Research map / Image generation / Text-to-image diffusion
Bridging Different Language Models and Generative Vision Models for Text-to-Image Generation
Shihao Zhao, Shaozhe Hao, Bojia Zi, Huaizhe Xu, Kwan-Yee K. Wong
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 |
|---|---|---|---|
| replace CLIP text encoder with Llama-2 | LaVi-Bridge with SD V1.4 U-Net · T2I-CompBench Color | score: 0.3578 -> 0.4859 | Improved |
| use larger T5 (Small to Base to Large) | LaVi-Bridge with SD V1.4 U-Net · COCO2014 short prompts | CLIP score: 0.3122 -> 0.3149 -> 0.3156 | Improved |
| replace CLIP with Llama-2 for long prompts | LaVi-Bridge with SD V1.4 U-Net · Long prompts | CLIP score: 0.3120 -> 0.3231 | Improved |
| replace SD U-Net with PixArt transformer | LaVi-Bridge with T5-Large · COCO2014 short prompts | aesthetic score: 5.881 -> 6.145 | Improved |
| replace LDM U-Net with SD V1.4 U-Net | LaVi-Bridge with T5-Large · COCO2014 short prompts | FID: 25.94 -> 23.11 | Improved |
| replace SD U-Net with PixArt transformer | LaVi-Bridge with T5-Large · T2I-CompBench Color | score: 0.3889 -> 0.3689 | Hurt |
| apply LaVi-Bridge training to the same CLIP and U-Net weights as SD V1.4 | Stable Diffusion V1.4 · COCO2014 short prompts | FID: 20.32 -> 23.57 | Hurt |
| apply LaVi-Bridge training to the same CLIP and U-Net weights as SD V1.4 | Stable Diffusion V1.4 · T2I-CompBench | compositional scores: Texture 0.4028 -> 0.4506 | Mixed |
| add LoRA to language and vision models (ablation) | LaVi-Bridge T5-Large + SD U-Net · T2I-CompBench Color | score: 0.2485 -> 0.3889 | Improved |
| use stacked feedforward adapter instead of a single linear layer (ablation) | LaVi-Bridge T5-Large + SD U-Net · T2I-CompBench Shape | score: 0.3044 -> 0.3552 | Improved |
| add LoRA (ablation) | LaVi-Bridge T5-Large + SD U-Net · COCO2014 short prompts | FID: 22.35 -> 23.11 | Hurt |
Closest papers by meaning
- Elucidating the design space of language models for image generation (ICML 2025 · similarity 0.91)
- Unified Language-Vision Pretraining in LLM with Dynamic Discrete Visual Tokenization (ICLR 2024 · similarity 0.91)
- MuLan: Adapting Multilingual Diffusion Models for Hundreds of Languages with Negligible Cost (ICML 2025 · similarity 0.90)
- LMFusion: Adapting Pretrained Language Models for Multimodal Generation (NeurIPS 2025 · similarity 0.90)
- LLM4GEN: Leveraging Semantic Representation of LLMs for Text-to-Image Generation (AAAI 2025 · similarity 0.90)
- LLM Blueprint: Enabling Text-to-Image Generation with Complex and Detailed Prompts (ICLR 2024 · similarity 0.90)
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.
