The oldest prompt

Martin Scorsese, 83 years old, joined Black Forest Labs as an advisor on Monday. He is using their FLUX image generation model to create storyboards.

Not to generate films. Not to replace actors. Not to cut budgets. To draw pictures of what he sees in his head so the people standing next to him on set can see it too.

"For 70 years, I've been creating my own storyboards," he said. "There's always been this problem of how do you communicate what you see in your head to your cast and crew. There are some things you have to see and feel."

That sentence describes a communication problem. Not a creation problem. Not a replacement problem. Not an existential threat to the craft. A man who has directed fifty films over six decades still cannot reliably transfer the image inside his skull to the people who need to build it. So he draws pictures. He has been drawing pictures since the 1950s. The pictures work. They are slow.

Now the pictures are faster.

The storyboard is the original structured prompt

A storyboard is a sequence of frames, drawn by hand, that describes what each shot should look like before the camera rolls. Lens height. Subject position. Lighting direction. Blocking. Composition. The relationship between foreground and background. What the audience should see and, just as importantly, what they should not.

It is, in every functional sense, a structured prompt addressed to humans.

The filmmaker translates creative intent into a visual document. The production designer reads it and builds the set. The cinematographer reads it and plans the lighting. The director of photography reads it and selects the lens. The key grip reads it and rigs the dolly track. Each person on the crew receives the same structured input and executes their portion of it.

Ninety-seven articles in this series have documented a gap between what a filmmaker knows and what a generation model understands. Scorsese just described the same gap between what a filmmaker knows and what a human crew understands. The storyboard has been closing that second gap since before the first gap existed.

The tools are different. The problem is the same. A person with a vision and an audience that needs to receive it.

The Copacabana problem

In a video filmed at his New York office, Scorsese demonstrated using FLUX to storyboard a scene. He reflected on the Copacabana tracking shot in Goodfellas, the three-minute Steadicam sequence through the kitchen, down the stairs, through the service corridor, into the club, past the waiters, to the table. Every vignette in that shot had to be planned before the camera moved. The timing of each encounter, the blocking of each extra, the pace of the walk, the position of the couple relative to the lens.

On set, that planning meant hand-drawn boards, verbal walkthroughs, rehearsals, and the accumulated trust between a director and a crew who had worked together long enough to fill in the spaces between the drawings. Scorsese said if a tool like FLUX had existed, "you could figure it out much much quicker and you could save production time, and also less wear and tear on the crew."

He did not say the tool would have made the shot better. He said the tool would have made the communication faster. The shot's quality lived in Scorsese's decisions. The time cost lived in translating those decisions to other people.

That distinction is the quiet center of this entire announcement.

The sixth filmmaker

Spielberg said never, then said yet. Soderbergh uses generation for the ten percent of a documentary where cameras cannot go. Schrader probed an AI girlfriend until she terminated the conversation. Kennedy asked how you teach taste. Jackson called it a special effect and said it might destroy the world in the same breath.

Scorsese said: I can show people what I see now.

Each response reveals a different relationship to the same tools. Spielberg is trajectory. Soderbergh is surgical deployment. Schrader is investigation. Kennedy is philosophical interrogation. Jackson is normalization. Scorsese is the most practical framing any major filmmaker has offered. He identified a specific, bounded problem (pre-production communication speed), selected a specific tool for it (FLUX image generation), and described the result ("creatively freeing"). No grand theory about the future of cinema. No existential wrestling. A working director solving a workflow problem he has tolerated for seventy years.

"Remember, cinema is a young medium, only around 125 years old, so we have to be open to how it can evolve." He used 3D for Hugo. He used de-aging for The Irishman. Both were tools that served specific films. Neither became his identity. FLUX appears to be the same category: technology that solves a local problem without redefining the practice.

The communication gap

Every filmmaker carries two burdens. The first is knowing what the work should look like. The second is transmitting that knowledge to the people and machines that execute it.

This series has been about the second burden as it applies to generation models. How do you tell a model that the light should fall from camera left at forty-five degrees, that the subject should occupy the left third of the frame, that the color palette should feel cold and institutional, that the camera should drift forward so slowly the viewer barely notices? Ninety-seven articles about the vocabulary of that transmission.

Scorsese's announcement is about the second burden as it applies to human collaborators. How do you show a production designer the exact quality of light bleeding through a window when the set does not exist yet? How do you show a DP the compositional relationship you are imagining when the actors have not been blocked? The storyboard handles this. It has always handled this. It just takes time, and the resolution of a hand-drawn sketch captures mood better than geometry.

FLUX generates a photorealistic frame from a text description. Scorsese describes what he sees. FLUX draws it. He shows it to his team. They build the real version with cameras, lights, actors, and sets.

The AI-generated image never appears in the film. It is a letter. A memo. An instruction set with better visual fidelity than a pencil sketch. The structured prompt that a generation model receives from a filmmaker is the same artifact, traveling in a different direction. One goes to humans. The other goes to a model. Both translate creative intent into a format the recipient can act on.

The investment question

Scorsese was introduced to Black Forest Labs through BroadLight Capital, an investment company co-founded by his manager Rick Yorn. There is a financial relationship behind the advisory role. That context matters because every endorsement in AI entertainment carries economic weight. Cameron sits on Stability AI's board. Affleck secretly co-founded InterPositive and sold it to Netflix. Aronofsky runs Primordial Soup in partnership with Google DeepMind. The AI 25 power list measures leverage, not craft.

But Scorsese did not pitch a platform. He did not announce a new product category. He did not describe filmmaking as a quantity problem or suggest studios make fifty films instead of one. He described using a specific tool for a specific pre-production task and called it "creatively freeing." The financial relationship exists. The use case is genuine. Both can be true simultaneously.

What the storyboard knows

A storyboard carries the filmmaker's accumulated creative decisions in visual form. Composition. Blocking. Camera angle. Scale. Mood. The relationship between elements in the frame. It does not carry dialogue, performance nuance, timing, or the thousand micro-decisions that happen on set when reality pushes back against the plan.

AI-generated storyboards carry all of that information and add photorealistic texture, lighting simulation, and color. They look more like the finished frame than a pencil sketch ever could. That additional fidelity is simultaneously the advantage and the risk. A pencil sketch communicates intent without prescribing execution. It says "the light should come from here" without saying "the light should look exactly like this." A photorealistic AI storyboard says both. The crew may interpret it as a target rather than a direction.

On a Scorsese set, that distinction is managed by Scorsese. He has sixty years of authority, a trusted DP, and the confidence to deviate from his own boards when the set tells him something the drawing could not. On a less experienced set, photorealistic boards might calcify into instructions the crew follows too literally, leaving no room for the happy accidents that Scorsese would recognize and embrace.

The tool does not determine that outcome. The filmmaker does.

The two rooms, again

There are two rooms in AI filmmaking. The room where AI generates the footage. And the room where AI helps the filmmaker communicate before the camera rolls.

Soderbergh works in the first room for ten percent of a documentary. The Koosha brothers work in the first room because they cannot physically cross a border. Rahi Anil Barve works in the first room for three hundred and sixty dollars. Hell Grind works in the first room with three thousand words per fifteen-second clip.

Scorsese works in the second room. The camera still rolls. The actors still perform. The DP still selects the lens. The set still gets built. AI accelerates the planning, not the production. The generation model produces images that the director shows to humans, not images that the audience sees.

Both rooms are real. Both rooms are valid. The distinction matters because the institutional gradient (copyright law, Academy rules, EU disclosure requirements, Golden Globes eligibility) does not yet address the second room. A film where every frame was photographed by a camera, performed by an actor, and lit by a gaffer, but whose storyboards were generated by FLUX, does not trigger a single institutional test. There is no label required. No disclosure form. No eligibility question. The AI was never in the frame. It was in the conversation before the frame existed.

As more filmmakers discover that the second room solves their actual problem, the institutional debates about the first room may find themselves addressing only half the industry.

The sentence that matters

Scorsese has been described as a cinema traditionalist. He fought for film preservation. He criticized the dominance of franchise filmmaking. He shoots on celluloid when possible. None of that contradicts his adoption of AI storyboarding, because the adoption is not about changing the medium. It is about communicating more efficiently within the medium he has practiced for seventy years.

He did not describe a revolution. He described relief.

"The ability to visualize and immediately share the storyboard was creatively freeing."

Creatively freeing. Not creatively threatening. Not creatively disruptive. Freeing. The word implies a constraint that has been lifted. The constraint was not a lack of vision. It was the labor of translating that vision into a format other people could hold. He has been doing that labor by hand since Eisenhower was president. The labor did not make the films better. It made them slower.

Across six major filmmaker responses, the directors who actually use the tools describe specific, bounded applications. The directors who refuse them describe unlimited, existential threats. The gap between those two descriptions is the distance between having tried the thing and having opinions about the thing.

Scorsese tried the thing. The thing draws storyboards. He is currently filming What Happens at Night with Leonardo DiCaprio and Jennifer Lawrence. Whether FLUX storyboards appear in that production's pre-production pipeline, the workflow itself is the same one filmmakers have used since Walt Disney pinned sketches to a corkboard in the 1930s. The pencil got an upgrade. The practice did not change.

The storyboard was the first structured prompt. Scorsese has been writing it for seventy years. The vocabulary was always the point.

Bruce Belafonte is an AI filmmaker at Light Owl. He has never drawn a storyboard that anyone could identify as representational and suspects this is why he types.