Research map / Image generation / Text-to-image diffusion
Investigating Style Similarity in Diffusion Models
Gowthami Somepalli, Anubhav Gupta, Kamal Gupta, Shramay Palta, Micah Goldblum, Jonas A. Geiping, Abhinav Shrivastava, Tom Goldstein
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 |
|---|---|---|---|
| train with multi-label style contrastive loss plus style-preserving SSL on ContraStyles starting from CLIP ViT-L | CLIP ViT-L · WikiArt zero-shot | mAP@1: 59.4 -> 64.56 | Improved |
| train CSD ViT-B versus CLIP ViT-B/16 | CLIP ViT-B/16 · DomainNet | mAP@1: 73.7 -> 78.3 | Improved |
| use CSD instead of CLIP ViT-L for content-constrained style retrieval | CSD vs CLIP ViT-L · SD 2.1 content-constrained, woman | mAP@1: 7.4 -> 10.8 | Improved |
| use CSD instead of CLIP ViT-L on user-generated prompts | CSD vs CLIP ViT-L · SD 2.1 user-generated prompts | mAP@1: 6.1 -> 5.7 | No clear effect |
| use CSD instead of CLIP ViT-L on simple prompts | CSD vs CLIP ViT-L · SD 2.1 simple prompts | mAP@1: 22.3 -> 24.5 | Improved |
| use CSD ViT-B instead of SSCD RN-50 on DomainNet | CSD ViT-B · DomainNet | Recall@10: 95.0 -> 94.3 | Hurt |
| compare untrained humans against feature extractors on style matching | human study · Artchive 4-way style matching | accuracy | No clear effect |
| exclude photometric augmentations (blur, color jitter) in SSL branch | CSD training · style retrieval | retrieval | Improved |
Closest papers by meaning
- Stylebreeder: Exploring and Democratizing Artistic Styles through Text-to-Image Models (NeurIPS 2024 · similarity 0.90)
- Style Injection in Diffusion: A Training-free Approach for Adapting Large-scale Diffusion Models for Style Transfer (CVPR 2024 · similarity 0.89)
- SigStyle: Signature Style Transfer via Personalized Text-to-Image Models (AAAI 2025 · similarity 0.89)
- FiVA: Fine-grained Visual Attribute Dataset for Text-to-Image Diffusion Models (NeurIPS 2024 · similarity 0.88)
- Conditional Balance: Improving Multi-Conditioning Trade-Offs in Image Generation (CVPR 2025 · similarity 0.88)
- Gender Artifacts from Art History to Text-to-Image Generation (NeurIPS 2026 · similarity 0.88)
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.
