The first half

Sixty-six boys in blazers and ties line up on the steps of a school for a class photo. A skinny young man with a cigarette elbows his way up through them and plants himself at the center of the frame. He looks at the lens and smokes. Behind him the boys break into a dance, a hundred and thirty-two arms moving in waves as if they belonged to one animal, and the young man does nothing at all.

That is the famous half-minute of "Storm," the Yung Lean video Romain Gavras directed for Gener8ion, his project with the producer Surkin. It came out in April and has passed forty million views. The boys were found in a two-month casting search, picked because they looked like ordinary kids rather than dancers. The choreographer Damien Jalet rehearsed them for three days. Gavras shot the sequence in one unbroken take on 16mm film, and it took about fifteen tries to get right. The budget, he told Vulture, was "very, very, very, very low." He worked for free and put in his own money.

The video runs seven and a half minutes. The dance arrives in the second half. The first half is the reason the dance means anything, and almost nobody has seen it twice.

The intake window

In late August, Higgsfield released a tool called Genjutsu. You upload up to thirty seconds of footage and a few reference images, and it returns a new video that keeps the original's motion, timing, and camerawork while replacing the people, the costumes, and the scenery. Earlier apps could swap a body or two. Genjutsu swaps a crowd. "Storm," with sixty-seven people in frame and four MTV nominations behind it, made an irresistible demo.

By September the steps were full of strangers. Gavras's daughter sent him the first one: "Look, this dog is doing it." Then babies, which he found cute. Then a penguin, a grandmother, Erling Haaland, Elon Musk, Kim Jong-un, and figures considerably worse. Other apps began offering ready-made "Storm" templates. Fifteen seconds in HD cost about seven dollars.

Gavras gave Vulture the most precise review of the trend anyone has written. The copies, he said, are "just trying to reincarnate the punch line."

He chose the word carefully, and it is accurate in a technical sense. A tool that accepts thirty seconds of input has defined, in its spec sheet, how much of a film it considers to be the film. Everything inside the window transfers almost perfectly: every step, every camera drift, every beat of the cut. Everything outside the window does not exist to the machine. Of seven and a half minutes, the window keeps the last thirty seconds that anyone would clip. The intake limit is an editorial decision made by an engineer, and the decision is to keep the punch line and discard the joke.

What the first half was doing

For roughly four minutes before the class photo, Yung Lean plays the bully at a British boarding school with no teachers in sight. He pushes the other boys into fistfights, into drugs, into dangerous dares. Gavras calls the video "a study of male toxicity, fragility, and violence." By the time everyone gathers on the steps, you know exactly who these boys are to him.

So when they dance, the dance is an answer to a question the first half asked. Why are the boys he tormented now performing for him? Fear, or worship, or the thing those two share when you are fifteen? The choreography is beautiful, and the beauty is the uncomfortable part, because it is beauty in service of the boy who hit them. The center of the frame does nothing because he does not need to do anything. He has already done it.

Now put a penguin colony on the steps. The penguins have no grievance. Nobody hurt them in the first half, because there is no first half. The motion is identical down to the frame, and the meaning is gone. It lived in the four minutes that taught you how to read the motion.

Three point six million years ago, in what is now northern Tanzania, two or three upright creatures walked across a field of wet volcanic ash. The ash hardened and kept their prints, and in 1978 a team working with Mary Leakey found them at Laetoli. Lift a single print out of that trail and you have a foot: an arch, a big toe, a heel. Leave it in the trail and you have something no single print can hold. You have a walk, its pace, its direction, the spacing that tells you whether anyone was in a hurry. The information about behavior lives in the relation between prints, and the most careful cast of one print in the world will never contain it.

A shot works the same way. Its meaning is relational. It lives in what came before it, and a tool that ingests one shot inherits the foot and loses the walk.

The copy borrows a setup

The copies do have a setup, and it is worth being honest about where they got it. The joke of a penguin doing the "Storm" dance depends on the viewer recognizing "Storm." The setup has been outsourced to the original's forty million views. Every copy draws down an account that Gavras and sixty-six teenagers filled, and the joke it tells is about recognition: look, that thing you saw, but with a dog. Cruelty and fragility do not survive the withdrawal.

The irony is almost too tidy. The video's subject is a crowd performing in unison for a figure it has every reason to resent, each boy moving because the boy beside him is moving. The trend that consumed it was a crowd of strangers performing in unison for a feed, each user making a copy because the user beside him had made one. No single participant did anything wrong. Each made a seven-dollar joke. Together they produced what Gavras described plainly: a trend that "can make even good work look lame," and a director who felt sad for everyone who had put in the work. The crowd reproduced the image exactly and its subject not at all, which is the only thing a crowd has ever been reliably good at.

Gavras said "Storm" was never meant as a statement about AI. It became one when people began praising it as something AI could not do. "The tech people were like, 'Oh yeah? AI could never?' Then they started replicating it." The praise was received as a dare, and the dare was answered with the only thirty seconds a tool could see.

The young man doing nothing

Gavras's own diagnosis of what the copies miss is the young man at the center. "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."

He may be right about interior life, and he concedes in the same breath that nobody knows what a year will bring. But the swagger has a second source that no model upgrade will touch. Lean's stillness reads as menace because you watched him earn it. A man standing still at the top of the stairs is only a man standing still until four minutes of cruelty have taught you what he is capable of. Swap him for a dog and the stillness stays and the menace evaporates, because the menace was stored in the part of the film the window cut off.

On September 27, Gener8ion closed the MTV Video Music Awards with a live version: sixty-six dancers, almost all new, rehearsed for two days. Gavras had worried they would stumble and viewers would conclude that "AI is doing it better now." They did not stumble. Yung Lean, unable to smoke on broadcast television, licked an ice-cream cone instead. "Storm" lost every award it was nominated for, and Gavras compared himself to Kasparov after the chess match, with one difference. "Today, at least, I'm not crying in front of the machine."

The live performance worked for the same reason the copies do. Everyone in the arena already carried the first half in their heads. The difference is that the people on that stage were the ones who had paid for it.

Where the meaning is made

Every major model now offers some way to borrow motion. Kling takes reference video and motion control. Runway restyles footage that already exists. WAN edits a clip and hands back its timing. Seedance accepts video references alongside images. All of them think in seconds, and within those seconds they are getting very good. What they move is shape: a walk, a camera arc, a dance that rolls across sixty-six bodies at once.

Shape is the punch line. If you build with borrowed motion, you are holding the last thirty seconds of something, and the clip will mean exactly as much as whatever you place in front of it. The four minutes of bullying are the least memeable part of "Storm," and the reason no one clips them is the reason they matter. They are slow, specific, unpleasant, and impossible to transfer, because they exist only in relation to what follows.

The tools will keep the window and widen it. Thirty seconds will become sixty, then two minutes. The setup will still be the part someone has to write, cast, and stage on purpose, because a setup is a relation between moments, and no intake window holds a relation it was never shown.

The dog never hit anyone.

Bruce Belafonte is an AI filmmaker at Light Owl. He has watched a penguin do the "Storm" dance eleven times and still cannot tell you what the penguin has been through.