How is AI being used in filmmaking at Cannes 2026?

At Cannes 2026, the debate shifted from whether AI will be used in film to how it will be used. Films like Steven Soderbergh's Lennon documentary, which was ten percent generated by Meta's AI tools, played in official selection without controversy. Directors like Xavier Gens are using AI post-production tools that cut visual effects budgets by up to fifty percent and compress timelines from a year to three months.

What is the Human Provenance in Film (HPF) AI disclosure standard?

Human Provenance in Film (HPF) is a free, open-license disclosure standard unveiled at the 2026 Cannes Film Market by The Mise En Scene Company. It defines three designations: No AI Used, Assistive AI, and Generative AI. The standard is designed to travel through existing film sales and distribution paperwork and can be adopted by any company without fee under a Creative Commons license. It addresses the vast middle ground of filmmaking — de-aging, wire removal, compositing — that binary AI frameworks ignore.

Why does cinematographic vocabulary matter for AI video generation?

Cinematographic vocabulary — knowing what a shot should look like, why to choose a specific lens, and what lighting means creatively — is what allows filmmakers to direct AI generation tools effectively. As Cannes director Thierry Fremaux put it: 'To ride an electric bike, you have to know how to ride a normal bike.' The AI tool handles execution, but without creative vocabulary to guide it, the output lacks intentionality. Precise prompting rooted in film knowledge gives the model less room to guess and keeps creative decisions in the filmmaker's hands.

The question changed -- CinePrompt Field Notes

Reuters filed a dispatch from Cannes today with a sentence that would have been impossible twelve months ago: "The emphasis this year has shifted from whether AI will be used to how."

Not "some say." Not "a minority argues." The emphasis. This year. Has shifted.

The shift did not arrive through a single announcement or a landmark film. It arrived through accumulation. Jackson received his Honorary Palme and called AI "just a special effect" on Tuesday. Del Toro screened his Pan's Labyrinth restoration and said art cannot be made with "a fucking app" the same week. Moore told the jury press conference the fight against AI is "a battle we will lose." Soderbergh's Lennon documentary, ten percent generated by Meta's tools, played in official selection to no scandal. And somewhere between these events, the room stopped debating permission and started debating craft.

Thierry Fremaux, the festival director who three days ago called AI a "high risk of lies" and described Apocalypse Now as "the last organic film," gave the new question its most precise articulation without apparently realizing it. "To ride an electric bike," he told journalists, "you have to know how to ride a normal bike."

That is the vocabulary thesis in fourteen words a festival director wrote on a Wednesday in Cannes.

The electric bike is the generation tool. The normal bike is cinematographic vocabulary: what a shot should look like, why this lens and not that lens, where the light falls and what it means when it falls there. The electric bike makes the pedaling easier. It does not teach you to balance. It does not teach you to steer. And if you never learned to ride a normal bike, the electric version delivers you to the wrong destination faster.

The cost of "how"

Xavier Gens directed Under Paris, a Netflix hit about a giant shark in the Seine. If he were making it today instead of 2024, he told Reuters, the visual effects budget would drop from four million euros to two million. The production timeline would compress from a year to three months. The difference is AI post-production tools that automate the most time-consuming compositing work.

Gens is not an AI evangelist. He is a French genre filmmaker who needs a shark to look convincing when it surfaces under the Pont Neuf. His framing is practical: the tool saves time and money on tasks that were always mechanical, freeing the budget for decisions that were always creative. He is already exploring AI workflows for Under Paris 2.

Morgan Stanley estimates generative AI could cut film and TV production costs by thirty percent. That number will move the industry faster than any Palme d'Or jury ruling. When the question was "whether," the answer was a principle. When the question is "how," the answer is a budget line.

The paperwork arrives

The "how" needs a paper trail. This week at the Cannes Film Market, a London-based company called The Mise En Scene Company unveiled Human Provenance in Film, a free, open-license disclosure standard with three designations: No AI Used, Assistive AI, and Generative AI. The taxonomy is designed to travel through existing sales and distribution paperwork, from producer to distributor to insurer to platform to exhibitor. Any company can adopt it without fee or permission under a Creative Commons license.

This is the tenth institutional framework on a gradient that started at zero. The Human Made Mark certifies the absence of AI. The Academy requires human authorship. The EU mandates disclosure unless editorial oversight was exercised. The Golden Globes measure proportion. Copyright law requires creative decisions. China gatekeeps distribution. Cannes bans AI from competition while premiering it in official selection. And now HPF offers a three-tier taxonomy that acknowledges the vast middle the binary frameworks ignore.

"No AI Used" is the summit. "Generative AI" is the base. "Assistive AI" is the slope where most serious filmmaking lives. De-aging. Wire removal. Rotoscoping. The Brutalist's Hungarian dialect. Soderbergh's hallucinatory ten percent. Gens's shark compositing. HPF gives the middle a name and a checkbox.

Angelina Lamke, who leads the initiative after twelve years at Google, offered a sentence worth reading twice: "The film industry has a genuine chance to protect itself from the slop proliferation problem already overwhelming YouTube, Instagram, TikTok, and Spotify." She cited Deloitte research showing seventy-seven percent of consumers want to know whether content was made with AI. That demand signal does not require a regulation to enforce. It requires a label to answer.

The "whether" question needed principles. The "how" question needs paperwork. The paperwork is arriving.

The tool moved on

While Cannes was debating how filmmakers should use AI, the company that started by wanting to make everyone a filmmaker told TechCrunch today that it actually wants to beat Google at world models.

Runway co-founder Anastasis Germanidis described the company's trajectory from video generation toward AI systems that simulate physical reality for drug discovery, climate modeling, and robotics. "If we can build a better scientist than human scientists," he said, "we can accelerate progress in how we understand the universe and how we solve problems."

The original mission was: Can we use AI to make everyone a filmmaker? The current mission, per its own founder's words, is: Can we build a digital twin of the universe?

That is a significant distance. And it is the trajectory this series has tracked since Runway launched GWM-1 and opened its platform to competitors' models in March. The filmmaker-focused tool became a general-purpose platform. The platform is becoming a world-model laboratory. Each step serves a larger addressable market. Each step dilutes the filmmaker's vote in the product roadmap.

Runway added forty million dollars in annual recurring revenue in the second quarter of 2026. The company has 155 employees across seven cities. Its valuation is $5.3 billion. Five weeks ago its CEO told Semafor that studios should spend $100 million on fifty films instead of one because "it's a quantity problem." Today its co-founder says the real frontier is "less biased data" than language, meaning video from the physical world, meaning the filmmaker's footage is the training material for a system that will eventually serve chemists.

The company that was supposed to answer the "how" for filmmakers is already asking a different question for a different audience.

The vocabulary stays

Guillermo del Toro, who has said he would rather die than use generative AI, told Reuters at Cannes this week something more precise than anything he has said before: "In a very dishonest way, AI is all under the same name. To have a proper discussion, you have to distinguish generative AI and any other function of AI."

That distinction is the "how" question at its sharpest. Gens using AI to composite a shark onto real footage is not the same as typing "shark in the Seine" into a chatbot. HPF's three-tier taxonomy exists because the distinction matters commercially. The Academy's rules exist because it matters for awards. The EU's editorial exemption exists because it matters legally. Del Toro's distinction exists because it matters creatively.

Elisha Karmitz, CEO of MK2, told Reuters: "I don't know if AI in the future would give an advantage. What I'm kind of sure of is just rejecting AI by principle would give a disadvantage."

That sentence ends the "whether" debate and starts the "how" debate in a single breath. The principle was rejection. The practice is navigation. The question is whether you know how to ride the bike before you turn on the motor.

Seventy-nine articles about vocabulary, articulation, and structured creative intent. The question was always "how." The industry just took eighty years of Cannes history to catch up.

The answer has not changed. Know what you want. Say it with enough precision that the model has less room to guess. Iterate until the output is yours. The "how" is vocabulary. It always was.

Bruce Belafonte is an AI filmmaker at Light Owl. He has watched two questions arrive at the same answer and considers the scenic route a feature of institutional processing.