A Training-Free Style-Personalization via SVD-Based Feature Decomposition

2025/07/06 by Kyoungmin Lee, Ji‐Hun Park, Lee, Kyoungmin +11
Computer Science · #Authorship Attribution and Profiling #Computer Vision and Pattern Recognition (cs.CV) #FOS: Computer and information sciences #Recommender Systems and Techniques #Speech Recognition and Synthesis

paper · pdf · doi:10.48550/arxiv.2507.04482

openalex publication_date 2025/07/06 · openalex created_date 2025/10/20 · openalex updated_date 2026/08/03

Abstract

We present a training-free framework for style-personalized image generation that operates during inference using a scale-wise autoregressive model. Our method generates a stylized image guided by a single reference style while preserving semantic consistency and mitigating content leakage. Through a detailed step-wise analysis of the generation process, we identify a pivotal step where the dominant singular values of the internal feature encode style-related components. Building upon this insight, we introduce two lightweight control modules: Principal Feature Blending, which enables precise modulation of style through SVD-based feature reconstruction, and Structural Attention Correction, which stabilizes structural consistency by leveraging content-guided attention correction across fine stages. Without any additional training, extensive experiments demonstrate that our method achieves competitive style fidelity and prompt fidelity compared to fine-tuned baselines, while offering faster inference and greater deployment flexibility.

Citations

• AlignedGen: Aligning Style Across Generated Images
• Exploring Multimodal Diffusion Transformers for Enhanced Prompt-based Image Editing
• CSD-VAR: Content-Style Decomposition in Visual Autoregressive Models
• StyleAR: Customizing Multimodal Autoregressive Model for Style-Aligned Text-to-Image Generation
• A Training-Free Style-aligned Image Generation with Scale-wise Autoregressive Model
• Infinity: Scaling Bitwise AutoRegressive Modeling for High-Resolution Image Synthesis
• UnZipLoRA: Separating Content and Style from a Single Image
• Switti: Designing Scale-Wise Transformers for Text-to-Image Synthesis
• HART: Efficient Visual Generation with Hybrid Autoregressive Transformer
• CSGO: Content-Style Composition in Text-to-Image Generation
• Visual Autoregressive Modeling: Scalable Image Generation via Next-Scale Prediction
• Implicit Style-Content Separation using B-LoRA
• VideoPoet: A Large Language Model for Zero-Shot Video Generation
• Style Injection in Diffusion: A Training-free Approach for Adapting Large-scale Diffusion Models for Style Transfer
• PhotoMaker: Customizing Realistic Human Photos via Stacked ID Embedding
• Style Aligned Image Generation via Shared Attention
• ZipLoRA: Any Subject in Any Style by Effectively Merging LoRAs
• AesPA-Net: Aesthetic Pattern-Aware Style Transfer Networks
• SDXL: Improving Latent Diffusion Models for High-Resolution Image Synthesis
• StyleDrop: Text-to-Image Generation in Any Style
• BLIP-Diffusion: Pre-trained Subject Representation for Controllable Text-to-Image Generation and Editing
• Training-free Content Injection using h-space in Diffusion Models
• Zero-Shot Contrastive Loss for Text-Guided Diffusion Image Style Transfer
• Scaling up GANs for Text-to-Image Synthesis
• ELITE: Encoding Visual Concepts into Textual Embeddings for Customized Text-to-Image Generation
• Muse: Text-To-Image Generation via Masked Generative Transformers
• Plug-and-Play Diffusion Features for Text-Driven Image-to-Image Translation
• InstructPix2Pix: Learning to Follow Image Editing Instructions
• Scaling Instruction-Finetuned Language Models
• LAION-5B: An open large-scale dataset for training next generation image-text models
• Diffusion-based Image Translation using Disentangled Style and Content Representation
• DreamFusion: Text-to-3D using 2D Diffusion
• DreamBooth: Fine Tuning Text-to-Image Diffusion Models for Subject-Driven Generation
• DreamBooth: Fine Tuning Text-to-Image Diffusion Models for Subject-Driven Generation
• Scaling Autoregressive Models for Content-Rich Text-to-Image Generation
• Photorealistic Text-to-Image Diffusion Models with Deep Language Understanding
• Text2LIVE: Text-Driven Layered Image and Video Editing
• MaskGIT: Masked Generative Image Transformer
• High-Resolution Image Synthesis with Latent Diffusion Models
• High-Resolution Image Synthesis with Latent Diffusion Models
• CLIPstyler: Image Style Transfer with a Single Text Condition
• AdaAttN: Revisit Attention Mechanism in Arbitrary Neural Style Transfer
• Learning Transferable Visual Models From Natural Language Supervision
• Zero-Shot Text-to-Image Generation
• Conceptual 12M: Pushing Web-Scale Image-Text Pre-Training To Recognize Long-Tail Visual Concepts
• Taming Transformers for High-Resolution Image Synthesis
• An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale
• Arbitrary Style Transfer via Multi-Adaptation Network
• A Closed-form Solution to Universal Style Transfer
• Attention-aware Multi-stroke Style Transfer
• Arbitrary Style Transfer with Style-Attentional Networks
• Neural Discrete Representation Learning
• Attention Is All You Need
• Universal Style Transfer via Feature Transforms
• Arbitrary Style Transfer in Real-time with Adaptive Instance Normalization
• Microsoft COCO: Common Objects in Context

Related

• A Training-Free Style-aligned Image Generation with Scale-wise Autoregressive Model
• Only-Style: Stylistic Consistency in Image Generation without Content Leakage
• CSD-VAR: Content-Style Decomposition in Visual Autoregressive Models
• LLM-Enabled Style and Content Regularization for Personalized Text-to-Image Generation
• AttenST: A Training-Free Attention-Driven Style Transfer Framework with Pre-Trained Diffusion Models