The dataset consisting of 500 images for the conversion from anime to realistic style was trained using Sam3.1 and Flux2Klein. Various styles of materials were utilized for conversion to maximize its versatility.

After testing, it was unexpectedly found that this LORA could optimize the performance of nipples, fingers, and toes.