Research map / Image generation / Text-to-image diffusion
ZipLoRA: Any Subject in Any Style by Effectively Merging LoRAs
Viraj Shah, Nataniel Ruiz, Forrester Cole, Erika Lu, Svetlana Lazebnik, Yuanzhen Li, Varun Jampani
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| merge LoRAs with learned per-column coefficients and cosine-similarity penalty instead of direct arithmetic sum | ZipLoRA · Subject-style merging | subject-alignment (DINO): 0.357 -> 0.420 | Improved |
| merge LoRAs with ZipLoRA instead of direct arithmetic sum | ZipLoRA · Subject-style merging | style-alignment (CLIP-I): 0.702 -> 0.699 | No clear effect |
| use ZipLoRA instead of direct merge, judged by users | ZipLoRA · User study | % preference: 82.7% prefer ZipLoRA | Improved |
| use ZipLoRA instead of joint training | ZipLoRA · Runtime | seconds: 3540 -> 560 | Improved |
| optimize only merger coefficient vectors with frozen LoRAs | ZipLoRA · Efficiency | GPU memory / storage: 38GB -> 21GB; 360MB -> 6.5MB | Improved |
| use SDXL as base instead of SD v1.5 for style LoRAs (ablation) | ZipLoRA · Style learning from one image | stylization quality | Improved |
| apply ZipLoRA to styles whose reference image contains content such as a cliff | ZipLoRA · Failure cases | content leakage | Hurt |
| zero out 80-90% of lowest-magnitude LoRA update elements (ablation) | LoRA update matrices · Qualitative (watercolor teddy bear) | generation quality | No clear effect |
Later papers that compared with this method
5 later accepted papers report ZipLoRA as a baseline in their results tables; 4 of them report a better number on most of the results they share with it.
- DCoAR: Deep Concept Injection into Unified Autoregressive Models for Personalized Text-to-Image Generation (CVPR 2026 · better on 2 of 3 shared results)
- Contrastive Test-Time Composition of Multiple LoRA Models for Image Generation (ICCV 2025 · better on 6 of 6 shared results)
- LoRA.rar: Learning to Merge LoRAs via Hypernetworks for Subject-Style Conditioned Image Generation (ICCV 2025 · better on 5 of 6 shared results)
- K-LoRA: Unlocking Training-Free Fusion of Any Subject and Style LoRAs (CVPR 2025 · better on 4 of 5 shared results)
- Implicit Style-Content Separation using B-LoRA (ECCV 2024 · better on 1 of 2 shared results)
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- K-LoRA: Unlocking Training-Free Fusion of Any Subject and Style LoRAs (CVPR 2025 · similarity 0.88)
- NP-LoRA: Null Space Projection for Subject-Style LoRA Fusion (NeurIPS 2026 · similarity 0.87)
- LoRACLR: Contrastive Adaptation for Customization of Diffusion Models (CVPR 2025 · similarity 0.85)
- Merging LoRAs like Playing LEGO: Pushing the Modularity of LoRA to Extremes Through Rank-Wise Clustering (ICLR 2025 · similarity 0.84)
- DynASyn: Multi-Subject Personalization Enabling Dynamic Action Synthesis (AAAI 2025 · similarity 0.84)
- Text-to-LoRA: Instant Transformer Adaption (ICML 2025 · similarity 0.84)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
