If you are interested in LLMs, you can look into this paper from ByteDance (also from CVPR 22) - https://openaccess.thecvf.com/content/CVPR2022/papers/Liu_XM...
The second paper seems to be more specific to Chinese fonts.
"Our goal is to generate fonts with specific impressions, by training a generative adversarial network with a font dataset with impression labels". The PDF contains interesting examples of Latin-based fonts generated by suggesting different impressions.