Growing Images: Spatial Scheduling in Diffusion Inpainting

Matteo Vilucchio, Luca Pesce, Florent Krzakala

Keywords: Image inpainting, Spatial scheduling, Stochastic interpolants, Statistical physics of generative models, Blume–Capel

Abstract:

Diffusion inpainting implicitly treats the order in which masked pixels are denoised as irrelevant. We argue in this manuscript that it is not.
Holding the model, prompt, and per-patch compute budget fixed in Stable Diffusion inpainting, we show that progressively filling a mask from its boundary measurably shifts reconstruction, perceptual, and distributional metrics relative to standard parallel denoising.
We study this analytically by recasting inpainting as \emph{finite-depth approximate inference}. Specifically, we consider a block-structured Blume--Capel spin system: a diffusion-like stochastic interpolant tilts the prior into a time-dependent Gibbs posterior with block-dependent random fields.
In this picture,
schedule-induced bias turns into a predictable consequence of finite-depth inference.

Submission Number: 39