At Runway’s showcase at Cinesite in London this August, I caught a glimpse of where generative AI filmmaking might be headed.
One film stood out. Costa Verde, by Léo Cannone, is shot on glitchy Hi-8, documenting a childhood holiday in rural France. Cut into this mosaic of the mundane were surreal images; a giant frog in a paddling pool, a wandering giant lit with fire, a witch flying her broomstick through a forest. A chronicle of small wonders, rendered through the slippery truth of childhood memory.
The film works because of its construction. It does not build scenes; it accumulates fragments, the way memory does. And it was all made with generative AI.
The industry mostly assumes generative AI films like these are warm-ups; that synthetic films, as I tend to call them, are traditional movies on the cheap. But I’d argue this is a medium still finding its form, and it’s most apparent in short films, where experimentation is low-to-no cost. Those sparks of originality on display at the Runway event owe little to classic film structure. Instead those works are starting to find out what the medium is good at.
From watching various films online, and from making synthetic film work at Wise Media, my production company, it’s becoming apparent what gen AI is good, and not so good at. The standard grammar of Hollywood is designed for continuity, of space and of performance, built so an audience can inhabit a stable world to allow focus on the people inside it changing. It relies on a contínuous presence; sustained performance, realistic speech, and the emotional ebb and flow within a scene.
But in synthetic films the smallest is a discreet clip of a few seconds, one that feels distinct from another created with exactly the same starting materials. A timeline is assembled from a mosaic of clips, hence its fragmentary feel.
What works well in synthetic films is different from a traditional feature; mood, memory, dreams, the past, the imagined future, macro details, vast non-human landscapes. There are gaps, a tension between geography and subject, and where a face is more often a question than a performance. Subjective, discontinuous, interior.
New mediums began by borrowing. Photography’s first bid for respectability was made in painting’s vocabulary; one of the pioneers of early photography, Julia Margaret Cameron, was notable for dressing her friends and her household staff in costumes for tableaux composed like canvases. Early cinema borrowed the grammar of the stage that was in vogue at the time. It took both mediums years to work out what the new art form could give an audience the old one could not.
Traditional film does continuity, stability and the accumulation of performance supremely well, and synthetic films that imitate that approach often feel labored, fragile, even inauthentic. I believe the ideal form is still evolving. For now the films that work interrogate the world through memory, through the past and the future, through the imagined realm and the logic of dreams. A medium of the mind rather than of reality.
Like a giant frog in a child’s paddling pool.
Mateo Willis is a director with a career spanning film and television, and the founder of Wise Media, an AI-first production company that uses both traditional and generative tools.
