I can confirm it's strongly biased to western women. I tested it out, the lora works great but even when using prompts with very strong wording to indicate east Asian or a specific type like Japanese, Chinese, Vietnamese etc. most of the time it generates a western woman. And sometimes when i get Asian woman it looks mixed, half Asian half European. If faces aren't in the training data then i assume that this AI model is making assumption based on the body type in the training data and skin color that it's likely western woman. I did get a few results of a truly Asian woman but it's difficult with this Lora. Which is interesting because base model of z-image favors Asian over Western women. Perhaps if OP can, make sure to train some Asians in training data even if no face shown, perhaps that would help? overall i do like this Lora.

i tried a specific Asian name i know for sure gives 100% Asian result every time. i tested 5 different prompts. only 1 out of 5 actually looked anything like her and Asian. i can for sure say this Lora has western women bias

@ann_willnn I suddenly feel so happy. It used to be so difficult to generate a graph of this quality using open-source models, but now with Z, it only takes a few seconds.