LanPaint: Training-Free Diffusion Inpainting with Asymptotically Exact and Fast Conditional Sampling

TL;DR

LanPaint introduces a training-free, asymptotically exact method for image inpainting using diffusion models, enabling fast, accurate, and gradient-free partial conditional sampling with superior results.

Contribution

It presents LanPaint, a novel training-free approach for partial conditional sampling in diffusion models, overcoming previous limitations of intractability and inefficiency.

Findings

Achieves accurate partial conditioning in inpainting tasks.

Provides faster, backpropagation-free Monte Carlo sampling.

Demonstrates superior visual coherence across diverse inpainting scenarios.

Abstract

Diffusion models excel at joint pixel sampling for image generation but lack efficient training-free methods for partial conditional sampling (e.g., inpainting with known pixels). Prior work typically formulates this as an intractable inverse problem, relying on coarse variational approximations, heuristic losses requiring expensive backpropagation, or slow stochastic sampling. These limitations preclude: (1) accurate distributional matching in inpainting results, (2) efficient inference modes without gradient, (3) compatibility with fast ODE-based samplers. To address these limitations, we propose LanPaint: a training-free, asymptotically exact partial conditional sampling methods for ODE-based and rectified flow diffusion models. By leveraging carefully designed Langevin dynamics, LanPaint enables fast, backpropagation-free Monte Carlo sampling. Experiments demonstrate that our approach achieves superior performance with precise partial conditioning and visually coherent inpainting across diverse tasks.