Diffusion Image Editing via Asynchronous Token Decoding

2026-08-10

AI Summary

This work addresses the challenge of global semantic drift in text-guided diffusion-based image editing, which often compromises the identity and background of unedited regions. To mitigate this issue, the authors propose ATDEdit, a novel framework that treats each sampling step during inference as a globally coupled token matrix, enabling simultaneous local editing and background preservation. ATDEdit introduces an asynchronous token decoding mechanism that dynamically identifies editable positions based on conditional surprise, applies conditional corrections to target tokens, and reuses source key/value memories while back-projecting latent representations in preserved regions—eliminating the need for spatial masks or model fine-tuning. Experiments demonstrate that ATDEdit achieves state-of-the-art fidelity on PIE-Bench (PSNR 27.44 dB, LPIPS 0.055) while maintaining strong semantic alignment.

Abstract

Text-guided diffusion image editing aims to modify semantic attributes of an image while preserving its identity, layout, and background. However, naïvely switching the text condition during sampling often causes global drift, as denoising dynamics propagate changes across tokens and can disrupt unedited regions. To address this issue, we propose \textbf{A}synchronous \textbf{T}oken \textbf{D}ecoding \textbf{Edit} (ATDEdit), an inference-time framework that views each sampler step as a parallel update of a globally coupled token matrix and enables token-indexed condition switching with differentiated update policies. Instead of applying synchronous target-conditioned updates to all tokens, ATDEdit estimates editable locations using token-wise conditional surprisal and applies target-conditioned corrections to the selected token set. It supplies source key/value memory at keep-token positions and projects selected keep-token latent rows back to their source values; these operations promote background preservation but do not constitute a pixel-level invariance guarantee. This approach combines local editing and background preservation without external or user-provided spatial masks and without model fine-tuning. On PIE-Bench, ATDEdit achieves the strongest reported preservation metrics, including 27.44~dB PSNR and 0.055 LPIPS, while retaining competitive semantic alignment.

Problem

• diffusion image editing
• text-guided editing
• global drift
• background preservation
• semantic attribute modification

Innovation

• asynchronous token decoding
• diffusion image editing
• conditional surprisal
• background preservation
• text-guided editing