How to train chinese handwritten model?

[yanziwu621]

Is this project suitable to train chinese handwritten model with my own data without changing the model? If I can add your QQ or Wechat ?

[Metavers1]

中文的和手写体公式的我都试了 不太行，限于扩散模型本身的限制，所以这也可能是他们一直没放出权重和代码的原因，

[dailenson]

Hi, thanks for your attention. The Chinese checkpoints are related to our future business plans, so we cannot release them now. However, you can easily modify the open-sourced English version to implement the Chinese version. If you encounter any issues during this process, I am more than happy to assist you.

[aigc2025]

@dailenson 您好，我想请教一个问题，就是在训练中文数据集的时候，

1. unifont.pickle是否可以直接使用？还是需要换成其他的？
2. 换成中文数据集的时候，letters应该换成什么呢？
   letters = '_Only thewigsofrcvdampbkuq.A-210xT5\'MDL,RYHJ"ISPWENj&BC93VGFKz();#:!7U64Q8?+*ZX/%'

祝好，谢谢

---

补充：对于第2点，是需要对我的中文数据集的label做一下预处理吗？提取出所有出现的汉字，然后做成一个letters吗？