Diffusion models can be challenged in the low signal-to-noise regime, where they have to make pixel-level predictions despite the presence of high noise. The geometric intuition is akin to using the finest stroke for oil painting throughout, which may be ineffective. We therefore study stroke-size control as a controlled intervention that changes the effective roughness of the supervised target, predictions and perturbations across timesteps, in an attempt to ease the low signal-to-noise challenge.
@article{arxiv.2603.26783,
title = {Can We Change the Stroke Size for Easier Diffusion?},
author = {Yunwei Bai and Ying Kiat Tan and Yao Shu and Tsuhan Chen},
journal= {arXiv preprint arXiv:2603.26783},
year = {2026}
}