Trained from art by Mitsuba Choco, a visual novel kind of style
Trained on AbyssOrangeMix2_NSFW
For whatever reason it can generate grainy looking results, using "<hypernet:Mitsuba_Choco:0.9>, (masterpiece, best quality:1.2), extremely detailed, jewel like eyes" As your positive prompts and "(bad_prompt_version2:0.6), (worst quality, low quality:1.0)" as your negative makes the images a lot clearer. Alternatively, upscaling the images in extras helps a lot too.