Stable and Near-Reversible Diffusion ODE Solvers for Image Editing
This work improves image editing quality for users of diffusion models by providing a more stable inversion method that handles larger semantic changes.
The paper addresses instability in reversible diffusion ODE solvers for text-guided image editing, showing a trade-off between background preservation and prompt alignment. By using near-reversible Runge-Kutta methods with vector-field smoothing, they improve edit fidelity and stability under large edits while retaining background preservation.
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.