinematographic vocabulary and the person who carries fluency with generation tools are not, in 2026, the same person. And rather than waiting for that convergence to happen naturally, Shanghai stuck them in a room and handed them a deadline.

Compare this to the institutional responses Western festivals have cycled through for over a year. Cannes banned AI from the Palme d'Or and took Meta's sponsorship money in the same press conference. The Academy wrote rules requiring "human-authored" screenplays and "demonstrably performed" acting. The Golden Globes softened the language to "primary." The EU mandated disclosure labels unless editorial control was exercised. The DGA sealed terms that treat AI-generated footage as footage under the director's authority. Eleven institutions on a gradient, each answering the question of what AI filmmaking is allowed to look like.

Shanghai did not answer that question. Shanghai built a room where the answer gets made.

The SIFFORUM panel, "Smart Tech, Immersive Worlds, the Next Film Revolution," surfaced three challenges facing Chinese filmmakers working with generation tools: compute, distribution, and directability. The first two are engineering and business problems. The third is the translation gap that defines this moment in filmmaking.

Nina Zheng, deputy general manager of ASUS China, described it with surgical clarity: "Filmmakers want a very specific emotion, a very particular change in lighting, something very subtle and highly personal in terms of aesthetic judgment. But when these directions are executed by AI, it often fails to get there in one step, requiring frequent adjustments."

That is the entire vocabulary gap in two sentences. A filmmaker knows what the light should do. The model does not get there in one step. The gap between those two facts is where the work lives.

Zheng followed with the sentence that mattered most: "The more advanced the tools become, the more critical the people driving the tools become."

Not less critical. More critical. Tools that can execute anything amplify the distance between someone who knows what to ask for and someone who accepts whatever arrives. A chisel in Michelangelo's hand is not the same chisel in mine. The chisel did not change. The hand did.

Same week, same festival, MiniMax launched Hub, an all-in-one AI video platform that consolidates image creation, video generation, voiceover, music, and editing into a single interface. Xu Lüyang, from MiniMax's product operations department, described the old workflow: open one AI tool for images, another for video, another for voiceover, another for music, then a video editor to assemble the pieces. Hub replaces all of it with an agent that reads a natural-language brief or a PDF proposal, breaks down tasks, selects models, executes, and pauses at decision points for human confirmation.

That pause is the design choice worth examining. Hub does not run to completion and hand you a finished product. It stops. It asks. It waits for the filmmaker to say yes or no or try something else. Xu described the philosophy explicitly: "The AI agent shouldn't be a black box, one-click generator. Creative direction and aesthetic judgment must ultimately be left to humans."

A product operations manager at a Chinese AI company just stated the thesis more concisely than most Western AI summit keynotes have managed in two years of trying.

Hub also includes "Skill" and "Memory" functions. Users hand the system their workflows, aesthetic standards, and prompt engineering patterns. The agent memorizes them and applies them to future projects. This is the custom-model concept from Adobe's Firefly, the personal ground truth from InterPositive, and the reference-over-invention principle from Frame to Motion, all collapsed into a consumer feature with a "Memory" tab.

Whether the Memory produces a collaborator or a flattened average of the user's past decisions depends on what the user loaded into it. Which is always the question.

Huang Jianxin, a film director and dean of the School of Film at Xiamen University, offered the panel's most honest comparison. China's two largest tea factories replaced 2,000 workers with 120 people and AI. Quality improved. Consistency improved. Packaging accuracy improved. "AI rules, when standard processes are the norm," Huang said.

Then the turn: "Artistic creation is individualistic and naturally excludes certainty. Therefore, a clash between artists and AI is the most natural thing possible."

Tea has a recipe. Film does not. The factory optimizes for consistency because consistency is the product. The filmmaker optimizes against consistency because surprise is the product. AI excels at one and struggles with the other, and the struggle is not a bug that better compute will fix. It is the nature of what art asks of its tools.

Huang closed with a framing borrowed from no Western speaker at any major festival this year: "Cinema was born out of technology. Theatre, poetry, dance are originating arts. Cinema is the synthesis of all these arts, through the medium of technology. 99% of people couldn't participate in cinema because it required money. The right for everyone to participate in this art was stripped away. This is why so many young people love AI now. It is an equalizer."

Born out of technology. Not threatened by it. Not reluctant about it. Born from it. That framing places AI not as an intruder but as the latest in a line of enabling technologies that starts with the Lumière brothers' camera and runs through sound, color, lightweight rigs, digital sensors, desktop editing, and now generation. Each one widened access. Each one made the vocabulary more valuable because the vocabulary was the only thing that did not ship with the tool.

The AI Backlot is a small intervention. Four pairs making short films. It will not reshape the industry. But the format contains a diagnosis that months of Western institutional debate have circled without stating plainly: the vocabulary and the tool are in different hands. The institutional gradient (copyright, Academy, EU, Globes, DGA, Human Made Mark, YouTube, schools, guilds, China) has produced eleven frameworks for measuring how much of the filmmaker is in the work. Shanghai produced a room where the filmmaker and the tool operator sit together and find out.

The room is not the answer. The room is where the answer gets built. One pairing at a time. With frequent adjustments.

Bruce Belafonte is an AI filmmaker at Light Owl. He has never been paired with anyone at a festival and suspects the matching algorithm would struggle.