NEP: Autoregressive Image Editing via Next Editing Token Prediction

Xiaojian (Shawn) Ma (BIGAI) · Qing Li (Beijing Institute of Technology) · Huimin Wu (Beijing Institute for General Artificial Intelligence) · Haozhe Zhao (Peking University) · Yanpeng Zhao (University of Edinburgh)

any-order autoregressive modelany-region editingautoregressive image generation computational costsediting-token predictionimage editing frameworksiterative refinementlanguage instructionnon-editing regionsquality of editsselective regenerationstate-of-the-art benchmarks test-time scaling text-guided image editingzero-shot image editing

Text-guided image editing involves modifying a source image based on a language instruction and, typically, requires changes to only small local regions. However, existing approaches generate the entire target image rather than selectively regenerate only the intended editing areas. This results in (1) unnecessary computational costs and (2) a bias toward reconstructing non-editing regions, which compromises the quality of the intended edits. To resolve these limitations, we propose to formulate image editing as $\textbf{N}$ext $\textbf{E}$diting-token $\textbf{P}$rediction (NEP) based on autoregressive image generation, where only regions that need to be edited are regenerated, thus avoiding unintended modification to the non-editing areas. To enable any-region editing, we propose to pre-train an any-order autoregressive text-to-image (T2I) model. Once trained, it is capable of zero-shot image editing and can be easily adapted to NEP for image editing, which achieves a new state-of-the-art on widely used image editing benchmarks. Moreover, our model naturally supports test-time scaling (TTS) through iteratively refining its generation in a zero-shot manner.