Before a single light is rigged or a camera lens chosen, a film often exists as a wall of small, hurried drawings: a figure falling in a stairwell, a shadow across a doorway, an arrow showing which way the dolly will glide. Saul Bass's meticulous panels for the shower scene in Hitchcock's Psycho are the most celebrated example, but the tradition runs from Max Douy's lavish boards for Jodorowsky's Dune, a film that was never made, to the quick pencil sketches pinned up on sets from Seoul to Los Angeles. It is this quiet craft, the first step in carrying a script off the page, that generative AI now appears to have in its sights. The signs are hard to ignore, and they come from unexpected quarters.

James Cameron sits on the board of a generative-AI company, Steven Soderbergh used Meta-powered tools for the animations in his documentary about John Lennon, and Martin Scorsese, hardly a technophile, announced in June a partnership with a firm whose text-to-image model he used to storyboard his next picture. Even Christopher Nolan, who calls the notion that AI will replace humans "nonsense", concedes it will yield some useful imaging tools. What might be lost is easy to overlook, because it was never on the screen. A storyboard session is typically a crowded conversation among director, artist, cinematographer, editor, production designer and stunt co-ordinator, in which ideas are acted out, sound effects improvised and dialogue read aloud.

Alex Clark, the British artist behind boards for Mickey 17, The Gorge and The Last House, describes it as a creative relationship built between people. A text prompt can produce an image in seconds, but it is far less clear that it can reproduce the argument, the laughter and the happy accidents that surround the drawing.