Martin Scorsese spent the back half of his career insisting cinema is a craft, not a content pipeline. So when he signed on as an adviser to Black Forest Labs this month and turned up in a video using their FLUX model to storyboard his next feature, the reflex take was that something had broken. It hadn't. Look at what he's actually doing with it, and the story is smaller and more useful than the headline.
The film is What Happens at Night, a drama with Leonardo DiCaprio and Jennifer Lawrence. Scorsese isn't generating footage. He's generating storyboards, still frames he uses to show his production designer, art director, and cinematographer what's in his head before a single set gets built. "During the pre-production process, time costs money," he said, "and this allowed us to move faster without sacrificing quality or craft." That isn't a manifesto. It's a line producer's sentence.
Pre-production was always the soft entry point
Everyone arguing about AI in film has been arguing about the wrong frame. The loud fight is over final pixels: will audiences accept generated performances, will actors be replaced, can a model hold a face across a feature. Those are real questions, and the honest answer to most of them in 2026 is still "not reliably." But final output was never where AI was going to land first. Pre-production was.
Storyboards and previz have a forgiving error budget. Nobody buys a ticket to see a storyboard. A board that's 80 percent right is enormously more useful than no board at all, because its job is to align a department, not to ship. A smudged proportion or an inconsistent prop costs nothing when the artifact is a conversation starter that gets thrown away once the shot is designed. That's exactly the tolerance band where current image models are strong and current video models are still shaky.
What changed isn't the tool, it's who's holding it
Directors have used AI-adjacent previz for a while now. The interesting part of the Scorsese news is the source. Here's a filmmaker with nothing to prove and every incentive to stay clear of a polarizing technology, and he put his name on it anyway, framed not as the future of art but as a faster way to communicate a shot. He even reached for his own precedent, noting he used 3D for Hugo and de-aging for The Irishman. The throughline is that pre-production tools are tools, and a director who can show instead of describe saves his department weeks.
That reframing matters more than another model release. When the conversation is "AI replaces the crew," crews resist, correctly. When it's "the director can hand the art department a clearer reference on day one," the same crew leans in, because it makes their work better and their iterations cheaper. The Scorsese partnership is a permission structure as much as a product placement.
The previz gap is where the value actually is
Most independent productions skip proper storyboarding entirely. Not because it's useless but because a good storyboard artist is expensive and slow, and the budget gets spent on things that end up on screen. The result is a lot of films that walk onto set without a shared visual language, and you can see it in the footage. Departments guess. Coverage gets reshot. Days slip.
Closing that gap is unglamorous, and it's where generated imagery earns its keep right now. A director who can produce fifty reference frames in an afternoon, consistent enough in tone, lighting, and blocking to brief a crew, has changed the economics of pre-production without touching the final image at all. This is the same logic behind treating a prompt as the start of a real pipeline rather than a slot-machine pull. The value is in the structure around the generation, the consistency and the orchestration, not the single pretty frame. It's why a previz-to-shot workflow, where the boards actually carry into the shoot, beats a folder of disconnected stills. Promvie is built on that idea, the pipeline, not the lone clip.
What to do about it
If you're making something this year, sort your AI use by error budget. Use generated imagery freely where being roughly right is enough: mood boards, storyboards, location and costume references, blocking diagrams. Keep your skepticism for final footage, where consistency and performance still break under scrutiny. Brief your collaborators with boards, not adjectives. The point isn't to look like Scorsese. It's to walk onto set knowing what the film looks like.
The headline says a legendary director embraced AI. The actual news is quieter and more durable: the part of filmmaking AI is genuinely good at is the part nobody was fighting over.