Research map / Image generation / Text-to-image diffusion
ComFusion: Enhancing Personalized Generation by Instance-Scene Compositing and Fusion
Yan Hong, Yuxuan Duan, Bo Zhang, Haoxing Chen, Jun Lan, Huijia Zhu, Weiqiang Wang, Jianfu Zhang
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 |
|---|---|---|---|
| replace DreamBooth class prior loss with class-scene prior loss (ablation) | ComFusion · Personalization (seen scenes) | CLIP-T: 0.229 -> 0.301 | Improved |
| add visual-textual matching losses (LIF, LSF) (ablation) | ComFusion · Personalization (seen scenes) | CLIP-I: 0.771 -> 0.814 | Improved |
| add visual fusion loss LIF (ablation) | ComFusion · Personalization (seen scenes) | DINO: 0.586 -> 0.658 | Improved |
| add textual fusion loss LSF (ablation) | ComFusion · Personalization (seen scenes) | CLIP-T: 0.189 -> 0.321 | Improved |
| add fusion losses to DreamBooth (ablation) | DreamBooth · Personalization (seen scenes) | CLIP-T: 0.229 -> 0.232 | No clear effect |
| increase coarse denoising steps tau from 3 to 5 (ablation) | ComFusion · Personalization (seen scenes) | DINO: DINO 0.658 -> 0.698, CLIP-T 0.321 -> 0.309 | Mixed |
| increase lambda_C^S from 1 to 10 (ablation) | ComFusion · Personalization (seen scenes) | DINO: 0.658 -> 0.619 | Hurt |
| increase lambda_F^S from 0.01 to 0.1 (ablation) | ComFusion · Personalization (seen scenes) | DINO: DINO 0.658 -> 0.529, CLIP-T 0.321 -> 0.351 | Mixed |
| ComFusion versus Break-A-Scene on user preference | ComFusion · User study | scene fidelity preference: 20.2% -> 62.1% | Improved |
Closest papers by meaning
- JeDi: Joint-Image Diffusion Models for Finetuning-Free Personalized Text-to-Image Generation (CVPR 2024 · similarity 0.88)
- PaRa: Personalizing Text-to-Image Diffusion via Parameter Rank Reduction (ICLR 2025 · similarity 0.87)
- Direct Consistency Optimization for Robust Customization of Text-to-Image Diffusion models (NeurIPS 2024 · similarity 0.87)
- XVerse: Consistent Multi-Subject Control of Identity and Semantic Attributes via DiT Modulation (NeurIPS 2025 · similarity 0.87)
- Tailored Visions: Enhancing Text-to-Image Generation with Personalized Prompt Rewriting (CVPR 2024 · similarity 0.87)
- MS-Diffusion: Multi-subject Zero-shot Image Personalization with Layout Guidance (ICLR 2025 · similarity 0.86)
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.
