<a target="_blank" rel="ugc" href="https://civarchive.com/models/1585859/xiaohuoguo">https://civarchive.com/models/1585859/xiaohuoguo</a>从该lora蒸馏出来的手绘风格,具有特殊的彩铅手绘笔触。经试验个人认为有相当不错泛化性，在训练素材为纯上半身肖像下训练的情况下能出中远景的图片，且于三种风格lora在1权重叠加时图像线条仍顺畅（但此时已经出现相当程度的性能下降）The hand-drawn style distilled from this Lora features special colored pencil hand-drawn strokes. Through experiments, I personally believe it has quite good generalization ability. When trained with training materials that are purely upper-body portraits, it can generate mid-to-long shot images. Moreover, it can still generate images normally when overlaid with three style Loras at a weight of 1 (but at this point, there has been a considerable degree of performance degradation).

铅笔绘画风格Pencil style

这轮训练在ep1时就有初步笔触了，拟合速度远超zimage等模型。重新测试，ep10在1.0权重时单独拟合度略差，但似乎和artist风格有更好的融合。（dino，黑星红白，鸣门卷等）This round of training showed initial results when ep1, with a fitting speed far exceeding that of models like zimage。etc. After re-testing, at ep10 with a weight of 1.0, the individual fitting degree is slightly poor, but it seems to have a better integration with the artist's style.（dino，kuroboshi，naruto05）

Files with SHA256: c67d646573ce91280bec3384d21ff53c371c1467cf64f07f97ff64e7d5246745