Scorsese Made the AI Debate Serious
What Black Forest Labs, FLUX and a master filmmaker’s storyboards actually tell us about cinematic intelligence
Martin Scorsese using AI for storyboarding is a significant moment, but it is not the end of the argument. It does not prove that generative AI has suddenly become cinema, and it does not prove that the old craft of cinema has been rendered obsolete. It proves something more specific, and for that reason more interesting: one of the great living filmmakers has publicly treated an AI image model not as a toy, not as a gimmick, and not as a replacement for filmmaking, but as a possible extension of the language of preproduction.
That distinction matters because the public conversation around AI in film still suffers from a chronic lack of proportion. Each new development is either inflated into a technological liberation myth or condemned as the next step in the industrial death of art. The Scorsese case deserves better than both reactions, partly because Scorsese is not a random celebrity endorsement, and partly because the actual use case is much narrower than some of the language around it suggests. The problem is not that the announcement is meaningless. The problem is that its meaning becomes distorted when a practical workflow is turned into a grand historical verdict.
The basic facts are public. Black Forest Labs, the Freiburg company behind FLUX, announced that Scorsese is joining the company as an advisor. The company frames the collaboration around “visual intelligence for cinema” and shows Scorsese in a working storyboarding session with FLUX. In his statement, Scorsese describes a very old directing problem: how to communicate the image inside the director’s mind to the cast and crew before that image becomes expensive reality. He says that he has created his own storyboards for decades, and that the tool allows him to share what he is visualizing more clearly and efficiently with the production designer, art designer and cinematographer. He also describes the ability to visualize and immediately share a storyboard as “creatively freeing.”
That is not a trivial claim. Anyone who has worked around film production understands the importance of getting an image out of one person’s head and into a form that other people can challenge, enrich, price, schedule, light, build, shoot and edit. Storyboards, concept art, animatics, look books, mood boards, previs and production design references exist because cinema is collaborative, material and brutally concrete. Before a shot becomes a shot, it must become a shared object of discussion. That object does not have to be final, but it has to be visible enough to organize decisions around it.
This is where Scorsese’s interest in FLUX becomes easy to understand. He is not describing AI as a replacement for actors, writers, designers, cinematographers or editors. He is describing it as a way of communicating visual intention during preproduction. That is a meaningful use case, especially because it comes from a filmmaker who has spent his life thinking through image, performance, montage, memory and cinema history with unusual seriousness. Yet the usefulness of the tool at this stage does not automatically justify the much larger conclusion that some people seem eager to draw from it.
The debate is not over because Scorsese used FLUX. In some ways, it has become harder to avoid.
The phrase that needs the most care is “cinematic intelligence.” It sounds attractive, and in Scorsese’s mouth it has weight, because he is not a marketing person borrowing cinema as decoration. He has earned the right to talk about the intelligence of the medium. But the phrase becomes dangerous if it is transferred too quickly from the filmmaker to the model. A generated image may look cinematic. It may contain lens language, atmosphere, costume, texture, light direction, scale and emotional suggestion. It may even be more immediately persuasive than a rough pencil sketch. But cinematic intelligence is not the same as producing an image that resembles cinema.
Cinema is not the surface of a frame. Cinema is the relation between image, duration, movement, body, rhythm, performance, memory, cut, sound, intention and consequence. It is not only what a frame looks like, but why that frame exists at that point in the scene, what it allows the actor to do, what it withholds from the audience, what it prepares in the next cut, and what kind of moral, emotional or psychological pressure it creates. A model can offer visual possibilities. It can accelerate exploration. It can put options on the table. But it does not understand why one option matters more than another inside a living film.
That judgment still belongs to the filmmaker and to the collaborators around the filmmaker. This is the central point that should not be lost in the excitement. The model does not become Scorsese because Scorsese uses it. The model becomes useful because Scorsese brings cinematic memory, taste, discipline and selection to it. He knows what to ask, what to reject, what to ignore, what to refine and when to return the work to the human conversation of production. The intelligence is not located in the production of images alone. It is located in the ability to decide what an image is for.
This is also why the Scorsese example is more serious than most AI hype. It does not ask us to admire the machine in isolation. It asks us to examine what happens when a very experienced director uses a generative system at a precise point in the production process. The business context around this development matters too, because AI in film is not only a question of imagination. It is also a question of power. The same technology that can help one director communicate with a production designer can also be used by a company to reduce departments, replace labor, compress development time, weaken craft roles or turn human interpretation into a cost center.
That does not mean every use of AI is morally identical. It means every use of AI has to be located inside a production system. The tool does not arrive as pure possibility. It arrives inside budgets, contracts, hierarchies, rights structures and anxieties about who will still be paid to make images tomorrow.
The labor context is therefore not an optional footnote. The Writers Guild of America describes AI as a key issue for writers and says the 2023 agreement established significant protections around its use. These were not theoretical concerns invented by people who dislike technology. They were contractual responses to a real shift in power, especially around authorship, consent, compensation and the possibility that previous human work could become raw material for future automated production.
That is why it is intellectually weak to say that the debate is over. It is not over. It has only become more specific.
It is no longer enough to ask whether AI belongs in cinema as a broad yes or no question. That question was always too blunt. Cinema has always absorbed tools, and some of those tools transformed not only production methods but the grammar of the medium itself. Sound, color, optical compositing, motion control, digital editing, CGI, digital intermediate, performance capture, virtual production and real time rendering all changed what filmmakers could imagine and how they could organize the work. Scorsese himself points to his use of 3D in Hugo and de aging technology in The Irishman as part of his wider openness to technical evolution.
But historical continuity should not become lazy equivalence. Not every new tool changes cinema in the same way, and not every change has the same ethical meaning. A camera changes what can be seen. A digital compositor changes what can be combined. A virtual production wall changes how space and light can be coordinated. A generative image model changes the speed and authority with which visual proposals appear before they have passed through the slower filters of research, drawing, discussion, design and production reality. That difference deserves careful attention.
The seduction of generative imagery is that it often arrives looking more finished than it is. A rough storyboard carries its incompleteness honestly. Its lines tell the team that this is a proposal, a direction, a sketch of intention. It leaves space for the production designer to interpret, for the cinematographer to disagree, for the actor to complicate the blocking, and for the editor to later discover that the scene wants to live differently. A generated frame can be more visually persuasive, but that persuasion can become a trap if the image begins to feel like a decision before the thinking behind it has matured.
This is one of the most important production questions raised by AI storyboarding. Does the generated image open the conversation or close it too early? Does it help the director communicate intention, or does it quietly become a visual command that everyone else is expected to execute? Does it help the production designer build on an atmosphere, or does it reduce design to matching a prompt driven reference? Does it help the cinematographer understand the emotional ambition of a scene, or does it replace the cinematographer’s interpretive role with an already polished mood?
The answer will not be decided by the model. It will be decided by production culture.
That is also why other filmmakers’ reactions matter. The current landscape is not divided neatly into enlightened adopters and frightened traditionalists. It is much messier than that. At AI on the Lot, the Los Angeles Times reported, more than 2,400 filmmakers and executives gathered to discuss how AI is reshaping Hollywood’s identity, and Paul Schrader moved between fascination, provocation and alarm. His comments were not calm policy language. They were closer to gallows humor from someone who understands both the lure and the danger of the machine.
The Guardian’s recent coverage of Ash Koosha’s Dreams of Violets shows another side of the same transformation. Koosha’s AI made feature, accepted at Tribeca, is framed not as a studio replacement strategy but as an attempt to make a politically urgent film under conditions where conventional production might have been impossible. This is a very different case from Scorsese’s storyboard workflow, but it demonstrates why the conversation cannot remain abstract. AI is entering cinema through multiple doors, from master directors’ preproduction language to independent filmmakers trying to make impossible films with almost no money.
The European dimension of the Black Forest Labs story is also real, but it should not be turned into a fairy tale. It matters that a company based in Freiburg has become visible at this level in a field often dominated by American and Chinese infrastructure narratives. It matters that FLUX is not just another imported layer in someone else’s platform economy. TechCrunch reported that Black Forest Labs raised 300 million dollars at a 3.25 billion dollar valuation and listed investors including Salesforce Ventures, a16z, NVIDIA, Creandum, Earlybird VC, BroadLight Capital, General Catalyst, Temasek, Canva and Figma Ventures.
But this is not a pure sovereignty story. Black Forest Labs may have a German center of gravity, but it is also a globally financed, commercially ambitious AI company operating inside international technology and media networks. That does not make the achievement smaller. It makes it more accurate. Europe should be glad when serious AI infrastructure is being built on the continent, but it should not confuse a Freiburg address with cultural independence from global capital, global platform economics or global entertainment interests.
Accuracy matters because the AI conversation is already overloaded with myth. The pro AI side often behaves as if every new tool automatically expands human creativity, while the anti AI side often treats every use case as if it were identical to theft, replacement and cultural collapse. Both positions are too simple. A director using FLUX to communicate a scene is not the same as a studio using synthetic performers to avoid paying actors. A production designer using a generated image as one layer in research is not the same as a company replacing concept artists with prompt labor. An AI assisted storyboard is not the same as an AI generated final film. But these differences must be argued honestly, not used as camouflage.
The ethical question is not merely whether AI was involved. The ethical question is what process changed, who gained power, who lost leverage, what material the model depends on, what rights were respected, what labor was displaced, what authorship was clarified or blurred, and whether the final work became more thoughtful or merely more efficient. This is where the conversation has to move if it wants to become serious. Not toward slogans about inevitable progress. Not toward blanket rejection. Toward a craft based understanding of where intelligence actually sits in the production chain.
Scorsese’s example is valuable precisely because it helps locate that intelligence. His use of FLUX is not a demonstration that the model understands cinema. It is a demonstration that a filmmaker with a deep internal cinema can use a model to externalize part of that cinema earlier in the process. The tool becomes meaningful because it enters a disciplined human practice. Without that discipline, it is merely a machine producing plausible images from a culture it has absorbed. With that discipline, it may become a new kind of sketchbook, a faster form of visual conversation, or a preproduction instrument that helps a team find a shared direction sooner.
That possibility should not be dismissed. There are real advantages here. A director who cannot draw may communicate a composition more clearly. A small production may test visual ambition before spending money it does not have. A cinematographer may see the emotional intention of a scene earlier. A production designer may receive not only words but atmosphere, scale, proportion and light direction as a starting point for discussion. For filmmakers working outside the most privileged production environments, this kind of visual articulation could become genuinely empowering.
But every advantage has a shadow. The same speed that helps a director communicate can also pressure others to accept the first impressive image. The same visual clarity that reduces misunderstanding can reduce creative ambiguity too soon. The same cheapness that helps independent filmmakers can also tempt producers to squeeze craft departments harder. The same abundance of options that seems liberating can produce a culture in which nothing is felt deeply because everything can be regenerated instantly. In film, friction is not always waste. Sometimes friction is where the work becomes intelligent.
A film is not made by executing a private image exactly as it first appeared in someone’s head. A film is made by transforming that image through bodies, places, materials, accidents, disagreements, money, weather, performance, timing and the stubborn resistance of reality. The best collaborators do not merely receive the director’s intention. They test it, misunderstand it productively, improve it, contradict it, protect it from its own weaknesses and return it in a richer form. If AI helps that process begin earlier, it may strengthen cinema. If AI makes the first visual proposal feel too complete, too authoritative or too cheap to question, it may weaken the very collaboration it claims to support.
That is the professional test hidden inside the Scorsese announcement. When the generated storyboard appears, does the room become more alive or less alive? Does the image invite interpretation, or does it end interpretation? Can the production designer still say that this is not the right world? Can the cinematographer still say that the scene should breathe differently? Can the actor still discover behavior that breaks the planned composition? Can the editor later abandon the storyboard because the scene found another truth?
These questions matter because they separate a tool from a command structure. The same image can function as an invitation or as an instruction, depending on the culture around it. In the hands of a serious filmmaker, it may become a way of beginning a richer conversation. In the hands of a nervous production system looking for speed, certainty and savings, it may become a way of reducing the number of people allowed to think.
This is why Scorsese made AI serious. Not because he settled the debate, and not because his endorsement turns generative AI into an artist. He made it serious because his use case is concrete, professional and difficult to dismiss. It is neither stupid nor apocalyptic. It does not fit the easy caricature of AI as empty content sludge, and it does not justify the opposite fantasy that models now possess the soul of cinema. It sits in the uncomfortable middle, where most important technological changes actually happen: inside a real workflow, with real benefits, real risks and real consequences for how people work together.
A more mature reading of this moment would therefore resist both cheap conclusions. Scorsese using AI does not mean that AI now simply belongs in cinema, as if one great name could settle an industrial and ethical debate. It also does not mean that cinema has betrayed itself, as if the medium had ever been pure from technology. What it means is more demanding: generative tools are beginning to enter the working language of serious filmmakers, and the industry now has to decide what kind of language that will become.
Will it become a language of clearer intention, richer collaboration and more precise preproduction? Or will it become a language of premature certainty, reduced labor and polished surfaces that arrive before the hard thinking has happened? Will AI help filmmakers communicate the image they are searching for, or will it encourage production systems to mistake a convincing visual proposal for a finished artistic decision? Will it expand the room, or quietly shrink it?
The most important thing to protect is not nostalgia for old tools. It is the human chain of responsibility that turns images into cinema. AI can help make an intention visible. It cannot decide whether that intention is worth following. It can produce variations. It cannot know which variation carries the scene. It can imitate atmosphere. It cannot be accountable for meaning. It can make the director’s first image easier to share. It cannot replace the living process by which that image is challenged, transformed and finally made.
Scorsese’s involvement matters because it prevents the AI question from remaining abstract. It shows that generative imagery is no longer only a tech demo, a culture war topic or a speculative threat. It is entering the working vocabulary of filmmakers who understand cinema deeply. That should neither terrify us into rejection nor seduce us into worship. It should force better thinking.
The future question is not whether AI can generate cinematic images. It can.
The real question is whether filmmakers, producers and institutions can still tell the difference between an image that looks like cinema and a process that deserves to be called cinema.