To assess text rendering performance in text-to-image generation, we introduce LeX-Bench, a benchmark comprising 1,310 carefully designed prompts.
Each prompt contains two parts: an Image Caption describing the image content, and a Text Caption specifying the text to be rendered. The combined format is:
{Image Caption}, with the text on it: {Text Caption}., e.g., A picture of a blue and green abstract people logo on a purple background, with the text on it: "AREA", "PEOPLE".

This dataset is introduced in the paper LeX-Art: Rethinking Text Generation via Scalable High-Quality Data Synthesis.

Project page: https://zhaoshitian.github.io/lexart/ Code: https://github.com/zhaoshitian/LeX-Art

Fig. 1: Overview of LeX-Bench.

@article{zhao2025lexart,
    title={LeX-Art: Rethinking Text Generation via Scalable High-Quality Data Synthesis},
    author={Zhao, Shitian and Wu, Qilong and Li, Xinyue and Zhang, Bo and Li, Ming and Qin, Qi and Liu, Dongyang and Zhang, Kaipeng and Li, Hongsheng and Qiao, Yu and Gao, Peng and Fu, Bin and Li, Zhen},
    journal={arXiv preprint arXiv:2503.21749},
    year={2025}
}