Research map / Image generation / Text-to-image diffusion
Linguistic Binding in Diffusion Models: Enhancing Attribute Correspondence through Attention Map Alignment
Royi Rassin, Eran Hirsch, Daniel Glickman, Shauli Ravfogel, Yoav Goldberg, Gal Chechik
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 SynGen syntax-guided attention loss instead of Attend-and-Excite | Stable Diffusion 1.4 · DVMP | Proper Binding: 52.47 -> 74.90 | Improved |
| use SynGen instead of Attend-and-Excite | Stable Diffusion 1.4 · A&E dataset | Improper Binding: 63.81 -> 23.81 | Improved |
| use SynGen instead of Attend-and-Excite | Stable Diffusion 1.4 · ABC-6K | concept separation, human majority vote (%): 11.17 -> 28.00 | Improved |
| use SynGen instead of Attend-and-Excite | Stable Diffusion 1.4 · DVMP | visual appeal, human majority vote (%): 12.17 -> 16.00 | Improved |
| use SynGen instead of Attend-and-Excite | Stable Diffusion 1.4 · DVMP | Entity Neglect: 10.77 -> 16.26 | Hurt |
| use SynGen instead of Attend-and-Excite | Stable Diffusion 1.4 · ABC-6K | Entity Neglect: 33.18 -> 34.41 | No clear effect |
| add negative loss term to positive-only loss (ablation) | SynGen · DVMP (100 prompts) | human preference among 4 variants | Improved |
| add positive loss term to negative-only loss (ablation) | SynGen · DVMP (100 prompts) | human preference among 4 variants | Improved |
| intervene for 25 of 50 steps instead of fewer steps (ablation) | SynGen · DVMP (100 prompts) | qualitative binding | Improved |
| intervene for all 50 steps instead of 25 (ablation) | SynGen · DVMP (100 prompts) | image blur rate | Hurt |
| raise scale factor from 1 to 20 (ablation) | SynGen · DVMP (100 prompts) | qualitative binding | Mixed |
Closest papers by meaning
- Magnet: We Never Know How Text-to-Image Diffusion Models Work, Until We Learn How Vision-Language Models Function (NeurIPS 2024 · similarity 0.89)
- Text Embedding is Not All You Need: Attention Control for Text-to-Image Semantic Alignment with Text Self-Attention Maps (CVPR 2025 · similarity 0.89)
- Localizing and Editing Knowledge In Text-to-Image Generative Models (ICLR 2024 · similarity 0.89)
- CONFORM: Contrast is All You Need for High-Fidelity Text-to-Image Diffusion Models (CVPR 2024 · similarity 0.88)
- Beyond Bag-of-Words: Diagnosing Compositional Binding Failures in Vision-Language Models (NeurIPS 2026 · similarity 0.88)
- The Hidden Language of Diffusion Models (ICLR 2024 · 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.
