Zacian LoRA trained on NovelAI with Kohya training scripts using the DAdaptAdam optimizer at 768x768 resolution. The high amount of intricate detail in Zacian's design coupled with the vastly different ways Zacian was anthrofied in art made this a difficult model to train. Does a pretty good job keeping the crown consistent, but the wings will have a lot of variation due to the inconsistent ways they were depicted in the source artwork. Sometimes they're large, sometimes they're small, and occasionally you'll get a derpy mix of the two. Adding clothing will weaken the character detail, so you may need to add some reinforcement tags to make clothing work. Look at my example prompts to get an idea of what tags to add. Unable to do a proper rear view, the tail turns into a wing nearly every time. Despite the numerous issues, you can get some cool looking results! Should have wide compatibility with model mixes based on NovelAI, which is most of them. Trained on a mixed male/female dataset.