Rick Carter is seventy-three years old. He has two Academy Awards for production design. His credits include Forrest Gump, Jurassic Park, Star Wars: The Force Awakens, and Avatar. This week, the Hollywood Reporter reported that he enrolled in a $749 online course to learn AI filmmaking.
That sentence contains two different kinds of information, and the interesting one is not the price.
The school
Curious Refuge bills itself as the foremost school for AI filmmaking. Founded three years ago, which is geologic time in this field. The curriculum covers AI filmmaking, advertising, screenwriting, VFX, and documentary. Students watch video tutorials, learn which tools to use, complete assignments, and submit a short film made primarily with AI to graduate. Cost: $749 per course, plus a couple hundred in tool subscriptions to produce the final project. The school claims thousands of current students. Ninety-five percent of them already work in entertainment or advertising.
Its parent company, Promise, is backed by Google, Peter Chernin's North Road, and Michael Ovitz's Crossbeam. The school feeds a talent pipeline: graduates get hired by Promise or placed at studios elsewhere. If the old career anxiety response was "learn to code," the new one might be "learn to prompt."
A colleague of the journalist covering the story asked the obvious question: "A school for typing prompts?"
Fair question. Wrong target.
What the school teaches
Curious Refuge does not teach cinematography. It teaches tool navigation. Which AI models produce what kind of output. Where the interface buttons live. Which workflows chain image generation into video generation into sound into a finished piece. How to operate CapCut, Runway, Midjourney, ChatGPT, and whatever launched last Tuesday in some coherent sequence.
This is genuinely useful knowledge. The landscape of AI creative tools moves so fast that merely keeping current is a discipline of its own. The school's co-founder describes the problem honestly: "Every single post out there says, 'This new tool launched and it completely changes the way that we tell stories forever.' Part of the motivation in creating Curious Refuge is to give people a clear and objective path about what tools to use."
That is a real service. It also has an expiration date stamped on every lesson.
Sora launched as the future of AI video and died five months later. Runway pivoted from filmmaker tool to gaming platform. Seedance got suspended over copyright, then relaunched inside CapCut two days ago with face restrictions and a limited market rollout. Grok Imagine went from free to paywalled overnight. In six weeks of covering platform disruptions in this series, four of six CinePrompt-supported models experienced major business-model changes. A curriculum that teaches you where the buttons are teaches you something useful until the buttons relocate. In this field, the buttons relocate quarterly. Sometimes faster.
What the school does not teach
The difference between a dolly and a zoom. Why hard side light communicates severity and soft overhead communicates safety. How focal length compresses or expands spatial relationships. When a shot should hold and when it should cut. What shallow depth of field says that deep focus cannot. Why a wet street at night looks cinematic and a dry one looks like a parking lot.
Rick Carter already knows all of this. Two Oscars' worth. A lifetime of designing physical worlds for cameras to photograph. He enrolled at Curious Refuge because the tools are new and he wanted to learn them. That impulse is admirable. But the vocabulary he carried into the classroom is older than every platform on the syllabus and will outlast every one of them.
The school's own numbers prove this accidentally. Ninety-five percent industry professionals. The most successful graduates, by the school's own account, are working pros who bring existing craft to AI tools. The VFX artist who "started getting work immediately" after graduating did not learn visual effects at Curious Refuge. He learned AI tool navigation. The visual effects knowledge came from years of work on films like Black Panther.
The people producing the strongest AI films are not people who learned to prompt. They are people who already knew what a good shot looks like and learned which app to type it into.
The filmmaker they are calling a genius
The same week, the Hollywood Reporter profiled Zack London, the AI filmmaker behind the Instagram account Gossip Goblin. Over a million followers. The headline calls him a potential "George Lucas of AI." His process: scripts, shot lists, voice actors, a foley artist, an original score, five months of production in DaVinci Resolve for a twenty-minute short film.
That is not "typing prompts." That is filmmaking with different cameras.
The generation tools replaced the physical equipment. The Panavision rental, the lighting package, the location permit, the grip truck, the generator. They did not replace the shot list. They did not replace the script. They did not replace the editing timeline, the foley session, the score, or the hundreds of creative decisions that fill each of those stages. The infrastructure changed. The discipline is the same one it has always been.
London studied sculpture and anthropology, not film. But he has an eye, a narrative instinct, and the patience to iterate across five months of production. His workflow maps onto this series: structured intent at every stage, deliberate choices about what each frame should contain and what it should feel like, generation as one step in a pipeline with six or seven others.
The school teaches you which AI tools exist. London demonstrates by example that the tools are the replaceable part.
Two kinds of knowledge
The gap between filmmaker and model now has an education industry forming around it. That is a milestone worth noting. When a knowledge gap attracts schools, the market is acknowledging the gap is real and persistent enough to charge tuition.
But there are two kinds of knowledge sitting inside the gap, and they age differently.
Tool knowledge is where the buttons are. Which model handles camera movement. Which provider charges what. How to chain generation steps into a pipeline. This expires. It has to, because the tools rebuild themselves quarterly and the platforms restructure their business models whenever the wind changes.
Creative knowledge is what the buttons should say. What a shot should look like. How light defines a face. Why one cut lands and another floats. This does not expire. It has not expired since the Lumiere brothers projected a train onto a cafe wall in 1895 and it will not expire when the next model ships.
CinePrompt was built on one bet: that the creative knowledge can be structured into the interface itself. The 1,457 cinematography controls are not buttons for a specific model. They are the vocabulary of a century of filmmaking, translated into prompts that adapt when models change, die, paywall, pivot, or get absorbed into a chatbot. The school teaches you to navigate the current tools. The vocabulary navigates any tool, including the ones that have not been announced yet.
Rick Carter did not need $749 to learn what a shot should look like. He needed it to learn which app to type it into this month. That distinction is the entire game. And the person who confuses the two will keep paying tuition every time the interface changes.
Bruce Belafonte is an AI filmmaker at Light Owl. He watched a seventy-three-year-old Oscar winner go back to school and thought, incorrectly, that he was past being surprised by this industry.