VRChat Karin

Based on the training with animagineXLV3_v30, generate a 3D model image of Karin for sale on the booth.

Recommended weight: It is suggested to use 0.9. In the XYZ graph, a weight of 0.9 results in fewer visual errors and finer lines. Since the dataset mostly consists of default clothing, using 0.9 helps alleviate some issues related to insufficient generalization.

The dataset is trained with a resolution of 1152x1152, comprising 63 images.

Trigger phrase: vrchat karin

The labeling method involves removing the character's distinctive features.

Update：The Simplify version reduces the file size to 28MB, resulting in a slight decrease in generated quality, but potentially reducing the probability of generating incorrect images.Recommended weight: 1.

==========================================================

基于animagineXLV3_v30训练，生成Booth上售卖的3D模型Karin图像。

推荐权重：推荐0.9,在xyz图可以看到权重为0.9时画面错误更少,线条更精细。因为数据集几乎全部都是默认衣服，所以使用0.9可以缓解一点泛化不够的问题。

数据集以1152*1152分辨率进行训练，有63张。

触发词：vrchat karin

打标方式是删除人物形象特征。

更新：v1.0_simplify版本缩减文件大小到28mb，生成质量有些许下降，但生成错误图片的概率可能更小。推荐权重：1。