Heliolisk LoRA trained on NovelAI with Kohya training scripts using the DAdaptAdam optimizer at 512x512 resolution. Trained on a low quality, inconsistent dataset, expect issues. Works pretty well when nude, but breaks down once you prompt for clothing or more complex scenes. Increasing the LoRA strength and prompting for the incorrect features will help with consistency, see my example pictures to get an idea of what that looks like. Should have wide compatibility with model mixes based on NovelAI, which is most of them. Trained on a mixed male/female dataset.