Generates Mitsuka Yoshimine from the DearS anime in 2 outfits. Heads up. I used 44 screencaps to generate this image, and due to the fact that this is a secondary character, and her personality as an exhibitionist, she is rarely alone. A good portion of the images used have the backs of students heads in frame, and that has had an effect on this lora. I used "deserted" and "alone" prompts to combat this effect, however if you use the word classroom, it will generate other people with their backs to the viewer at a very high rate, despite weighted prompts to the contrary.Other prompts that help are "vibrant" and "hyper realism", since the screencaps can appear washed out, and a lot of detail isn't used in the full body shots from the image dataset. I used the vintage anime lora I linked in the suggested resources, as well as the disney princess lora, as style augmentations in my own image generation, and was pleased with the results. The animes' style is a bit dated, and for me at least, became a bit unattractive.I've also found that the negative prompts "blurry" and "bad hands", weighted at 1.2 or 1.3, help a lot to combat the bad screenshot look, but perhaps that's common knowledge at this point.Lastly, I can't recommend the adetailer enough for this, I found it really making the faces much better than what the standard image generation was doing. ExpressiveH and StSdetailer help with that exponentially.

Files with SHA256: 05f5a3049c01edf7fc9adcf61ffb680b97cc71bf42556f43361b7901d9cd1e49