Research map / Image generation / Text-to-image diffusion
EruDiff: Refactoring Knowledge in Diffusion Models for Advanced Text-to-Image Synthesis
Xiefan Guo, Xinzhu Ma, Haoxiang Ma, Zihao Zhou, Di Huang
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 EruDiff (DK-DM + NO-RL) to FLUX.1-dev | FLUX.1-dev · WISE | WiScore overall: 0.50 -> 0.64 | Improved |
| train EruDiff directly on WISE prompts instead of Knowledge-10K | FLUX.1-dev + EruDiff · WISE | WiScore overall: 0.64 -> 0.73 | Improved |
| add EruDiff to FLUX.1-dev | FLUX.1-dev · Science-T2I S | SciScore: 23.31 -> 33.70 | Improved |
| add EruDiff to Qwen-Image | Qwen-Image · Science-T2I S | SciScore: 25.38 -> 35.43 | Improved |
| add EruDiff to Qwen-Image | Qwen-Image · Science-T2I C | SciScore: 31.56 -> 38.48 | Improved |
| replace GPT-4o prompt rewriting with EruDiff | FLUX.1-dev · Science-T2I S | SciScore: 32.90 -> 33.70 | Improved |
| add NO-RL to EruDiff (ablation) | EruDiff on FLUX.1-dev · Science-T2I S | SciScore: 31.99 -> 33.70 | Improved |
| add NO-RL to EruDiff (ablation) | EruDiff on FLUX.1-dev · Science-T2I C | SciScore: 32.82 -> 34.54 | Improved |
| add positive-sample learning to NO-RL (ablation) | EruDiff on FLUX.1-dev · Science-T2I S | SciScore: 33.70 -> 33.83 | No clear effect |
| add positive-sample learning to NO-RL (ablation) | EruDiff on FLUX.1-dev · Science-T2I C | SciScore: 34.54 -> 34.19 | No clear effect |
| use SciScore reward instead of CLIPScore for NO-RL filtering (ablation) | EruDiff on FLUX.1-dev · Science-T2I S | SciScore: 32.93 -> 33.70 | Improved |
| add EruDiff instead of GRPO | FLUX.1-dev · Science-T2I S | SciScore: 31.68 -> 33.70 | Improved |
Closest papers by meaning
- Prompt-Free Diffusion: Taking "Text" out of Text-to-Image Diffusion Models (CVPR 2024 · similarity 0.90)
- Prompting Hard or Hardly Prompting: Prompt Inversion for Text-to-Image Diffusion Models (CVPR 2024 · similarity 0.90)
- I Think, Therefore I Diffuse: Enabling Multimodal In-Context Reasoning in Diffusion Models (ICML 2025 · similarity 0.90)
- DreamDistribution: Learning Prompt Distribution for Diverse In-distribution Generation (ICLR 2025 · similarity 0.90)
- On Discrete Prompt Optimization for Diffusion Models (ICML 2024 · similarity 0.90)
- Localizing and Editing Knowledge In Text-to-Image Generative Models (ICLR 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.
