Defining AI Disclosure in Filmmaking
AI disclosure in filmmaking means clearly informing audiences, collaborators, and commissioners when generative or assistant AI tools contribute to a film’s images, sound, or editing decisions, especially when those contributions are hard to detect with the naked eye and might otherwise be mistaken for traditional craftsmanship, archival material, or live performance captured on set. That definition goes to the heart of the debate that followed Steven Soderbergh’s documentary John Lennon: The Last Interview at Cannes. Around ten percent of its images were built with Meta’s video tools, yet the AI sequences were designed as surreal, non-literal visuals: circles of light, morphing flowers, and abstract diptychs rather than deepfaked Lennons. Soderbergh described himself as “my own whistle blower,” arguing that transparency about AI-assisted production work is as important as the ethical choices about when to use the tools in the first place.
A Festival Divided on AI Visual Effects Disclosure
Cannes became a testing ground for cinema transparency standards around AI. Soderbergh’s Meta-backed film sat between sharply opposed public statements. Guillermo del Toro dismissed generative tools with an expletive, while Peter Jackson told the crowd he would be open to AI performance work, even digital resurrections, if legacy guardians agreed. Inside the film market, Darren Aronofsky and Google’s James Manyika framed AI as a tool in the lineage of sound and portable cameras, not a replacement for human talent. At the same time, festival leaders Thierry Frémaux and Iris Knobloch said they aimed to exclude films “primarily driven by generative AI” from competition and floated a “made without artificial intelligence” label. Yet these ideas are not formal film industry AI guidelines, leaving filmmakers to guess how much AI visual effects disclosure is expected and how it might shape selection and awards.
Audience Trust vs Creative Freedom
Soderbergh’s ethical test for AI-assisted images is blunt: “It has to be necessary. Is it the only way to accomplish what I want to see?” He describes using Meta’s tools to solve “images that are impossible to shoot” after the project had run out of time and money, positioning AI as a last resort rather than a shortcut. Still, his choice to spell out that about ten minutes of the documentary rely on generative visuals raises a key tension. Some viewers worry that invisible AI undercuts authenticity, especially in documentary, where trust rests on how images relate to reality. Others see AI as another stylistic option, akin to animation or collage, that needs no special warning. For now, AI disclosure filmmaking is voluntary, and each director must decide how much process detail audiences are entitled to know without spoiling the film’s spell.
Invisible Artists and New Questions of Attribution
While Cannes argued about AI, the Commission Supérieure Technique de l’image et du son used its CST Awards to honour human crafts that usually stay unseen. Editor Nicolas Rumpl was praised for work that reveals a film’s visual identity and emotional depth, while production designer Esther Mysius was celebrated for turning physical spaces into storytelling. Their recognition underlines how many hands shape a film without public credit beyond a brief roll of names. As generative tools enter pipelines, this raises awkward questions. If AI helps design a shot or transition, does it displace a craftsperson, or does someone still guide and refine the result in ways that deserve attribution? The CST’s focus on invisible artists suggests a broader shift: cinema transparency standards may soon need to cover not only when AI is used, but how and by whom its outputs are supervised, edited, and owned.
Towards Shared Cinema Transparency Standards
The debate around Soderbergh’s Lennon film shows how far the industry is from shared rules on AI visual effects disclosure. Cannes can hint at policies and labels, but there is no widely accepted framework for how productions should log AI involvement, how festivals should request that information, or how credits might reflect hybrid workflows. One practical direction would be a simple, standardised disclosure note in end credits and festival materials, specifying whether generative AI contributed to images, sound, or editing decisions, without turning every film into a technical report. Another is to fold AI into existing guild and craft discussions, so editors, designers, and VFX supervisors can define ethical lines for their disciplines. Until such film industry AI guidelines exist, individual transparency experiments — like Soderbergh calling out his own ten percent — will continue to set the informal benchmarks others are measured against.