Research map / Image generation / Text-to-image diffusion
Be Yourself: Bounded Attention for Multi-Subject Text-to-Image Generation
Omer Dahary, Or Patashnik, Kfir Aberman, Danny Cohen-Or
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 |
|---|---|---|---|
| use Bounded Attention in place of vanilla Stable Diffusion for counting | Stable Diffusion · DrawBench counting | Recall: 0.78 -> 0.88 | Improved |
| use Bounded Attention in place of Layout-guidance for spatial relations | Stable Diffusion · DrawBench spatial | Accuracy: 0.35 -> 0.36 | Improved |
| generate each subject separately and merge (MultiDiffusion) instead of single process | layout-guided generation · DrawBench counting | Recall: 0.78 -> 0.55 vs vanilla SD | Hurt |
| remove Bounded Guidance (guide without bounded masks) (ablation) | Bounded Attention · qualitative SD/SDXL examples | layout alignment | Improved |
| add Bounded Denoising (ablation) | Bounded Attention · qualitative SD/SDXL examples | semantic leakage | Improved |
| add subject mask refinement in later denoising stages (ablation) | Bounded Attention · qualitative SD/SDXL examples | fine detail preservation | Improved |
| choose Bounded Attention over Layout-guidance in user study | Stable Diffusion · user study | preference probability: 0.85 vs LG, 0.72 vs BD, 0.95 vs MD | Improved |
Closest papers by meaning
- Devil is in the Detail: Towards Injecting Fine Details of Image Prompt in Image Generation via Conflict-free Guidance and Stratified Attention (CVPR 2025 · similarity 0.90)
- OneActor: Consistent Subject Generation via Cluster-Conditioned Guidance (NeurIPS 2024 · similarity 0.89)
- CONFORM: Contrast is All You Need for High-Fidelity Text-to-Image Diffusion Models (CVPR 2024 · similarity 0.89)
- Grounded Text-to-Image Synthesis with Attention Refocusing (CVPR 2024 · similarity 0.89)
- R&B: Region and Boundary Aware Zero-shot Grounded Text-to-image Generation (ICLR 2024 · similarity 0.89)
- MS-Diffusion: Multi-subject Zero-shot Image Personalization with Layout Guidance (ICLR 2025 · 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.
