The Gates Are Open

Making films with AI, from the toy to your own show. The Vibe Filming series opens here.

Five weeks earlier, most of the fifteen people in that room had never generated a single frame of video. On July 11th, we sat around a conference table eating donuts and watched all five of their finished shows back to back — a murder mystery on a wedding cruise, a fantasy-crime revenge piece, an anime sci-fi about a burning town, a dark comedy about a golfer who gets away with murder, a children's cartoon about a grandfather's bedtime stories brought to life. Then two guests joined the table: George Huang, who's been a professor at UCLA's film school since long before any of this existed, and Bernie Su, a three-time Emmy winner who's spent a decade making interactive series for YouTube and Twitch. Neither of them teaches our cohort. Neither had seen this specific work before that afternoon. And within the first twenty minutes of sitting down, they'd both said the same thing in two different ways.

Huang said it as history. He told us about Georges Méliès, the filmmaker who shot A Trip to the Moon in 1902 — wrote it, directed it, acted in it, built the sets, hand-tinted every frame with color himself, one at a time. Nobody specialized yet because there was nothing to specialize into. Then movies started making money, and the people funding them looked at what Henry Ford was doing to car manufacturing and asked why filmmaking couldn't work the same way. So it did. Writers became a department. Directors became a department. The person who used to do everything got broken into forty job titles, and somewhere in that breakup a new class of people appeared whose entire job was deciding who got to make anything at all. "That became this institution of gatekeepers," Huang said, "of studio heads, of agents, of power brokers — people who get to decide what is the film we make, what is the film we're going to serve to our audiences." He's spent thirty years inside that system, doctoring scripts for movies you've heard of and plenty you haven't, most recently cutting a film's budget with Luc Besson from fifty million dollars to ten with AI's help. He is not a tourist making a hot take. "It's refreshing to see that now there are tools that are breaking down those barriers, those gates as it were," he told us. "We're almost going back to the invention of film, where you have all the tools you need to do this yourself."

Bernie Su said the same thing without any of the history, because he didn't need it — he just needed to point at what had happened in the room already. "You are more armed today as a single creator, with tool sets and abilities and technology and knowledge, than anybody who's ever had," he said. "So now it's simply up to you." He wasn't being generous. He'd just spent an hour hearing fifteen people who'd never touched this technology explain, unprompted, exactly what went wrong on their shots and exactly how they'd fixed it — one student figured out she could photograph a stranger's laptop screen at an industry event, feed the photo into ChatGPT, and turn it into a reusable "hook analyzer" skill for her scripts. Another spent sixteen takes chasing a shot of a golf ball that the model kept insisting on sinking into the hole. That's not beginner's luck. That's the actual muscle memory of production, acquired in five weeks by people who started with none of it.

Here's the part that made me want to write this down instead of just being glad it happened: neither of them was talking about a toy. Huang has been openly skeptical of AI for most of the time he's used the word — he told us his own students have shot vertical drama pilots without it, because "AI is still a controversial subject among academia." He's watched a filmmaker at Annecy get booed off the stage a few weeks before our showcase for using an AI-trained model to bring hand-painted frames to life, in a technique that literally couldn't have been done any other way. He knows exactly how contested this is inside the industry he's spent his career in. And he sat at that table anyway and told a room of first-time creators that what they'd just done put them closer to the birth of cinema than to the assembly line that followed it. That's not hype. That's a historian telling you which era you're actually standing in.

I built Leyline because I had the paranoid version of this thought two and a half years early. At the end of 2023, right after ChatGPT and Midjourney both broke into the mainstream in the same few months, I became convinced the film industry was going to get hit by something it wasn't ready for — and that it would happen inside five years. It happened in two and a half. I'd had a small independent studio before that, developing IP the traditional way: get representation, wait, pitch, wait more. I know what the old gate felt like from the wrong side of it. What we built is the pipeline for the world on the other side of that gate — the one where a story you've been sitting on for years doesn't need someone else's permission to exist.

But a pipeline isn't a guarantee, and this is the part I don't want to skip past. Removing the gatekeeper doesn't remove the difficulty — it relocates it. Huang put it plainly when one of the students asked how you're supposed to pick up the institutional knowledge that used to come from twenty years of climbing a specialist ladder: he said the ladder itself might not be the answer anymore, and that the actual skill left standing is the one Méliès had — being able to do, or at least direct, everything at once. Bernie Su put the same relocation a different way, watching us debate whether to keep polishing an episode or ship it: "The idea of the gatekeepers are gone, or theoretically they are — that's awesome. So then it's all drive, talent, and your maneuvering." No gate to blame for staying stuck. No greenlight committee to blame if the story doesn't land, either. Both of them, independently, pointed at the same replacement problem: it used to be hard to get permission to make something; now the entire difficulty has moved downstream to getting anyone to watch it. "How do you connect with audiences," Huang said, "is now going to be the challenge."

That's the actual shape of what we're calling vibe filming, and it's the same shape as the vibe coding this company has been running on since before I could name it. Not "AI makes it easy." A non-technical founder and one engineer didn't scale a million-line system because a tool made it easy — we did it because the tool moved the bottleneck from can you write the code to can you tell, fast, whether what got written is actually right, and then we built the discipline to answer that second question over and over without the system falling over. Vibe filming moves the same bottleneck the same way. The hard part was never pointing a camera or, now, writing a prompt. The hard part was always deciding what's worth pointing at, noticing when a take is wrong for a reason you can't yet name, and doing that a thousand times without losing the thread of the story you set out to tell. Fifteen people in a room for five weeks just proved that skill transfers faster than anyone assumed, guided by two people who make their living inside the industry that's supposed to be threatened by exactly that proof.

Cohort two starts late August. If Huang is right about the history and Su is right about the math, the next five weeks belong to whoever shows up and starts.