The Director Caught in the AI-Slop Crosshairs

“Storm” was a music video AI couldn’t fake. Then came a tool custom-made to copy it into infinity.

Sixty-six boys, all wearing ties and blazers, line up on the steps of a school as if they’re posing for a class photo. Out in front is Yung Lean, the Swedish rapper, who drags on a cigarette, elbows his way up the stairs, and plants himself at center frame. Then the group around him breaks into dance — twisting, buckling, and bouncing in waves with 132 arms moving as if they belong to the same animal. Lean does almost nothing besides look at the camera and smoke. 

This is “Storm,” a seven-and-a-half-minute music video directed by Romain Gavras for Gener8ion, his audiovisual project with French producer Surkin, who does the music (Gavras makes the films; guests like Lean supply the vocals). For the video’s first half, Lean plays the bully at a British boarding school where, with no teachers in sight, he pushes his classmates into fistfights, drugs, and dangerous dares. Then comes the class photo, and the boys he’s been terrorizing turn into his backup dancers. (Lean also stars in Gavras’s new film, Sacrifice, which is out on Netflix this month.)

“Storm” is a throwback to the 1990s, when optical illusionists like Chris Cunningham and Michel Gondry directed music videos that forced viewers to doubt their own eyes. It’s also, in 2026, the kind of impossible-seeming spectacle you might think was made with AI. But the schoolboys are real, found in a two-month casting search and picked because they looked like ordinary kids instead of professional dancers. Choreographer Damien Jalet rehearsed the group for three days, and Gavras filmed the sequence in one unbroken take on 16-mm. film. It took about 15 tries to get right. The budget, Gavras says, was “very, very, very, very low”; Iconoclast, the production company he works with, covered some of the costs, and Gavras worked for free and chipped in his own money. Knowing what a pain in the ass it must’ve been to shoot makes it better. When “Storm” debuted in April, it seemed like a celebration of all the things AI supposedly couldn’t fake: sweat, luck, and the swagger of a guy who knows how cool he looks standing still. 

So then, of course, people started faking it. In September, social media was flooded with clips that re-created a half-minute from “Storm” — Lean’s cigarette, his walk up the steps, the dance explosion behind him — with the same choreography and camera moves but with Lean and his classmates replaced by strangers, pets, celebrities, and worse. “My daughter sent me the first one I saw,” says Gavras. “She was like, ‘Look, this dog is doing it.’” For a little while, he thought it was funny: “It was cute — dogs and then babies.” But soon it was everybody — a penguin, a grandmother, Erling Haaland, Elon Musk, Kim Jong-un, Hitler, Jeffrey Epstein — and the joke wore off. “It made me sad for all the people who put in so much work on our video,” Gavras says. “A trend like this can make even good work look lame. Part of me wanted to quit filmmaking, move back to Greece, and become an olive farmer.” 

Many of the clips were made with a tool called Genjutsu, which was released in late August by the AI video company Higgsfield. Apps that let you insert yourself into a video have been around since at least 2024, when one called Viggle turned a clip of Lil Yachty walking onstage into a meme template, but most could replace only a body or two at a time. Genjutsu can replace an entire crowd, including the 67 people in “Storm,” without a hiccup. A user uploads up to 30 seconds of footage and a few reference images, and the tool spits out a new video that preserves the original’s motion, timing, and camerawork while swapping out the people, costumes, and scenery. “Storm,” which had just been nominated for four MTV Video Music Awards, including Video of the Year, made for an irresistible demo. As the trend took off, other AI apps began offering ready-made “Storm” templates of their own. 

Gavras insists “Storm” was never intended to be a statement about AI. “It’s a study of male toxicity, fragility, and violence,” he says. Elaborate practical filmmaking is just his M.O. His 2012 video for M.I.A.’s “Bad Girls” features stunt drivers in Morocco balancing cars on two wheels; his 2022 film, Athena, opens with an 11-minute single-take shot of youths storming a police station and escaping to their housing project miles away. But once people started calling “Storm” something AI couldn’t do, AI companies seemed to take it personally. “The tech people were like, ‘Oh yeah? AI could never?’ Then they started replicating it. So we were forced into this conversation. It’s a nightmare.”

Gavras says a few AI companies approached him about licensing “Storm,” but he turned them down: “There’s an economic model coming that we didn’t want to be a part of.” He isn’t fussy about commerce; he pays for projects like “Storm” by directing ads. “I’m a mercenary when I do commercials,” he says. “But ‘Storm’ was supposed to be pure and I can’t stain it.” Which leaves him in a hard spot. Taking the money would feel like selling out. Turning it down doesn’t make the copies go away; it means somebody else gets paid. (Users can pay about $7 to make a 15-second HD Genjutsu video.) 

Why can’t he make the copies stop? Gavras says that’s a question for his lawyers. The choreography in “Storm” is almost certainly protected by copyright; in 2023, a federal appeals court revived a choreographer’s lawsuit against Epic Games over a couple seconds of a dance routine that he alleged showed up in Fortnite. “If two seconds of one dancer got past the motion to dismiss,” says Aaron Moss, a copyright lawyer in L.A., “30 seconds of dozens of dancers moving in formation should get past one easily.”

The hard part is figuring out who to sue. Copyright law usually blames whoever presses the button — if you photocopy a copyrighted photo, Moss says, “you’re the one that’s liable for infringement, not the person who made the copy machine” — and with Genjutsu, the button-pressers are the users who upload “Storm.” One way to hold a company responsible is to show it actively encouraged the copying by, say, using the dance in its marketing, which may be what those AI companies wanted a license for. Otherwise, Gavras is left chasing thousands of anonymous users in dozens of countries. “The practical ability to sue over most of these copies is close to zero,” Moss says. 

Even if Gavras found someone to sue, he could be an old man, or an olive farmer, before he got any satisfaction. On September 29, a federal appeals court issued the first ruling on whether training an AI model on copyrighted material is fair use (it wasn’t, at least this time) in a case Thomson Reuters filed in 2020, before ChatGPT existed, against a legal-research start-up that went out of business the following year. Meanwhile, AI companies seem to find exciting new ways to bend copyright law every day. Recently, ChatGPT was caught signing fake New Yorker cartoons with the names of real cartoonists, and an AI-generated comedian with an OnlyFans account was accused of stealing stand-up jokes; it racked up more followers than the comedian who wrote them. “It feels like watching a huge swarm of locusts descend on a field and eat absolutely everything there,” says James Grimmelmann, a professor at Cornell Law School. “They’re hungry, and they’ll move on, and it will be some other field tomorrow.”

An AI joke thief or a penguin doing the “Storm” dance may seem silly, but it’s a glimpse of a future in which people can lift an artist’s style, or even their entire personality, as easily as they once downloaded mp3s. Gavras finds it “fascinating and depressing — it’s like Prometheus stealing the fire.” Still, he doesn’t think the machines have completely pulled it off. AI can copy “Storm” but not make it; the knockoffs are “just trying to reincarnate the punch line,” he says. “They’re missing the swagger of Yung Lean. He looks like a dangerous baby. There’s something in the imperfection of humanity, and the interior life of someone who dances or even someone who just smokes a cigarette, that’s not quite there yet. But who the fuck knows what will happen in a year.” 

In the meantime, Gavras settled on the only response available. Gener8ion had been invited to perform at the VMAs, and he’d been on the fence about it. “Storm” would be hard to re-create live, and he wasn’t sure he belonged at an MTV awards show. As the copies spread, however, he changed his mind. He and Jalet auditioned 66 dancers, almost all new, and rehearsed them for two days. The danger, he says, was that they’d screw up and viewers would decide that “AI is doing it better now.” On September 27, in L.A., Gener8ion closed the show with a dance sequence executed just as well as in the video, though Yung Lean had to lick an ice-cream cone instead of smoke a cigarette. “Storm” lost all the awards it was nominated for, but Gavras says that wasn’t the point. “Remember Kasparov crying when he lost the chess match to a machine? Today, at least, I’m not crying in front of the machine.” 

Thank you for subscribing and supporting our journalism. If you prefer to read in print, you can also find this article in the October 19, 2026, issue of New York Magazine.

Want more stories like this one? Subscribe now to support our journalism and get unlimited access to our coverage. If you prefer to read in print, you can also find this article in the October 19, 2026, issue of New York Magazine.