Research map / Image generation / Text-to-image diffusion
StyleDrop: Text-to-Image Synthesis of Any Style
Kihyuk Sohn, Lu Jiang, Jarred Barber, Kimin Lee, Nataniel Ruiz, Dilip Krishnan, Huiwen Chang, Yuanzhen Li, Irfan Essa, Michael Rubinstein, Yuan Hao, Glenn Entis and 2 more
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 |
|---|---|---|---|
| style-tune Muse with StyleDrop (HF) instead of plain Muse prompting | Muse · 6 styles, Parti prompts | CLIP Style score: 0.556 -> 0.694 | Improved |
| style-tune Muse with StyleDrop (HF) instead of plain Muse prompting | Muse · 6 styles, Parti prompts | CLIP Text score: 0.323 -> 0.322 | No clear effect |
| StyleDrop on Muse instead of DreamBooth on Imagen | StyleDrop · 6 styles, Parti prompts | CLIP Style score: 0.644 -> 0.694 | Improved |
| StyleDrop (R1) instead of DreamBooth on Imagen | StyleDrop · User study | style preference (%): 9.7% -> 86.0% | Improved |
| add iterative training with human feedback to round 1 (ablation) | StyleDrop · 6 styles, Parti prompts | CLIP Text score: 0.313 -> 0.322 | Improved |
| add iterative training with human feedback to round 1 (ablation) | StyleDrop · 6 styles, Parti prompts | CLIP Style score: 0.705 -> 0.694 | Hurt |
| use CLIP feedback instead of human feedback for iterative training (ablation) | StyleDrop · 6 styles, Parti prompts | CLIP Text score: 0.322 -> 0.329 | Improved |
| use human feedback instead of CLIP feedback for iterative training (ablation) | StyleDrop · 6 styles, Parti prompts | CLIP Style score: 0.673 -> 0.694 | Improved |
| select synthetic images by human feedback instead of random for iterative training (ablation) | StyleDrop · 6 styles, Parti prompts | CLIP Style score: 0.678 -> 0.694 | Improved |
| use iterative training with careful selection (CLIP/human) instead of random (ablation) | StyleDrop · banana/bottle ablation | CLIP Text score: 0.322 -> 0.339 (CLIP), 0.328 (HF) | Improved |
| use descriptive style descriptor instead of rare token on Muse (ablation) | StyleDrop · 6 styles, Parti prompts | CLIP Style score: 0.654 -> 0.705 | Improved |
| use descriptive style prompt instead of rare token on Muse (ablation) | StyleDrop · qualitative, melting golden 3d | visual style consistency | No clear effect |
Closest papers by meaning
- Style Aligned Image Generation via Shared Attention (CVPR 2024 · similarity 0.88)
- CSGO: Content-Style Composition in Text-to-Image Generation (NeurIPS 2025 · similarity 0.88)
- StyleStudio: Text-Driven Style Transfer with Selective Control of Style Elements (CVPR 2025 · similarity 0.87)
- SigStyle: Signature Style Transfer via Personalized Text-to-Image Models (AAAI 2025 · similarity 0.87)
- FineStyle: Fine-grained Controllable Style Personalization for Text-to-image Models (NeurIPS 2024 · similarity 0.87)
- StyleSSP: Sampling StartPoint Enhancement for Training-free Diffusion-based Method for Style Transfer (CVPR 2025 · similarity 0.87)
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.
