-v2模型简介： 1-增加了更多的图片进行训练，生成全身的图片质量有一定的改善，但还存在一些问题； 2-减少了图片的光污染，在机甲金属感上做了一定的优化； 3-建议尺寸是576x1024，建议打开高分辨率修复，408x728或576x1024 ， 打开高分辨率修复后质量和细节有一定提升。 生成建议： 1-反向词可以试试：EasyNegative,bad-artist,bad_prompt_version2,signature； 2-提高采样算法：首先默认的采样次数调到40； 3-再把高分辨率修复采样次数调到20； 那就基本可以得到比较满意的质量效果了
Description
----------------------------------------------------------------------
-v3 model introduction: Added more pictures for training, the number of training steps increased to 15120;
2- Remove all transparent mecha, if the graphics card is good enough and the setting is high enough, the resulting graphics are really handsome;
3- The recommended size is 576x1024, it is recommended to open high resolution repair, 408x728 or 576x1024 (toast does not support that large, local deployment can try 576x1024)
----------------------------------------------------------------------