Jul 2026· Neural Networks· Vol 205 Pt A, pp.
109412
· 1 citation· 52 references
Medicine
TL;DR
A novel training-free tri-branch denoising framework that effectively decouples structure from style via attention manipulation is proposed that outperforms existing state-of-the-art approaches and achieves consistent improvements in structural consistency and stylistic coherence for image composition.
Abstract
Leveraging the robust generative priors of diffusion models, image composition has achieved remarkable progress. However, existing approaches continue to grapple with a persistent dilemma: the trade-off between maintaining the structural fidelity of the source object and achieving deep stylistic harmonization with the background. We attribute this limitation to two primary factors: 1) the insufficient disentanglement of geometric structure and visual appearance in current architectures, leading to conflicts during the generation process; and 2) the reliance on global statistical alignment techniques, which merely adjust tonal distributions but fail to capture complex semantic stylistic patterns. To address these challenges, we propose a novel training-free tri-branch denoising framework that effectively decouples structure from style via attention manipulation. Specifically, we propose two core mechanisms. Semantic Injection employs self attention maps to separate an object's spatial structure from its visual appearance. Style Guidance adapts advanced attention based style transfer techniques to the composition task for the first time. Comprehensive experimental results show that our method outperforms existing state-of-the-art approaches and achieves consistent improvements in structural consistency and stylistic coherence for image composition.
A diffusion-based framework with a stage-wise multi-condition guidance mechanism that enhances both structural and textural fidelity and compares with recent image-to-image translation and diffusion-based baselines to observe competitive performance in both visual coherence and identity preservation.
Yue Que, Xuegui Cheng, Shuqian Shi et al.· Neural Networks· 0 citations
Image-guided style transfer aims to apply the artistic characteristics of a style image to a content image while preserving its semantic structure and layout. Despite advances in diffusion-based methods, existing approaches often face challenges in disentangling content and style, particularly when independently optimized adapters are naively combined, causing conflicts between adapters and limiting controllability over the content-style balance in inference. We further demonstrate that training-free structural guidance directly derived from the content image through the internal attention of pre-trained model outperforms a dedicated content LoRA adapter in terms of structural fidelity and computational efficiency. Building on these observations, we propose AnyStyle, a streamlined framework for image-guided style transfer. The framework adopts a unified single-adapter paradigm for coherent style capture from the style image and incorporates training-free structural guidance from the content image, thus avoiding complex entanglement between multiple adapters and improving controllability and stability. Extensive experiments show that our method delivers competitive quantitative performance and significantly improved perceptual quality. Code is available at https://github.com/Yvan1001/AnyStyle.
Face photo-sketch translation is a significant task in cross-domain image generation. Traditional methods often struggle to balance global structure and local details, and they lack the ability of adaptive cross-domain feature fusion. To address these challenges, this paper presents a novel image generation method based on generative adversarial networks (GANs). In the early stage of the encoder, a Global-Local Fast Fourier Convolution module is introduced. The global branch employs Fast Fourier Convolution to capture long-range dependencies, while the local branch utilizes depthwise separable and standard convolutions to extract local textures. This parallel approach enables the simultaneous representation of global and local features. Additionally, a bi-directional gated channel attention module is incorporated to aggregate complementary information from both branches, thereby balancing overall consistency and fine detail realism. Furthermore, an Adaptive Cross-Domain Attention module is designed to dynamically select reference image features based on regional correlation: real details are preserved in highly correlated regions, while style alignment and fusion are performed in weakly correlated regions, effectively reducing interference from irrelevant features. Experimental results on the CUFS and CUFSF datasets demonstrate that the proposed method achieves superior visual quality outperforms or ranks second in terms of the LPIPS, FID, and FSIM metrics, validating its effectiveness in high-quality cross-domain image generation.
Lei Zhang, Houpan Zhou· International Conference on...· 0 citations