Minority-Focused Text-to-Image Generation via Prompt Optimization (CVPR 2025 Oral)

Soobin Um and Jong Chul Ye

This repository contains the code for the paper "Minority-Focused Text-to-Image Generation via Prompt Optimization" (CVPR 2025 Oral).

Setup

First, create your environment. We recommand to use the following comments.

git clone https://github.com/anonymous5293/MinorityPrompt
cd MinorityPrompt
conda env create -f environment.yaml

Second, download sdxl_lightning_4step_unet.safetensors in ckpt.

Examples

• T2I generation

source scripts/text_to_img.sh

• MS-COCO

source scripts/text_to_mscoco.sh

Feel free to modify the scripts to fit your needs.

Citation

If you find this repository useful, please cite our paper:

@article{um2024minorityprompt,
  title={MinorityPrompt: Text to Minority Image Generation via Prompt Optimization},
  author={Um, Soobin and Ye, Jong Chul},
  journal={arXiv preprint arXiv:2410.07838},
  year={2024}
}