Generative Visual Chain-of-Thought for Image Editing

TL;DR

This paper introduces GVCoT, a novel framework for image editing that combines visual reasoning with spatial cues, trained on a large dataset, and demonstrates superior performance on new benchmarks.

Contribution

The paper proposes GVCoT, a unified visual reasoning framework for image editing that jointly optimizes reasoning and editing in an end-to-end manner, and introduces new datasets and benchmarks.

Findings

GVCoT outperforms state-of-the-art models on SREdit-Bench and ImgEdit.

Constructed GVCoT-Edit-Instruct dataset with 1.8 million samples.

Progressive training strategy enhances localization and editing quality.

Abstract

Existing image editing methods struggle to perceive where to edit, especially under complex scenes and nuanced spatial instructions. To address this issue, we propose Generative Visual Chain-of-Thought (GVCoT), a unified framework that performs native visual reasoning by first generating spatial cues to localize the target region and then executing the edit. Unlike prior text-only CoT or tool-dependent visual CoT paradigms, GVCoT jointly optimizes visual tokens generated during the reasoning and editing phases in an end-to-end manner. This way fosters the emergence of innate spatial reasoning ability and enables more effective utilization of visual-domain cues. The main challenge of training GCVoT lies in the scarcity of large-scale editing data with precise edit region annotations; to this end, we construct GVCoT-Edit-Instruct, a dataset of 1.8M high-quality samples spanning 19 tasks. We adopt a progressive training strategy: supervised fine-tuning to build foundational localization ability in reasoning trace before final editing, followed by reinforcement learning to further improve reasoning and editing quality. Finally, we introduce SREdit-Bench, a new benchmark designed to comprehensively stress-test models under sophisticated scenes and fine-grained referring expressions. Experiments demonstrate that GVCoT consistently outperforms state-of-the-art models on SREdit-Bench and ImgEdit. We hope our GVCoT will inspire future research toward interpretable and precise image editing.