Previsualization serves as a crucial intermediate layer between initial ideas and final production in fields such as film, games, architecture, and urban design. It allows creators to iteratively refine scenes, actions, cameras, and spatial-temporal dynamics. However, existing generative methods often rely on simple prompts to jointly control these factors through one-shot image or video synthesis, resulting in weak controllability and limited support for iterative editing.
StateFlow addresses this limitation by introducing a persistent 3D world state. The framework recognizes that a world comprises multiple elements with geometry, appearance, and other attributes, along with cameras. Different frames are produced through local modifications or recombinations of this shared state, which is otherwise largely reused. Thus, the missing component is an explicit and persistent working state.
StateFlow is a state-centric framework that uses an editable 3D world to organize scene structure, evolution, and cameras. It maintains a persistent structured 3D state of scene elements and camera configurations, serving as the core working representation for previsualization. Off-the-shelf video models can enhance visual quality when higher fidelity is desired.
The framework operates in three stages: state construction, state evolution, and state access. State construction lifts generated 2D content into a coherent 3D world through prior-guided, conflict-aware dual-view initialization. State evolution translates user intent into structured state transitions while preserving world memory, avoiding full-scene regeneration for each edit. State access uses render-feedback reflection to refine camera plans into visually feasible trajectories, avoiding reliance on VLM semantics alone.
Experiments demonstrate that StateFlow produces high-quality 3D worlds for video creation and game-like prototyping, showcasing its potential to enhance creative workflows.