Research map / Image generation / Text-to-image diffusion
Easier Painting Than Thinking: Can Text-to-Image Models Set the Stage, but Not Direct the Play?
Ouxiang Li, Yuan Wang, Xinting Hu, Huijuan Huang, Rui Chen, Jiarong Ou, Xin Tao, Pengfei Wan, Xiaojuan Qi, Fuli Feng
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 intermediate reasoning and prompt rewriting (Think mode) | BAGEL · T2I-CoReBench reasoning mean | reasoning mean: 34.1 -> 41.9 | Improved |
| add intermediate reasoning and prompt rewriting (Think mode) | BAGEL · T2I-CoReBench composition mean | composition mean: 46.4 -> 39.6 | Hurt |
| add o3 prompt rewriting (ablation) | FLUX.1-Krea-dev · T2I-CoReBench reasoning | reasoning mean: 44.3 -> 65.5 | Improved |
| add o3 prompt rewriting (ablation) | Qwen-Image · T2I-CoReBench reasoning | reasoning mean: 49.3 -> 72.7 | Improved |
| add o3 prompt rewriting (ablation) | Nano Banana · T2I-CoReBench reasoning | reasoning mean: 73.6 -> 78.2 | Improved |
| add o3 prompt rewriting (ablation) | GPT-Image · T2I-CoReBench reasoning | reasoning mean: 69.0 -> 79.9 | Improved |
| add o3 prompt rewriting (ablation) | Nano Banana · T2I-CoReBench GR and AR | GR / AR score: 83.2 (down 0.8), 81.4 (down 1.7) | Hurt |
| add o3 prompt rewriting (ablation) | GPT-Image · T2I-CoReBench PR | PR score: 87.1 (down 0.2) | No clear effect |
| add o3 prompt rewriting (ablation) | FLUX.1-Krea-dev · T2I-CoReBench LR | LR score: 30.3 -> 64.9 | Improved |
| switch from diffusion to autoregressive/unified paradigm | open-source T2I models · T2I-CoReBench | overall | Mixed |
| gap between reasoning and composition in best closed model | Nano Banana Pro · T2I-CoReBench | mean score: 89.7 -> 82.7 | Hurt |
| gap between reasoning and composition in best open model | Qwen-Image-2512 · T2I-CoReBench | mean score: 83.7 -> 51.7 | Hurt |
Closest papers by meaning
- OneIG-Bench: Omni-dimensional Nuanced Evaluation for Image Generation (NeurIPS 2025 · similarity 0.91)
- ConceptMix: A Compositional Image Generation Benchmark with Controllable Difficulty (NeurIPS 2024 · similarity 0.91)
- DynT2I-Eval: A Dynamic Evaluation Framework for Text-to-Image Models (NeurIPS 2026 · similarity 0.90)
- Davidsonian Scene Graph: Improving Reliability in Fine-grained Evaluation for Text-to-Image Generation (ICLR 2024 · similarity 0.90)
- Compositional Chain-of-Thought Prompting for Large Multimodal Models (CVPR 2024 · similarity 0.89)
- ConceptBed: Evaluating Concept Learning Abilities of Text-to-Image Diffusion Models (AAAI 2024 · similarity 0.89)
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.
