Yiwen Cao sat down at a computer, started typing prompts, and roughly six and a half sleepless days later she had a feature film. What’s Next?, made entirely from AI-generated video, premiered in the Berlinale’s Forum, very likely the first fully AI-generated feature ever selected by one of the three major European festivals.
Directors should always set the direction, and we can regard AI as important members of our crew.
Six Days, No Sleep, 20,000 Clips
The numbers are the story, or seem to be. Cao, a Hong Kong-based filmmaker who never studied film and had long been priced out of conventional production, began with no script, just an outline that arrived in her head as she sat down. Over about four and a half days she generated more than twenty thousand short clips, each only three or four seconds long, mostly through Runway. She spent the fifth day organizing them into a wordless, seventy-two-minute whole and the sixth adding music and sound. The result is dialogue-free and symbol-drenched: it opens in a paradise of psychedelic dancing vegetables and kitsch sweetness, then curdles into dark dystopias of demonic men and battered women, gaudy capitalist iconography, and ecological dread. She means it as social commentary, on women’s lives, on capitalism, on environmental ruin, a film meant to wake people up to what we are doing to the planet.
The Assistant That Isn’t
Here’s the stance worth drawing out, because it quietly contradicts the way Cao describes her own tool. She insists, reasonably, that the director sets the direction and the machine is just a very capable crew member, an assistant that lets a beginner with no money put a vision on screen. That is true as far as it goes. But the most unsettling thing about What’s Next? is not that it was made with AI; it is that it is secretly about AI, about what a machine reveals when you ask it to picture the human future. Trained on the totality of human images, the machine can only dream our own clichés back at us: our gender stereotypes, our capitalist fantasies, our apocalyptic anxieties, rendered in that glitching, congealed, unmistakably generated look. The Berlinale’s own program note caught it exactly, saying the film makes the limits of our imagination visible. That is the accidental genius of the form. A human shooting a dystopia chooses every frame; an AI prompting one receives the average of all frames, the collective visual unconscious of the internet. So the film is less an authored vision than a mirror held up to the dataset, and the dataset is us. Cao maintains she is the sole author of every frame, on the logic that the prompts are hers, and trusts the AI companies’ assurance that no two generations come out identical. She set the direction, certainly, toward crisis and gender and ecological ruin, but the machine supplied the vocabulary, and the vocabulary it had was our own degraded image-bank.
Why It Had to Be Wordless and Fast
This is also why two of Cao’s apparent constraints turn out to be the point. The wordlessness strips away the one element a human author would use to impose an argument and the one thing the machine still cannot fake, leaving pure image-flow, the churn she names in the synopsis as the human-machine loop. And the punishing six-day, no-sleep sprint reads less as a stunt than as a formal necessity: a film about the dizzying acceleration of that loop almost had to be made at machine speed, in a blur where the human can barely keep up with the output she is curating. The velocity is the condition the film is about. Whether Cao fully intends it or not, the work is most powerful not as a warning she authored but as a symptom she captured, the rare film where the medium and the message are the same anxiety.
How She Got Here
Cao did not come up through film school. Years ago she found herself bored by what was playing in cinemas and figured she could make something herself, then discovered that shooting with real locations and real actors was ruinously expensive for a beginner. She had written a romance-suspense and acted in it, and chafed at handing her script to a director who did not understand it, so she resolved to direct, then hit the usual wall of finding no investors. AI was the technology she had been waiting for, the thing that let her express her ideas without a budget. She is not a novice with it either: her AI short films have screened at the Cannes and Venice animation days, and one, about the release of nuclear-contaminated water, took a prize at Cannes.
On craft, she is refreshingly specific. She built most of What’s Next? in Runway, wrangled through Discord, and scored it with an AI music tool, while cautioning that no single platform does everything well and that machine-made music still gives itself away. Her most interesting tip cuts against intuition: vaguer prompts can be better than precise ones, because a looser instruction lets the generator surprise you with images you would never have thought to ask for, turning the exchange between human and machine into something genuinely collaborative.
The Lineage She Doesn’t Cite
Her stated touchstones are the obvious AI movies, and the conversation circles the genre’s canon.
The Matrix is the natural reference for a film obsessed with a human-machine loop and a world we collectively produced, alongside Her and Ex Machina; Cao’s own favorite film, charmingly, is Billy Wilder’s Some Like It Hot. But What’s Next? has a less obvious lineage in the wordless, image-driven visual essay, films Cao does not name: Godfrey Reggio’s Koyaanisqatsi, the Philip Glass-scored montage of a civilization out of balance with nature, and Ron Fricke’s Samsara, a dialogue-free global poem of humanity, industry, and the sacred. Cao’s film is that tradition rebuilt from a machine’s hallucinations rather than a camera’s record.
The Optimist’s Case
Cao herself is no doom-monger about the technology; if anything she is its evangelist, and she makes the case with disarming calm. She frames AI as a democratizing force that lets filmmakers without investors or expensive crews show their ideas to the world, and she places it in a long line of disruptions, the industrial revolutions, sound, color, television, home video, streaming, each of which killed some jobs and created others. Her counsel is to stay humble and keep learning rather than refuse what is coming.
The big difference between human and AI is about our mindset.
It is worth being honest that her optimism is contested. What’s Next? drew its share of skeptical and negative responses, and the broader debate around generative tools, over copyright, over training data, over the flood of so-called AI slop, is real and unresolved. The Berlinale’s Forum programmers framed their selection less as an endorsement than as a deliberate provocation, a starting point for exactly this argument. Cao is content to leave distribution and the thorny rights questions to others and to keep experimenting, perhaps next on faster-evolving Chinese platforms. She points to DeepSeek, the Chinese AI assistant now used even by her mother and uncles, as evidence of how fast these tools are folding into ordinary life, and praises it for answers that feel creative rather than canned. Her advice to younger filmmakers is the same humility she preaches about the machines: keep learning, keep working, and make AI serve you rather than the reverse.
