Use with weights 0.8 to 1.

Keywords are kirishimaakari and kirishimakotone. Most images trained with serafuku.

Note that this LoRA was trained at 768x768, so generate images with at least that resolution for best results.

Issues

• Both characters are in this LoRA, so their hair ornaments and colors can get mixed up. Honestly, though they're basically the same character visually anyway.

• None of the dataset had both characters in the same image, so 2girls prompts are more difficult to generate.