mlsanityData to 7 Oct 2026

K-LoRA: Unlocking Training-Free Fusion of Any Subject and Style LoRAs

arXiv 2502.18461 ↗First posted 25 Feb 2025cs.CVcited by 48

Not on the map. Why?

Abstract (arXiv)

Recent studies have explored combining different LoRAs to jointly generate learned style and content. However, existing methods either fail to effectively preserve both the original subject and style simultaneously or require additional training. In this paper, we argue that the intrinsic properties of LoRA can effectively guide diffusion models in merging learned subject and style. Building on this insight, we propose K-LoRA, a simple yet effective training-free LoRA fusion approach. In each attention layer, K-LoRA compares the Top-K elements in each LoRA to be fused, determining which LoRA to select for optimal fusion. This selection mechanism ensures that the most representative features of both subject and style are retained during the fusion process, effectively balancing their contributions. Experimental results demonstrate that the proposed method effectively integrates the subject and style information learned by the original LoRAs, outperforming state-of-the-art training-based approaches in both qualitative and quantitative results.

Key facts. Cited by 48 papers in the map (data through 7 Oct 2026). Builds on 42 of its 51 references that are in the map. First posted on arXiv 25 Feb 2025.

Builds on

42 of 51 references.

References · 42

  1. 2021Learning Transferable Visual Models From Natural Language Supervisioncited by 33,456
  2. 2021High-Resolution Image Synthesis with Latent Diffusion Modelscited by 18,694
  3. 2021LoRA: Low-Rank Adaptation of Large Language Modelscited by 14,084
  4. 2022DreamBooth: Fine Tuning Text-to-Image Diffusion Models for Subject-Driven Generationcited by 3,286
  5. 2017Arbitrary Style Transfer in Real-time with Adaptive Instance Normalizationcited by 1,998
  6. 2023IP-Adapter: Text Compatible Image Prompt Adapter for Text-to-Image Diffusion Modelscited by 1,439
  7. 2022DINO: DETR with Improved DeNoising Anchor Boxes for End-to-End Object Detectioncited by 1,306
  8. 2022Multi-Concept Customization of Text-to-Image Diffusioncited by 1,099
  9. 2024LoRA+: Efficient Low Rank Adaptation of Large Modelscited by 365
  10. 2023InstantBooth: Personalized Text-to-Image Generation without Test-Time Finetuningcited by 343
  11. 2023PhotoMaker: Customizing Realistic Human Photos via Stacked ID Embeddingcited by 324
  12. 2023FastComposer: Tuning-Free Multi-Subject Image Generation with Localized Attentioncited by 313
30 more
  1. 2023VeRA: Vector-based Random Matrix Adaptationcited by 294
  2. 2022Inversion-Based Style Transfer with Diffusion Modelscited by 290
  3. 2022SmartBrush: Text and Shape Guided Object Inpainting with Diffusion Modelcited by 276
  4. 2023Mix-of-Show: Decentralized Low-Rank Adaptation for Multi-Concept Customization of Diffusion Modelscited by 261
  5. 2023P+: Extended Textual Conditioning in Text-to-Image Generationcited by 242
  6. 2023Break-A-Scene: Extracting Multiple Concepts from a Single Imagecited by 230
  7. 2024Diffusion Model-Based Image Editing: A Surveycited by 212
  8. 2023StyleDrop: Text-to-Image Generation in Any Stylecited by 197
  9. 2023ZipLoRA: Any Subject in Any Style by Effectively Merging LoRAscited by 194
  10. 2023Style Injection in Diffusion: A Training-free Approach for Adapting Large-scale Diffusion Models for Style Transfercited by 194
  11. 2023Cones: Concept Neurons in Diffusion Models for Customized Generationcited by 150
  12. 2023Localizing Object-level Shape Variations with Text-to-Image Diffusion Modelscited by 142
  13. 2024Implicit Style-Content Separation using B-LoRAcited by 135
  14. 2023LoRA-FA: Efficient and Effective Low Rank Representation Fine-tuningcited by 132
  15. 2023A Neural Space-Time Representation for Text-to-Image Personalizationcited by 123
  16. 2017StyleBank: An Explicit Representation for Neural Image Style Transfercited by 89
  17. 2024Multi-LoRA Composition for Image Generationcited by 80
  18. 2024CSGO: Content-Style Composition in Text-to-Image Generationcited by 76
  19. 2023ProSpect: Prompt Spectrum for Attribute-Aware Personalization of Diffusion Modelscited by 75
  20. 2023Delta-LoRA: Fine-Tuning High-Rank Parameters with the Delta of Low-Rank Matricescited by 72
  21. 2020Arbitrary Style Transfer via Multi-Adaptation Networkcited by 70
  22. 2024MagMax: Leveraging Model Merging for Seamless Continual Learningcited by 67
  23. 2024LoRA-Composer: Leveraging Low-Rank Adaptation for Multi-Concept Customization in Training-Free Diffusion Modelscited by 55
  24. 2024Mixture-of-Subspaces in Low-Rank Adaptationcited by 47
  25. 2024LoRA-drop: Efficient LoRA Parameter Pruning based on Output Evaluationcited by 38
  26. 2024How to Continually Adapt Text-to-Image Diffusion Models for Flexible Customization?cited by 32
  27. 2024MC2: Multi-concept Guidance for Customized Multi-concept Generationcited by 26
  28. 2024FreeTuner: Any Subject in Any Style with Training-free Diffusioncited by 24
  29. 2024MiLoRA: Efficient Mixture of Low-Rank Adaptation for Large Language Models Fine-tuningcited by 19
  30. 2024MagicTailor: Component-Controllable Personalization in Text-to-Image Diffusion Modelscited by 14

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2025: 24; 2026: 24.

48 papers.

Sort:

By area:

  1. 2025Decouple and Orthogonalize: A Data-Free Framework for LoRA Mergingcited by 17
  2. 2025LoRA-Mixer: Coordinate Modular LoRA Experts Through Serial Attention Routingcited by 16
  3. 2025ConsisLoRA: Enhancing Content and Style Consistency for LoRA-based Style Transfercited by 9
  4. 2025MultLFG: Training-free Multi-LoRA composition using Frequency-domain Guidancecited by 6
  5. 2025LiON-LoRA: Rethinking LoRA Fusion to Unify Controllable Spatial and Temporal Generation for Video Diffusioncited by 6
  6. 2025FreeLoRA: Enabling Training-Free LoRA Fusion for Autoregressive Multi-Subject Personalizationcited by 5
  7. 2025Subject or Style: Adaptive and Training-Free Mixture of LoRAscited by 5
  8. 2025TARA: Token-Aware LoRA for Composable Personalization in Diffusion Modelscited by 5
  9. 2025SplitFlux: Learning to Decouple Content and Style from a Single Imagecited by 5
  10. 2025The Consistency Critic: Correcting Inconsistencies in Generated Images via Reference-Guided Attentive Alignmentcited by 5
  11. 2026ReStyle-TTS: Relative and Continuous Style Control for Zero-Shot Speech Synthesiscited by 5
  12. 2026DiffGraph: An Automated Agent-driven Model Merging Framework for In-the-Wild Text-to-Image Generationcited by 4
36 more
  1. 2025Calligrapher: Freestyle Text Image Customizationcited by 3
  2. 2025FantasyStyle: Controllable Stylized Distillation for 3D Gaussian Splattingcited by 3
  3. 2025LoRAtorio: An intrinsic approach to LoRA Skill Compositioncited by 3
  4. 2025AgeBooth: Controllable Facial Aging and Rejuvenation via Diffusion Modelscited by 3
  5. 2025LoRA on the Go: Instance-level Dynamic LoRA Selection and Mergingcited by 3
  6. 2026CleanStyle: Plug-and-Play Style Conditioning Purification for Text-to-Image Stylizationcited by 3
  7. 2026OrthoFuse: Training-free Riemannian Fusion of Orthogonal Style-Concept Adapters for Diffusion Modelscited by 2
  8. 2025Localized LoRA: A Structured Low-Rank Approximation for Efficient Fine-Tuningcited by 1
  9. 2025CraftGraffiti: Exploring Human Identity with Custom Graffiti Art via Facial-Preserving Diffusion Modelscited by 1
  10. 2025OmniStyle2: Learning to Stylize by Learning to Destylizecited by 1
  11. 2026Mixture of Style Experts for Diverse Image Stylizationcited by 1
  12. 2026SSR-Merge: Subspace Signal Routing for Training-Free LoRA Merging in Diffusion Modelscited by 1
  13. 2026DreamStyle3D: Efficient 3D Stylized Asset Generation via Dual-Attention Disentanglementcited by 1
  14. 2025RichControl: Structure- and Appearance-Rich Training-Free Spatial Control for Text-to-Image Generationcited by 0
  15. 2025SPG: Style-Prompting Guidance for Style-Specific Content Creationcited by 0
  16. 2025FreeFuse: Multi-Subject LoRA Fusion via Adaptive Token-Level Routing at Test Timecited by 0
  17. 2025Expanding the Content-Style Frontier: a Balanced Subspace Blending Approach for Content-Style LoRA Fusioncited by 0
  18. 2025V-Shuffle: Zero-Shot Style Transfer via Value Shufflecited by 0
  19. 2025NP-LoRA: Null Space Projection for Subject-Style LoRA Fusioncited by 0
  20. 2026Understanding Affective Adaptation in Multimodal Foundation Models: Emergent Functional Specializationcited by 0
  21. 2026GimmBO: Interactive Generative Image Model Merging via Bayesian Optimizationcited by 0
  22. 2026Training-Free Multi-Concept Image Editingcited by 0
  23. 2026A Diffusion-Driven Fine-Grained Nodule Synthesis Framework for Enhanced Lung Nodule Detection from Chest Radiographscited by 0
  24. 2026The Quadratic Geometry of Flow Matching: Semantic Granularity Alignment for Text-to-Image Synthesiscited by 0
  25. 2026SpiralDiff: Spiral Diffusion with LoRA for RGB-to-RAW Conversion Across Camerascited by 0
  26. 2026Abstraction in Style: Beyond Texture and Colorcited by 0
  27. 2026FREE-Switch: Frequency-based Dynamic LoRA Switch for Style Transfercited by 0
  28. 2026Mining Attribute Subspaces for Efficient Fine-tuning of 3D Foundation Modelscited by 0
  29. 2026AD-Relight: Training-Free Banner Relighting via Illumination Translation with Diffusion Priorscited by 0
  30. 2026Training-Free Multi-Concept LoRA Composition with Prompt-Aware Weightingcited by 0
  31. 2026Context Recycling for Long-Horizon LLM Inferencecited by 0
  32. 2026Disco-LoRA: Disentangled Composition of Content, Style, and Motion for Multi-concept Video Customizationcited by 0
  33. 2026LILAC: Layer-Wise Independent LoRAs and Cascaded Conditioning for Multi-Concept Customization of Diffusion Modelscited by 0
  34. 2026DeCo: Zero-Shot Industrial Anomaly Generation through Decoupling and Recouplingcited by 0
  35. 2026RefDiT: Local Attribute Guidance in Reference-Based Image Generationcited by 0
  36. 2026What Does the Rank Buy? A Spectral and Distributional Analysis of Low-Rank Adaptationcited by 0

Similar, not cited

Close in wording, no citation either way

  1. 2024LoRA.rar: Learning to Merge LoRAs via Hypernetworks for Subject-Style Conditioned Image Generation
  2. 2024A LoRA is Worth a Thousand Pictures
  3. 2025LoRAverse: A Submodular Framework to Retrieve Diverse Adapters for Diffusion Models
  4. 2025AttenST: A Training-Free Attention-Driven Style Transfer Framework with Pre-Trained Diffusion Models
  5. 2025Training-Free Multi-Style Fusion Through Reference-Based Adaptive Modulation
  6. 2026FreeStyle: Free Control of Style-Content Dual-Reference Generation from Community LoRA Mining
  7. 2024Block-wise LoRA: Revisiting Fine-grained LoRA for Effective Personalization and Stylization in Text-to-Image Generation
  8. 2026A Unified Study of LoRA Variants: Taxonomy, Review, Codebase, and Empirical Evaluation