A bossman LoRA trained on SDXL 1.0. (uses same dataset as https://civitaiarchive.com/models/181242, similar parameters work well for it). The original LoRA tends to give better results and is easier to use, but the SDXL LoRA (as expected) works with SDXL models.