Let’s be honest — do we need Style LoRAs for Z-Image? No.
Does it make sense to train on a distilled model? No.
Did I try it anyway? Yes.
All LoRAs reduce the default Asian look.
Turbo is, in my opinion, the LoRA with the best overall output. Natural look, lots of detail.
High-Turbo might be slightly overtrained, but it can sometimes deliver even better natural results.
Beauty-Turbo really shines with the DPM++ 2S a RF sampler and produces images that can look smoother. Recreates professional photography or Photoshop look.
Also to mention is that it is optimized for strength 1, but lighter weights in the 0.5–1.0 range can also be used if needed.
Additionally, I only tested everything with Forge Neo, so I can’t evaluate or guarantee similar results for ComfyUI or other setups.