Trained on the animefull-final model for 12 epochs at a network dim of 16, then resized to one of 8.Most of the images in the dataset were character portraits taken from <a rel="ugc" href="https://store.steampowered.com/app/345820/Shantae_and_the_Pirates_Curse/">the game</a> with a white background, backgrounds still generate fine but they seem to weaken the style a bit.The shading style seems to be best preserved with the Anything V3 model, as it keeps the flat coloring the best, the Based64 and 7th Anime models are not as accurate but still give very nice looking results.The LoRA is trained to work at a weight of 1 but you're free to adjust it depending on what model you're using it on and what results you want.There's no activation tag, just load the LoRA in and it should work right away.

Files with SHA256: 6615d8a2ee98daf5877dd6054269362e890eaff43486039bb7afb4bf1122ac2f