Beyond Blur: A Fluid Perspective on Generative Diffusion Models

TL;DR

This paper introduces a PDE-based corruption process for image synthesis using advection-diffusion, enhancing diversity and quality in generative diffusion models by integrating fluid dynamics principles.

Contribution

It presents a novel PDE-driven corruption process with a GPU-accelerated solver, generalizing previous PDE-based methods for improved image generation.

Findings

Advection improves image diversity and quality.

The framework generalizes prior PDE-based corruption techniques.

Physically motivated PDE enhances generative modeling.

Abstract

We propose a novel PDE-driven corruption process for generative image synthesis based on advection-diffusion processes which generalizes existing PDE-based approaches. Our forward pass formulates image corruption via a physically motivated PDE that couples directional advection with isotropic diffusion and Gaussian noise, controlled by dimensionless numbers (Peclet, Fourier). We implement this PDE numerically through a GPU-accelerated custom Lattice Boltzmann solver for fast evaluation. To induce realistic turbulence, we generate stochastic velocity fields that introduce coherent motion and capture multi-scale mixing. In the generative process, a neural network learns to reverse the advection-diffusion operator thus constituting a novel generative model. We discuss how previous methods emerge as specific cases of our operator, demonstrating that our framework generalizes prior PDE-based corruption techniques. We illustrate how advection improves the diversity and quality of the generated images while keeping the overall color palette unaffected. This work bridges fluid dynamics, dimensionless PDE theory, and deep generative modeling, offering a fresh perspective on physically informed image corruption processes for diffusion-based synthesis.