Research map / Image generation / Text-to-image diffusion
Text Embedding Knows How to Quantize Text-Guided Diffusion Models
Hongjae Lee, Myungjun Son, Dongjea Kang, Seung-Won Jung
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 QLIP to Q-diffusion on BK-SDM | Q-diffusion W4A16 · COCO2017 | FAB: 16.00 -> 12.14 | Improved |
| add QLIP to Q-diffusion on BK-SDM | Q-diffusion W4A16 · COCO2017 | FID: 30.02 -> 30.01 | No clear effect |
| add QLIP to PTQD on BK-SDM | PTQD W4A16 · COCO2017 | sFID: 77.18 -> 73.26 | Improved |
| add QLIP to Q-diffusion on Stable Diffusion with lower-bit options | Q-diffusion W4A8 · COCO2017 | FID: 23.40 -> 21.61 | Improved |
| add QLIP to PTQD on Stable Diffusion | PTQD W4A8 · Conceptual Captions | FID: 25.40 -> 20.42 | Improved |
| add QLIP to Q-diffusion on SDXL | Q-diffusion W4A16 · Conceptual Captions | FAB: 16.00 -> 11.34 | Improved |
| add QLIP to PCR on FLUX | PCR · COCO2017 (500 prompts) | ImageReward: 0.9986 -> 1.0214 | Improved |
| add QLIP to Q-diffusion for runtime | Q-diffusion W4A16 · Conceptual Captions | Runtime (s): 5.58 -> 4.85 | Improved |
| use GIQA instead of Realism score as T2Q quality measure (ablation) | QLIP · COCO2017 | FID: 22.18 -> 21.61 | Improved |
| use GIQA instead of CLIP-IQA as T2Q quality measure (ablation) | QLIP · COCO2017 | FID: 21.81 -> 21.61 | Improved |
| add time-step medium and high probabilities p_m and p_h to p_q (ablation) | QLIP Q2B · COCO2017 | FID: 26.91 -> 21.61 | Improved |
| add p_m^t to p_q + p_h^t (ablation) | QLIP Q2B · COCO2017 | FID: 29.37 -> 21.61 | Improved |
Closest papers by meaning
- On Discrete Prompt Optimization for Diffusion Models (ICML 2024 · similarity 0.89)
- AP-Adapter: Improving Generalization of Automatic Prompts on Unseen Text-to-Image Diffusion Models (NeurIPS 2024 · similarity 0.88)
- Get What You Want, Not What You Don't: Image Content Suppression for Text-to-Image Diffusion Models (ICLR 2024 · similarity 0.88)
- Würstchen: An Efficient Architecture for Large-Scale Text-to-Image Diffusion Models (ICLR 2024 · similarity 0.88)
- DGQ: Distribution-Aware Group Quantization for Text-to-Image Diffusion Models (ICLR 2025 · similarity 0.87)
- Prompting Hard or Hardly Prompting: Prompt Inversion for Text-to-Image Diffusion Models (CVPR 2024 · similarity 0.87)
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.
