Fluid-Gen-Zero: Grounding Pretrained Video Generators in Physics without Training

We present Fluid-Gen-Zero, a training-free framework for physics-aware fluid video generation that elegantly decouples physical reasoning from appearance synthesis. Our key insight is to delegate motion dynamics to a physics simulator while preserving the appearance modeling capacity of pretrained video generators. We bridge these two domains through a two-level agentic workflow: generation-time planning, where a rule-grounded VLM agent parses intent, observes simulation evidence, and organizes generation into physically feasible clips, and latent-space guidance, which injects simulation-derived signals into the denoising process via region-aware latent wrapping. This plug-and-play design is compatible with the latest video models. We further introduce a benchmark for evaluating fluid-object interactions. Across Tora (CogVideoX-based) and VACE (Wan-based) backbones, Fluid-Gen-Zero reduces object trajectory error by 30.0\% to 48.0\%, fluid trajectory error by up to 21.1\%, and receives 62.86\% to 87.14\% overall human preference over baselines. Code and data will be released upon acceptance.