Research map / Image generation / Text-to-image diffusion
Text-Anchored Score Composition: Tackling Condition Misalignment in Text-to-Image Diffusion Models
Luozhou Wang, Guibao Shen, Wenhang Ge, Guangyong Chen, Yijun Li, Yingcong Chen
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 |
|---|---|---|---|
| add attention realignment to score composition (ablation) | TASC with ControlNets · ControlNets ablation | image-condition distance: 9.59% -> 1.58% | Improved |
| add attention realignment to score composition (ablation) | TASC with GLIGEN · GLIGEN ablation | image-condition distance: 6.23% -> 4.34% | Improved |
| add attention realignment to score composition (ablation) | TASC with GLIGEN · GLIGEN ablation | image-text similarity: 74.81% -> 79.87% | Improved |
| add attention realignment to score composition (ablation) | TASC with ControlNets · ControlNets ablation | image-text similarity: 73.78% -> 73.88% | No clear effect |
| text-anchored score composition without realignment vs plain SD (ablation) | Stable Diffusion with ControlNets · ControlNets ablation | image-text similarity: 69.49% -> 73.78% | Improved |
| TASC vs Attend-and-Excite on CLIP score | Stable Diffusion · Misaligned-condition prompts | CLIP Score: 0.3253 -> 0.3323 | Improved |
| TASC vs Attend-and-Excite on FID | Stable Diffusion · MSCOCO | FID: 111.41 -> 107.39 | Improved |
| TASC vs Attend-and-Excite on controllability | ControlNets · Controllability with ControlNets | relative image-condition distance: 14.1% -> 1.6% | Improved |
| StructureDiffusion instead of plain SD | Stable Diffusion · Misaligned-condition prompts | CLIP Score: 0.2934 -> 0.2928 | No clear effect |
| TASC added compute versus plain SD | Stable Diffusion · Inference cost | Time (seconds): 9.70 -> 13.88 | Hurt |
Closest papers by meaning
- Improving Long-Text Alignment for Text-to-Image Diffusion Models (ICLR 2025 · similarity 0.89)
- Aligning Text to Image in Diffusion Models is Easier Than You Think (NeurIPS 2025 · similarity 0.88)
- CtrLoRA: An Extensible and Efficient Framework for Controllable Image Generation (ICLR 2025 · similarity 0.88)
- Information Theoretic Text-to-Image Alignment (ICLR 2025 · similarity 0.88)
- CoMat: Aligning Text-to-Image Diffusion Model with Image-to-Text Concept Matching (NeurIPS 2024 · similarity 0.88)
- Ctrl-X: Controlling Structure and Appearance for Text-To-Image Generation Without Guidance (NeurIPS 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.
