Nagatoro Hayase (TI)

详情

模型描述

Trained on ~90 images from the show at rate of .0005. SD struggles with high CFG and steps (see distribution image). I prefer 25 steps at a scale of 4 at clip skip 2 for Anythingv3.

Her earclip appears randomly if not specified, but the hair pins almost always work (sometimes on the left side). The one-piece tanline is there, but usually has problems.

Edit: Added training data

此模型生成的图像