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

Guandong Li

cs.CVFeb 20, 2026

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

Authors: Guandong Li

Organizations: iFLYTEK

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 $(δ_k, δ_v)$ 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).

Figures & tables

Figure 1: Overview of DCAG. The Key channel controls attention routing (coarse, nonlinear) and the Value channel controls feature aggregation (fine, linear). Both channels apply bias-delta rescaling independently before the joint attention computation.

Method	$\delta_{k}$	$\delta_{v}$	LPIPS $\downarrow$	SSIM $\uparrow$	PSNR $\uparrow$	MSE $\downarrow$
No Guidance	1.00	1.00	0.3523	0.6307	15.56	3902
$\delta_{k}=1.10$ group
GRAG	1.10	1.00	0.2588	0.7444	17.93	2588
DCAG	1.10	1.05	0.2575	0.7453	17.92	2585
DCAG	1.10	1.10	0.2555	0.7468	17.90	2570
DCAG	1.10	1.15	0.2542	0.7477	17.89	2557

Table 1: Main results on PIE-Bench (700 images). DCAG consistently outperforms GRAG (K-only) at matched

\delta_{k}

. Best results per group in bold .

Category	GRAG	DCAG	$\Delta$	$N$
Random	0.2676	0.2594	$\downarrow$ 3.1%	140
Change Object	0.2364	0.2299	$\downarrow$ 2.7%	80
Add Object	0.1097	0.1068	$\downarrow$ 2.7%	80
Delete Object	0.1752	0.1677	$\downarrow$ 4.3%	80
Change Attribute	0.1538	0.1554	$\uparrow$ 1.0%	40
Change Count	0.2688	0.2654	$\downarrow$ 1.3%	40

Table 2: Per-category LPIPS

\downarrow

comparison at

\delta_{k}\!=\!1.10

. DCAG (

\delta_{v}\!=\!1.15

) improves 8 out of 10 categories.

$\delta_{v}$	LPIPS $\downarrow$	SSIM $\uparrow$	PSNR $\uparrow$	MSE $\downarrow$
1.00	0.2588	0.7444	17.93	2588
1.05	0.2575	0.7453	17.92	2585
1.10	0.2555	0.7468	17.90	2570
1.15	0.2542	0.7477	17.89	2557
1.20	0.2546	0.7466	17.81	2563

Table 3: V-channel sweep at

\delta_{k}=1.10

. LPIPS improves monotonically up to

\delta_{v}=1.15

, then saturates.

Figure 2: V-channel sweep at fixed

\delta_{k}=1.10

. LPIPS (blue, left axis) improves monotonically up to

\delta_{v}=1.15

, then saturates. SSIM (red, right axis) follows the same pattern. The sweet spot at

\delta_{v}\approx 1.15

reflects the linear nature of the Value channel.

Figure 3: Per-category LPIPS comparison between K-only (

\delta_{k}\!=\!1.10

) and DCAG (

\delta_{k}\!=\!1.10,\delta_{v}\!=\!1.15

). Top: absolute LPIPS values. Bottom: relative improvement (%). DCAG improves 8 out of 10 categories, with the largest gains in Delete Object (

\downarrow

4.3%) and Change Background (

\downarrow

4.2%).

Figure 4: Delta-to-bias ratio heatmaps across 60 layers (x-axis) and 24 denoising steps (y-axis). Left: Key space. Right: Value space. Both spaces exhibit pervasive bias-delta structure, with Value ratios (mean 2.45) consistently higher than Key ratios (mean 1.79). The weak correlation ( $r=-0.17$ ) between the two confirms their structural independence.

Figure 5: Qualitative comparison across editing categories. DCAG ( $\delta_{k}\!=\!1.10,\delta_{v}\!=\!1.15$ ) better preserves non-edited regions compared to K-only ( $\delta_{k}\!=\!1.10$ ), while maintaining editing quality. K-only ( $\delta_{k}\!=\!1.15$ ) achieves higher fidelity but at the cost of reduced editing strength.

Diffusion-Based Image Editing

Diffusion-based image editing changes image content, style, or attributes while aiming to preserve the parts users want to keep.

See all 225 papers

Jun 11, 2026

DuET: Dual Expert Trajectories for Diffusion Image Editing

Recent diffusion editors perform diverse instruction-based edits while conditioning on the source image at every denoising step. Yet persistent source-image conditioning can limit how fully an edit is executed and how natural the result appears, especially when the target scene diverges substantially from the input. We introduce DuET (Dual Expert Trajectories), a training-free inference method that temporarily relaxes source-image conditioning by transitioning through a text-to-image phase before returning to edit mode, allowing the denoising trajectory to move toward the target distribution while retaining the structural benefits of image-conditioned editing. Without modifying model weights or increasing sampling cost, DuET consistently improves instruction relevance, semantic fidelity, and perceptual quality across diverse models and benchmarks. In some cases, these gains come with a modest reduction in source-image preservation, revealing a predictable trade-off between source preservation and edit fidelity.

Lidia Troeshestova, Alexander Ustyuzhanin, Sergey Kastryulin

HSE University · Yandex

Text-Guided Image Editing Diffusion-Based Image Editing Conditional Diffusion Models Training-Free Image Editing Diffusion-Based Denoising

Ask PDF

May 11, 2026

Masked Generative Transformer Is What You Need for Image Editing

Diffusion models dominate image editing, yet their global denoising mechanism entangles edited regions with surrounding context, causing modifications to propagate into areas that should remain intact. We propose a fundamentally different approach by leveraging Masked Generative Transformers (MGTs), whose localized token-prediction paradigm naturally confines changes to intended regions. We present EditMGT, an MGT-based editing framework that is the first of its kind. Our approach employs multi-layer attention consolidation to aggregate cross-attention maps into precise edit localization signals, and region-hold sampling to explicitly prevent token flipping in non-target areas. To support training, we construct CrispEdit-2M, a 2M-sample high-resolution (>1024) editing dataset spanning seven categories. With only 960M parameters, EditMGT achieves state-of-the-art image similarity on multiple benchmarks while delivering 6x faster editing, demonstrating that MGTs offer a compelling alternative to diffusion-based editing.

Wei Chow, Linfeng Li, Xian Sun +14

1ByteDance · 2National University of Singapore · 3Duke University +2

Transformer Text-Guided Image Editing Image Editing

Ask PDF

Sep 28, 2026

Attention-Scoped Guidance: Training-Free Spatial Control for Image Editing

Instruction-guided image editing should change what the instruction names and leave the rest of the image untouched. In dual classifier-free guidance (CFG), an editor combines two directions at every denoising step, one that pushes toward the instructed edit and one that pulls back toward the source image, using global weights. We introduce Attention-Scoped Guidance (ASG), a sampler wrapper that makes these weights spatial. It reads a soft support map from the instruction attention that the editor already computes, then weakens text guidance where support is low and strengthens image anchoring where support is high. The wrapper requires no training, no external mask, and no additional network evaluation. On the full MagicBrush and PIE-Bench++ splits, ASG improves preservation-oriented metrics, leading three of four MagicBrush metrics and PIE-Bench++ background PSNR. A dose-matched control that removes the spatial placement loses up to 0.73 CLIP on PIE-Bench++, confirming that the spatial allocation itself carries the gain.

Zeyan Li, Wei Zhou, Hadi Amirpour +3

Shanghai Jiao Tong University · Cardiff University · University of Klagenfurt +1

Classifier-Free Guidance Text-Guided Image Editing Diffusion Guidance Attention Control in Diffusion Models Training-Free Image Editing

Ask PDF

Abstract

Diffusion-Based Image Editing

Explore similar work

DuET: Dual Expert Trajectories for Diffusion Image Editing

Masked Generative Transformer Is What You Need for Image Editing

Attention-Scoped Guidance: Training-Free Spatial Control for Image Editing