This is a face LoRA for Z-Image Base. It doesn't reproduce specific faces, but rather applies subtle adjustments. It might feel a little different if you try to use it as a general face LoRA. It's closer to a doll-likeness type, and could be useful when you want to make an Asian face appear on a Western-style model.

It's trained on a Z-Image Base model, so it can be used with general ZiB models.

Although the faces are similar, each model was trained using images from different categories. It's intended to be mixed by specifying the ratio. The basic usage is to use two for positive and one for negative to highlight features. You don't need to use all three simultaneously, and using all positive is also fine. It's difficult to describe the effect of each LoRA, so please experiment in practice.