AI 3D simulation filmmaking

Today's AI video systems generate each shot independently, so characters, objects, scale and lighting drift between shots. This research program proposes a different architecture for AI filmmaking: a persistent, structured world in which events actually happen under explicit rules, and in which every shot is a camera observing that world rather than a fresh generation.The paper sets out the core concepts:Master reality and production state: a master world kept separate from per-scene production state.Events as first-class objects: each with an intent, preconditions, a resolution computed by rules, and the changes it makes.Causality and counterfactuals.Scale and ratio control: sizes stored once, in real-world units.Observation contracts for cameras.A proposed file format (.pworld) and a player that runs it.Every claim is labelled as established, existing research, inference, proposed or open. It also includes an honest comparison with game engines, and a Minimum Viable Production World test of one room, two characters, ten objects and ten events. A working browser prototype accompanies the paper.The concept and requirements are the author's. The document was researched, drafted and critiqued with the help of AI assistants.