Let it Snow! Animating 3D Gaussian Scenes with Dynamic Weather Effects via Physics-Guided Score Distillation

TL;DR We introduce Physics-Guided Score Distillation to animate static 3D Gaussian scenes with scene-wide dynamic weather effects.

Above, we demonstrate our framework generating effects such as snowfall, rainfall, fog, and sandstorms. By leveraging physics simulation as a motion prior to guide Video Score Distillation Sampling (Video-SDS), our approach jointly refines both motion and appearance, achieving physically plausible dynamics with photorealistic quality.

Abstract

3D Gaussian Splatting has recently enabled fast and photorealistic reconstruction of static 3D scenes. However, dynamic editing of such scenes remains a significant challenge. We introduce a novel framework, Physics-Guided Score Distillation, to address a fundamental conflict: physics simulation provides a strong motion prior that is insufficient for photorealism , while video-based Score Distillation Sampling (SDS) alone cannot generate coherent motion for complex, multi-particle scenarios. We resolve this through a unified optimization framework where physics simulation guides Score Distillation to jointly refine the motion prior for photorealism while simultaneously optimizing appearance. Specifically, we learn a neural dynamics model that predicts particle motion and appearance, optimized end-to-end via a combined loss integrating Video-SDS for photorealism with our physics-guidance prior. This allows for photorealistic refinements while ensuring the dynamics remain plausible. Our framework enables scene-wide dynamic weather effects, including snowfall, rainfall, fog, and sandstorms, with physically plausible motion. Experiments demonstrate our physics-guided approach significantly outperforms baselines, with ablations confirming this joint refinement is essential for generating coherent, high-fidelity dynamics.

Examples of Dynamic Weather Effects

Select a scene and effect to view the animation enabled by our Physics-Guided Score Distillation framework.
Our approach generates scene-wide weather effects with continuous particle emission that naturally interact with static geometry and evolve coherently over time.

How does it work?

🌍 Given multi-view images, we first reconstruct a static scene using 3D Gaussian Splatting (3DGS) and map it to a particle representation.

💡 We introduce dynamic particles and simulate their motion using the Material Point Method (MPM) to generate physically plausible reference trajectories with mesh-based collision handling.

🔍 These trajectories serve as a motion prior to guide our Neural Dynamics Model, which is optimized via Video-SDS to jointly refine both motion and appearance for photorealism.

📋 See our paper for details on how Physics-Guided Score Distillation leverages physics simulation to enable coherent, photorealistic scene-wide weather effects.

BibTeX


                @misc{fiebelman2025letsnowanimating3d, 

                     title={Let it Snow! Animating 3D Gaussian Scenes with Dynamic Weather Effects via Physics-Guided Score Distillation}, 

                     author={Gal Fiebelman and Hadar Averbuch-Elor and Sagie Benaim}, 

                     year={2025}, 

                     eprint={2504.05296}, 

                     archivePrefix={arXiv}, 

                     primaryClass={cs.GR}, 

                     url={https://arxiv.org/abs/2504.05296}, 

                }