Tag
This paper presents an agentic framework that uses coding agents to generate physically plausible world simulations from natural language prompts, outperforming video-based models in physical accuracy and instruction fidelity.