Recent studies have explored the combination of multiple LoRAs to simultaneously generate user-specified subjects and styles. However, most existing approaches fuse LoRA weights using static statistical heuristics that deviate from LoRA's original purpose of learning adaptive feature adjustments and ignore the randomness of sampled inputs. To address this, we propose a dynamic training-free fusion framework that operates throughout the generation process. During the forward pass, at each LoRA-applied layer, we dynamically compute the KL divergence between the base model's original features and those produced by subject and style LoRAs, respectively, and adaptively select the most appropriate weights for fusion. In the reverse denoising stage, we further refine the generation trajectory by dynamically applying gradient-based corrections derived from objective metrics such as CLIP and DINO scores, providing continuous semantic and stylistic guidance. By integrating these two complementary mechanisms-feature-level selection and metric-guided latent adjustment-across the entire diffusion timeline, our method dynamically achieves coherent subject-style synthesis without any retraining. Extensive experiments across diverse subject-style combinations demonstrate that our approach consistently outperforms state-of-the-art LoRA fusion methods both qualitatively and quantitatively.
FreeLoRA: Enabling Training-Free LoRA Fusion for Autoregressive Multi-Subject Personalization
School of Artificial Intelligence, Jilin University(吉林大学人工智能学院);Shanghai Innovation Institute(上海创新研究院);School of Computer Science, Fudan University(复旦大学计算机学院)
Few-shot multi-token DreamBooth with LoRa for style-consistent character generation
Institute of Smart Cities (ISC)(智能城市研究所);Department of Statistics, Computer Science and Mathematics(统计学、计算机科学与数学系);Public University of Navarre (UPNA)(纳瓦拉公共大学)
Training-Free Multi-Style Fusion Through Reference-Based Adaptive Modulation
University of Washington, Seattle, WA 98195, USA(华盛顿大学);Nanjing University of Posts and Telecommunications, Jiangsu, Nanjing, China(南京邮电大学);Nanjing Foreign Language School, Jiangsu, Nanjing, China(南京外国语学校)
SeNeDiF-OOD: Semantic Nested Dichotomy Fusion for Out-of-Distribution Detection Methodology in Open-World Classification. A Case Study on Monument Style Classification
FISTNet: FusIon of STyle-path generative Networks for Facial Style Transfer
University of Sindh(信德大学);West Virginia University(西弗吉尼亚大学);IKLab Inc.(IKLab公司);Tech University of Korea(韩国技术大学);University of Calabria(卡拉布里亚大学);Munster Technological University(芒斯特理工大学)
Semantic Library Adaptation: LoRA Retrieval and Fusion for Open-Vocabulary Semantic Segmentation
The Good AI Lab;University of Toronto(多伦多大学);KTH(瑞典皇家理工学院);AMD(超威半导体公司);Silo AI;King;Technical University of Munich(慕尼黑工业大学);Munich Center for Machine Learning(慕尼黑机器学习中心);Google(谷歌公司);ETH Zurich(苏黎世联邦理工学院);University of Bologna(博洛尼亚大学)
TriLoRA: Integrating SVD for Advanced Style Personalization in Text-to-Image Generation
Institute of Computing Technology, Chinese Academy of Sciences(中国科学院计算技术研究所);Southeast University(东南大学);Institute of Automation, Chinese Academy of Sciences(中国科学院自动化研究所);Shaanxi Normal University(陕西师范大学)
LoRCA: LoRA Cycle Adaptation for Histology to HiP-CT Translation with DINOv3
University College London(伦敦大学学院);University of Cambridge(剑桥大学);Cambridge Institute of Therapeutic Immunology and Infectious Diseases(剑桥治疗免疫学与传染病研究所);Wellcome Trust Sanger Institute(惠康基金会桑格研究所);Cambridge Stem Cell Institute(剑桥干细胞研究所);Centre for AI in Medicine(医学人工智能中心);Cavendish Laboratory(卡文迪什实验室);Medical University of Vienna(维也纳医科大学);European Synchrotron Radiation Facility(欧洲同步辐射装置);CIFAR(加拿大高级研究所);Cambridge University Hospitals NHS Foundation Trust(剑桥大学医院NHS基金会信托基金);NIHR Cambridge Biomedical Research Centre(英国国家健康与研究院剑桥生物医学研究中心)
Scale-Separated Conditioning for Style-Encoder-Free Diffusion Stylization
College of Computing, Georgia Institute of Technology(佐治亚理工学院计算学院);Courant Institute of Mathematical Sciences, New York University(纽约大学柯朗数学科学研究所);Department of Computer Science, Brown University(布朗大学计算机科学系)
Filtering Memorization from Parameter-Space in Diffusion Models
RIKEN AIP(理化学研究所Advanced Institute for Peripheral Research);Science of Tokyo(东京科学大学);University of California, Berkeley(加州大学伯克利分校);Zhejiang University(浙江大学)