Research map / Image generation / Text-to-image diffusion
Prompt Reinjection: Alleviating Prompt Forgetting in Multimodal Diffusion Transformers
Yuxuan Yao, Yuxuan Chen, Hui Li, Kaihui Cheng, Qipeng Guo, Yuwei Sun, Zilong Dong, Jingdong Wang, Siyu Zhu
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 Prompt Reinjection | SD3.5-large · GenEval | overall score: 0.7179 -> 0.7644 | Improved |
| add Prompt Reinjection | FLUX.1-dev · GenEval | overall score: 0.6613 -> 0.6986 | Improved |
| add Prompt Reinjection | Qwen-Image · GenEval Position | score: 0.7550 -> 0.8025 | Improved |
| add Prompt Reinjection | SD3.5-large · DPG-Bench | overall score: 83.64 -> 86.99 | Improved |
| add Prompt Reinjection | Qwen-Image · DPG-Bench | overall score: 88.90 -> 89.15 | Improved |
| add Prompt Reinjection | FLUX.1-dev · PickScore (COCO-5K) | PickScore: 23.05 -> 22.97 | Hurt |
| add Prompt Reinjection | Qwen-Image · CLIP score (COCO-5K) | CLIP: 0.2715 -> 0.2704 | Hurt |
| add Prompt Reinjection | SD3-medium · ImageReward (COCO-5K) | ImageReward: 0.9514 -> 1.0825 | Improved |
| add Procrustes rotation alignment on top of distribution anchoring (ablation) | Prompt Reinjection on SD3-medium · GenEval | overall score: 0.6897 -> 0.7054 | Improved |
| add distribution anchoring on top of rotation (ablation) | Prompt Reinjection on FLUX.1-dev · GenEval | overall score: 0.6877 -> 0.7002 | Improved |
| add alignment (anchoring and rotation) to plain reinjection (ablation) | Prompt Reinjection on SD3-medium · GenEval | overall score: 0.6849 -> 0.7054 | Improved |
| inject into all deeper blocks instead of only layers 12-23 (ablation) | Prompt Reinjection on SD3-medium · GenEval | overall score: 0.6923 -> 0.7054 | Improved |
Closest papers by meaning
- Diff-Prompt: Diffusion-Driven Prompt Generator with Mask Supervision (ICLR 2025 · similarity 0.91)
- Rare Text Semantics Were Always There in Your Diffusion Transformer (NeurIPS 2025 · similarity 0.91)
- Prompting Hard or Hardly Prompting: Prompt Inversion for Text-to-Image Diffusion Models (CVPR 2024 · similarity 0.91)
- Prompt-Free Diffusion: Taking "Text" out of Text-to-Image Diffusion Models (CVPR 2024 · similarity 0.90)
- On Discrete Prompt Optimization for Diffusion Models (ICML 2024 · similarity 0.90)
- Textualize Visual Prompt for Image Editing via Diffusion Bridge (AAAI 2025 · 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.
