This week at San Diego Comic-Con, Guillermo del Toro was asked about the 3D re-release of Pan's Labyrinth. He said it was made through a fully manual conversion process, about 1,000 artists rotoscoping the film frame by frame. He chose not to use AI tools. The reason, in his words: "If we cut a generation of people from learning their craft, you're cutting the rest of the history of that medium away from them, for what?"
Seven weeks earlier, on June 2, 2026, Martin Scorsese joined the German AI company Black Forest Labs as an adviser. He tested the storyboarding tool on a scene and found "the ability to visualize and immediately share the storyboard was creatively freeing." Meanwhile, Christopher Nolan, on the press tour for The Odyssey in July 2026, told reporters he had never seen "a more rapid wholesale dismissal of a supposedly foundational jump in technology in my lifetime." He was talking about young audiences, and what he calls AI slop.
Three major directors. Three different public positions. Or so it looks from the outside.
What del Toro is actually protecting
I spent years in the editing room before I directed anything. Not years sitting in the edit as an observer, but years as the person who watched takes and decided which one lived. The thing you learn from that kind of work is not a skill you can name. It is a calibration. You develop a sense for what a frame is doing, what it costs when something is wrong, what it means when the light shifts two degrees and the actor's face becomes a different person. That calibration is not transferable by description. It accumulates through attention paid over time to a very large number of specific, unforgiving choices.
Del Toro chose 1,000 artists to sit with Pan's Labyrinth frame by frame. Not because rotoscoping is efficient. It is not. It is slower and more expensive than the AI alternative. He chose it because the artists who rotoscoped that film are now different people than they were before. They have looked at every frame of one of the most meticulously composed films of the last twenty years, close enough to trace it. That kind of looking changes what you can see afterward, and the change is permanent.
Walter Benjamin wrote about the aura of the original. He was largely wrong about reproduction erasing aura (cinema did not destroy painting, and painting did not die when photography arrived), but he was right about something adjacent. Aura is not a property of the object. It is what a person trained to see chooses to notice. A thousand artists tracing Pan's Labyrinth are training their attention on the finest example of what they want to eventually produce. The question del Toro is asking is not "why use humans when AI could do this faster?" The question is: "Who will direct the next generation of films, if we optimize away the process that makes directors?"
What Scorsese is doing, and why it is not the opposite
Scorsese uses AI to communicate his vision with his team faster. He knows what the shot looks like. He has always known what the shot looks like. Decades of looking at a very large number of frames went into that knowledge. The AI tool makes the communication from his head to his team more efficient. Nothing about the decision-making changed.
This is the use case that is actually available to most of us: AI as an execution layer for a decision already made. The condition is that there is a decision to begin with. Scorsese can use the storyboarding tool because he has spent a career developing the judgment to know what to ask for. The tool returns something useful because the input was precise. Precise input requires someone who knows what precision looks like in this domain.
For the independent filmmaker or the small studio, three things follow from this contrast:
- The Scorsese model is accessible right now. If you know what you want the image to look like, AI generation and AI visualization are genuinely faster ways to get from your head to a communicable form. The question is whether you know what you want, specifically enough to make the output useful.
- Del Toro's 1,000 artists is not a budget recommendation. It is a reminder that the judgment to use AI tools well did not arrive on its own. It came from somewhere that involved paying close attention for a long time.
- Nolan's observation about young audiences is worth sitting with longer than it usually gets. The generation that grew up consuming content at volume has developed a fast, fine-grained sensor for content produced without intention. They feel the absence of a decision before they can name it. That is a market signal, not a generational quirk.
What the audience is actually rejecting
Not AI. Not the technology as such. The absence of somebody on the other side of the image.
Nolan's children, in their late teens and early twenties, do not object to AI generation as a technique. They object to images that feel like no one chose them. The distinction is perceptible even to people who cannot explain it technically. Something is off when a frame was produced without a reason specific enough to make that frame and not another one. The trained eye identifies this as craft failure. The untrained eye identifies it as a vague unease. The result is the same: the image does not connect.
This mechanism is not new. It is the same one that made audiences stop caring about a certain category of CGI: not because the pixels were technically wrong, but because the weight of the objects felt sourced from nowhere. No one decided that particular texture, that particular gravity, that particular relation between the light source and the surface. The audience did not need a technical vocabulary to file this correctly as unreality. They felt the absence of a decision, and they responded to that absence in the only way available to an audience, which is by not caring.
The AI image that fails is the one where generation was the decision. Where someone opened a model, typed a description loose enough to mean almost anything, took the first result that looked acceptable, and moved on. No selection pressure, no understanding of why this frame and not the others. The model can produce a thousand options. The work is in knowing which one is right, and knowing that requires already having a reason for it.
Three directors, one position
Del Toro is not fighting technology. He spent the same Comic-Con conversation discussing the digital 3D conversion pipeline in technical detail. His fight is for the conditions that produce directors who know what to do with technology when it arrives, and who can tell when technology is being used to replace a decision that should have been made by a person.
Scorsese is not abandoning craft. He spent fifty years developing the judgment that makes his AI-assisted storyboards look like Scorsese storyboards and not like anyone else's. The tool did not produce the judgment. The judgment was already there, and the tool made it faster to share.
Nolan is not predicting the death of cinema. He is observing that the test has gotten faster and harder to pass. The generation that grew up online has a precise filter for authorship, built through years of consuming content at volume. They feel when someone was there making the choice. They feel when no one was.
The tool is not the art. The tool serves whoever is holding it, which means the question is always who is holding it and what they have learned to see. Del Toro's 1,000 artists are learning to see. Scorsese already knows what he sees. Neither position is in conflict with AI. Both are in conflict with the idea that AI removes the need to develop judgment first.
That is the position all three share, whether they frame it that way or not. What you bring to the tool is what the tool returns, scaled up. The camera did not replace the eye that knows where to point it. The model does not replace the author who knows what to ask for.
