PhyRPR: Training-Free Physics-Constrained Video Generation

TL;DR

PhyRPR introduces a training-free, three-stage pipeline that decouples physical reasoning from visual synthesis, enabling more physically plausible and controllable video generation.

Contribution

It proposes a novel training-free, staged approach that separates physical understanding from visual synthesis in video generation.

Findings

Improves physical plausibility of generated videos.

Enhances motion controllability in video synthesis.

Operates without additional training or fine-tuning.

Abstract

Recent diffusion-based video generation models can synthesize visually plausible videos, yet they often struggle to satisfy physical constraints. A key reason is that most existing approaches remain single-stage: they entangle high-level physical understanding with low-level visual synthesis, making it hard to generate content that require explicit physical reasoning. To address this limitation, we propose a training-free three-stage pipeline,\textit{PhyRPR}:\textit{Phy\uline{R}eason}--\textit{Phy\uline{P}lan}--\textit{Phy\uline{R}efine}, which decouples physical understanding from visual synthesis. Specifically, \textit{PhyReason} uses a large multimodal model for physical state reasoning and an image generator for keyframe synthesis; \textit{PhyPlan} deterministically synthesizes a controllable coarse motion scaffold; and \textit{PhyRefine} injects this scaffold into diffusion sampling via a latent fusion strategy to refine appearance while preserving the planned dynamics. This staged design enables explicit physical control during generation. Extensive experiments under physics constraints show that our method consistently improves physical plausibility and motion controllability.