Bridging the Manifold Gap: Riemannian Residual Line Search for One-Step Image Editing
2026-06-23 • Computer Vision and Pattern Recognition
Computer Vision and Pattern Recognition
AI summaryⓘ
The authors focus on improving one-step diffusion image editors, which are fast but struggle to balance strong edits and preserving the original image. Instead of creating a new editing model, they introduce a method called Riemannian Residual Line Search that refines edits by better estimating how the prompt changes locally and then selecting the best result based on how well it matches the target prompt. Their approach helps choose the best edit from a set of candidates to get better results. They tested it on a large benchmark and found it works better than other current one-step methods.
diffusion modelsimage editingenergy-field transportCLIP alignmentone-step updateRiemannian geometryprompt-delta fieldimage synthesisPIE-Bench++line search algorithm
Authors
Hongzhu Yi, Zhongtian Luo, Tong Li, Yiyan Fan, Jungang Xu
Abstract
One-step diffusion editors are fast because they avoid inversion and iterative optimization, but a single transport update must be aggressive enough to realize the target prompt and conservative enough to preserve the source image--and no fixed update strength satisfies both demands across edit types. We treat this tension as a post-hoc candidate-selection problem on top of energy-field transport rather than as a new editing model. Our proposed method, Riemannian Residual Line Search, first builds a stronger edit by estimating the local time curvature of the prompt-delta field and projecting the corrected direction back onto the update norm of the original first-order energy-field transport estimation. It then forms a small residual path from the source image to this strong edit, retains the original first-order output as one candidate, and picks the final image by maximizing target-prompt CLIP alignment. On a 700-sample PIE-Bench++ evaluation across 10 edit type IDs, our method achieves state-of-the-art (SOTA) performance among current one-step update algorithms.