The model is trained using the base model "ACertainty.ckpt." Through testing, it has been observed that to generate pixel art-style effects, the weight should be maintained above 0.85. However, when the weight exceeds 1.3, issues such as color overflow and distortion in the generated images become more pronounced.
The LYCORIS model is trained on pixel art-style images, and the images used in training do not exhibit significant hand features. This results in a notable decline in the quality of hand representations within the model's generated output.
该模型是使用“ACertainty.ckpt”为底模训练而成,经过测试,使用时权重保持在0.85以上才会生成像素风格的效果,权重如果超过1.3则生成出的图像溢色问题与人物扭曲问题会较为严重。
因为此LYCORIS模型训练时用到的图片为像素画风,训练使用到的图片自身手部特征并不显著,致使模型本身生成出的手部质量奇差。
Description
FAQ
Comments (8)
Is it for SD1.5 or SD2.0? The model says 2.0, but the images say anything-v5-PrtRE
Sorry,this is SD1.5,I wrote it wrong,but the model training does not use "anything-v5-PrtRE".Thank you for your correction.
@tatsu_hiro nice :D
好看诶 方便问下大大有在国内的平台上传嘛想去膜拜
没有
这边方便获得您的联系方式吗 想和您交流学习
@bonniezhang 首先感谢您对吾的肯定,但吾训练微调模型只是闲暇之余的无聊之举,吾对于训练模型只了解一些皮毛,专业性严重不足,国内也有很多介绍ai模型训练的高人,例如秋叶、青龙......他们更值得阁下去交流学习,吾说到底只是门外汉,无法给予阁下多少帮助,所以就不提供联系方式了,请见谅。
好哦








