InstantRetouch: Efficient and High-Fidelity Instruction-Guided Image Retouching with Bilateral Space

2026-06-03

AI Summary

This work addresses the limitations of existing language-guided image editing methods, which often suffer from insufficient content fidelity due to the generative nature of diffusion models and inefficient iterative sampling. To overcome these issues, we introduce the bilateral grid mechanism into this task for the first time, enabling efficient full-resolution image editing by predicting a low-resolution bilateral affine transformation grid guided by a learned guidance map. Our approach integrates variational score distillation with prompt alignment loss, effectively suppressing content drift while preserving pretrained generative priors. Evaluated on a newly constructed benchmark, the proposed method significantly outperforms recent approaches such as Gemini-2.5-Flash in inference speed, while achieving superior performance in content fidelity, instruction following, and visual quality.

Abstract

Language-guided photo retouching aims to adjust color and tone while preserving geometry and texture. Recently, diffusion-based retouching shows a superior visual quality, but often struggles with both fidelity issues due to its generative nature and efficiency because of its iterative sampling process. In this work, we propose an efficient and fidelity-preserving retouching method using bilateral space manipulation, which is both compact and content-decoupled. Specifically, instead of directly editing pixels or image latents, our model predicts a low-resolution bilateral grid of affine transforms, which are sliced using a learned guidance map and then applied to the full-resolution image. This approach yields both high fidelity and improved efficiency. To retain strong priors of a pretrained generative model, we distill a multi-step diffusion model into our bilateral grid framework using Variational Score Distillation, complemented by a prompt alignment loss to guide instruction-following behavior. Additionally, we introduce a new benchmark and evaluate our method across multiple dimensions: fidelity, instruction following, and efficiency. Compared to the latest retouch methods, like Gemini-2.5-Flash (Nano-Banana), our method can avoid content drift, significantly improve latency, and generate visually pleasing edits, while maintaining a high level of fidelity. Project page: https://openimaginglab.github.io/InstantRetouch/.

Problem

• image retouching
• instruction-guided editing
• fidelity
• efficiency
• diffusion models

Innovation

• bilateral space
• affine grid
• Variational Score Distillation
• instruction-guided retouching
• efficiency-fidelity trade-off