Stable and Near-Reversible Diffusion ODE Solvers for Image Editing

2026-05-12

AI Summary

Diffusion models for large-scale text-guided image editing often face a trade-off between invertibility and numerical stability, which can lead to editing artifacts and degradation of background structure. This work proposes a near-invertible Runge–Kutta ODE solver that integrates vector field smoothing with a diffusion inversion strategy, enhancing numerical stability without imposing strict invertibility constraints. The study further uncovers an inherent trade-off between background preservation and prompt alignment—a relationship previously unexplored in the literature. By leveraging the proposed method, the approach achieves significantly improved editing robustness and output fidelity while maintaining strong consistency with the original background content.

Abstract

The inversion of diffusion models plays a central role in image editing. Algebraically reversible ODE solvers provide an appealing approach to diffusion inversion for text-guided image editing, by eliminating the inversion error inherent in DDIM-based editing pipelines. However, empirical results indicate that reversibility alone is insufficient. As edits require larger semantic or visual changes, reversible diffusion solvers often exhibit instabilities and suffer sharp drops in output quality. In this paper, we show that the trade-off between exact reversibility and numerical stability manifests empirically as a trade-off between background preservation and prompt alignment in image editing. We then investigate the use of near-reversible Runge-Kutta methods as a more stable alternative to exactly reversible diffusion schemes. When combined with a vector-field smoothing strategy, the resulting approach improves edit fidelity, remains stable under large edits, and largely retains the background-preservation benefits of reversible solvers.

Problem

• diffusion models
• image editing
• ODE solvers
• reversibility
• numerical stability

Innovation

• near-reversible ODE solvers
• diffusion inversion
• Runge-Kutta methods
• vector-field smoothing
• image editing