Programmable World Model
The paper separates game-like world rules from video generation so entities can keep state even when they leave the frame.
Its framework turns natural-language instructions into executable programs for entity states and transitions. A lightweight engine maintains a persistent global world state, including off-screen objects and non-visual attributes. That state is compiled through augmented 3D bounding boxes into conditioning signals for a pretrained video model. The authors report 94% Count Accuracy and 98% State Accuracy on their CombatStateBench benchmark. HF Daily Papers' note
Its framework turns natural-language instructions into executable programs for entity states and transitions. A lightweight engine maintains a persistent global world state, including off-screen objects and non-visual attributes. That state is compiled through augmented 3D bounding boxes into conditioning signals for a pretrained video model. The authors report 94% Count Accuracy and 98% State Accuracy on their CombatStateBench benchmark. HF Daily Papers' note
score 5