Edge-preserving noise for diffusion models

TL;DR

This paper introduces an edge-preserving noise scheme for diffusion models that enhances structural detail capture and improves performance in structure-guided tasks like stroke-to-image synthesis.

Contribution

The authors propose a hybrid noise scheme with an edge-aware scheduler, enabling diffusion models to better preserve structural details and be fine-tuned efficiently for improved task performance.

Findings

Enhanced structural detail in generated images.

Improved robustness and perceptual quality in structure-guided tasks.

Consistent improvements in FID, KID, and CLIP-score metrics.

Abstract

Classical diffusion models typically rely on isotropic Gaussian noise, treating all regions uniformly and overlooking structural information important for high-quality generation. We introduce an edge-preserving diffusion process that generalizes isotropic models via a hybrid noise scheme with an edge-aware scheduler that smoothly transitions from edge-preserving to isotropic noise. This enables the model to capture fine structural details while generally maintaining global performance. We evaluate the impact of structure-aware noise in both diffusion and flow-matching frameworks, and show that existing isotropic models can be efficiently fine-tuned with edge-preserving noise, making our framework practical for adapting pre-trained systems. Beyond unconditional generation, our method particularly shows improvements in structure-guided tasks such as stroke-to-image synthesis, improving robustness and perceptual quality, as evidenced by consistent improvements across FID, KID, and CLIP-score.