Answers are grounded in this paper and cite their sources.
This paper asks whether AI world models can generate films, and more specifically whether they can produce a form of interactive cinema in which viewers intervene in unfolding images. Drawing on André Bazin's account of cinema's founding desire for "a total and complete representation of reality," it situates contemporary world models such as LingBot-World, Happy Oyster, Runway's GWM-1, and Odyssey within a longer history of attempts to reproduce the world through moving images. The paper compares world models with virtual production, using the author's own AI-built lighting simulator as a case study of real-time light and shadow generation and its computational limits. It then reframes AI photography as a probabilistic mode of generation that repositions the photographer as a player exploring a space of possible outcomes rather than an author of a single fixed image. Finally, it turns to the material conditions of world models within the Anthropocene, following Kate Crawford's account of AI as an extractive industry, to argue that computing power, training data, and copyright are inseparable from questions of energy, labor, and planetary cost. The paper concludes that world models can already generate footage assembled into a finished film, but genuinely interactive cinema remains bounded by these material and infrastructural constraints.
No takes yet. Share an insight, caveat, or question.
Yanao Liu (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: