Research map / Image generation / Text-to-image diffusion
PEA-Diffusion: Parameter-Efficient Adapter with Knowledge Distillation in non-English Text-to-Image Generation
jian ma, Chen Chen, Qingsong Xie, Haonan Lu
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 |
|---|---|---|---|
| train a 6M-parameter MLP adapter with UNet distillation instead of translation | PEA-Diffusion (SDXL) · MC cultural prompts, user study | success rate: 0.2577 -> 0.4978 | Improved |
| use PEA-Diffusion instead of translation on general prompts | PEA-Diffusion (SDXL) · MG general prompts | ImageReward / PickScore | Hurt |
| use PEA-Diffusion instead of translation for Japanese | PEA-Diffusion · MC cultural prompts, Japanese | CLIPScore | Hurt |
| add KD of any kind versus plain fine-tuning of the adapter (ablation) | PEA-Diffusion · Chinese MG/MC | ImageReward, PickScore, CLIPScore | Improved |
| add UNet intermediate feature distillation (ablation) | PEA-Diffusion · Chinese MG/MC | ImageReward, CLIPScore | Improved |
| add UNet logit distillation only (ablation) | PEA-Diffusion · Chinese MG/MC | ImageReward, CLIPScore | No clear effect |
| distill text encoder features only (TKD) (ablation) | PEA-Diffusion · Chinese MG/MC | ImageReward, CLIPScore | Hurt |
| add text encoder KD on top of UNet KD (ablation) | PEA-Diffusion · Chinese MG/MC | ImageReward, CLIPScore | Hurt |
| replace adapter training with LoRA or W_K/W_V fine-tuning | SDXL language transfer · MG and MC | ImageReward, PickScore, CLIPScore | Hurt |
| use PEA-Diffusion instead of GlueGen | PEA-Diffusion · MG, five languages | ImageReward: GlueGen -0.5 ImageReward | Improved |
| train PEA on low-resolution (around 640) data (ablation) | PEA-Diffusion (SDXL, Chinese) · MG | ImageReward | Hurt |
| increase adapter parameters at equal training samples (ablation) | PEA · MG (Chinese) | ImageReward | Improved |
Later papers that compared with this method
1 later accepted paper reports PEA-Diffusion as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.
- Where Culture Fades: Revealing the Cultural Gap in Text-to-Image Generation (CVPR 2026 · better on 4 of 4 shared results)
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- MuLan: Adapting Multilingual Diffusion Models for Hundreds of Languages with Negligible Cost (ICML 2025 · similarity 0.91)
- Exploring the Deep Fusion of Large Language Models and Diffusion Transformers for Text-to-Image Synthesis (CVPR 2025 · similarity 0.90)
- AltDiffusion: A Multilingual Text-to-Image Diffusion Model (AAAI 2024 · similarity 0.89)
- LLM4GEN: Leveraging Semantic Representation of LLMs for Text-to-Image Generation (AAAI 2025 · similarity 0.89)
- Ranni: Taming Text-to-Image Diffusion for Accurate Instruction Following (CVPR 2024 · similarity 0.88)
- Mastering Text-to-Image Diffusion: Recaptioning, Planning, and Generating with Multimodal LLMs (ICML 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.
