Dual-Channel Attention Guidance for Training-Free Image Editing Control in Diffusion Transformers

2026-02-20

AI Summary

Existing training-free image editing methods based on Diffusion Transformers (DiTs) modulate only the Key channels in attention mechanisms, making it difficult to finely balance editing strength and content fidelity. This work proposes a Dual-Channel Attention Guidance (DCAG) framework, which reveals for the first time that both Key and Value projections in DiTs exhibit a bias-increment structure. Leveraging this insight, DCAG jointly controls the Key channels—determining where to attend—and the Value channels—determining what content to aggregate—thereby constructing a two-dimensional intervention parameter space (δₖ, δᵥ). Evaluated on PIE-Bench, the method significantly outperforms single-channel approaches, reducing LPIPS by 4.9% and 3.2% on object removal and addition tasks, respectively, and effectively enhancing local editing precision.

Technology Category

• Computer Vision: Diffusion Models for Vision
• Machine Learning: Large Multimodal Models (LMMs)
• Search and Optimization: Learning to Search

Application Category

• Search and Retrieval-Augmented AI: Retrieval-Augmented Generation (RAG) and multi-modal RAG
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Security and Privacy: Data transparency and provenance

Abstract

Training-free control over editing intensity is a critical requirement for diffusion-based image editing models built on the Diffusion Transformer (DiT) architecture. Existing attention manipulation methods focus exclusively on the Key space to modulate attention routing, leaving the Value space -- which governs feature aggregation -- entirely unexploited. In this paper, we first reveal that both Key and Value projections in DiT's multi-modal attention layers exhibit a pronounced bias-delta structure, where token embeddings cluster tightly around a layer-specific bias vector. Building on this observation, we propose Dual-Channel Attention Guidance (DCAG), a training-free framework that simultaneously manipulates both the Key channel (controlling where to attend) and the Value channel (controlling what to aggregate). We provide a theoretical analysis showing that the Key channel operates through the nonlinear softmax function, acting as a coarse control knob, while the Value channel operates through linear weighted summation, serving as a fine-grained complement. Together, the two-dimensional parameter space $(\delta_k, \delta_v)$ enables more precise editing-fidelity trade-offs than any single-channel method. Extensive experiments on the PIE-Bench benchmark (700 images, 10 editing categories) demonstrate that DCAG consistently outperforms Key-only guidance across all fidelity metrics, with the most significant improvements observed in localized editing tasks such as object deletion (4.9% LPIPS reduction) and object addition (3.2% LPIPS reduction).

Problem

• training-free control
• image editing
• diffusion transformer
• attention manipulation
• editing intensity

Innovation

• Dual-Channel Attention Guidance
• Training-Free Editing
• Diffusion Transformer
• Attention Manipulation
• Value Space Control