Every Shot, Zero-Shot: The Decoupling of Framing and Duration in Sound-Era Hollywood and Prestige Film

We study the visual pacing of 1,236 feature films from three overlapping film corpora: the Academy Award Best Picture nominees (1927–2009), the corresponding ten highest-grossing films in the United States in the same time period (1927–2009), and the British Film Institute's Greatest Films poll (2022). With zero-shot models, we tag the scale and the number of people on screen for every shot. In all three corpora, the relationship between shot pace and what appears on screen erodes progressively across the sound era. In the Hollywood films of the 1930s and 1940s, shot scale pulls the average film's shot about 28.1% away from the film's own central pace (bootstrap SE = 1.97) and the number of people on screen about 28.8% (1.33). By the 2000s these shared pulls shrink to 9.4% (0.59) and 5.8% (0.79). The pull inside individual films, each measured along its own axis, falls just as steeply, from 35.9% to 13.5% on scale and from 33.5% to 12% on people. The erosion cannot be explained simply by faster cutting. The prestige films decouple while slowing down, and the slowest third of the corpus shows the same decline. The coupling is diminished rather than abolished, with framing still pulling duration in modern films at roughly a third of its classical strength. These findings point to a steady breakdown of the classical relationships between certain formal elements of film and narrative goals. A cut to a close-up was once also a change of tempo, and is now increasingly only a change of view. Alongside these results, we present a granular, human-evaluated, reusable annotation pipeline and publish the supporting film- and shot-level metadata under an open-access license.