这次的训练集采用的是P站上mignon老师绘制的极高质量的插画,mignon老师绘制的人物面部细节，尤其是眼睛非常有特点,可惜的是作者技术能力比较有限，Lora仅能在portrait等怼脸距离才能较好的还原眼部细节,一旦镜头拉远就只能从轮廓上大致保留相似性了,

训练集采用了31张图,全身图13张，head out of fream 9张，face close-up 9张,在训练时除了总的关键词“mignon30”之外，全身图额外关键词"mignonf13",躯干图“mignonb9”,脸部特写“mignonh9”,

在正常绘图，尤其是进行面部绘画时可以同时添加“mignon30” & “mignonh9”,如果皮肤展露面积较大时，则可以添加“mignon30” & “mignonb9” & “mignonh9”,可以看情况和需求进行调整,虽然对画面影响不会非常显著，但多少会更趋向于原画风格,

推荐的权重是0.6,

这次依然是Z系列，也就是按照比较传统的画风提取方法进行训练,Z-000032和Z-000036在泛化表现上略有区别,作者一共拟定了4个大体人物形象和3种镜头，共12组xyz图，对Lora在纯随机种子条件下的出图表现进行测试,可以去Gallery中查看，并选择对应的版本进行下载,

模型本身以及模型生成的图片请勿用于任何商业或不正当用途！！

This training set is based on a very high-quality illustration from Pixiv by mignon, who has drawn the facial details of the characters, especially the eyes, which are very distinctive.

Unfortunately, the author's technical ability is relatively limited, Lora can only restore better eye details in portrait level. Once the camera is far away it can only roughly retain the basic similarity.

The training set used 31 images. 13 full body images, 9 head out of fream images, and 9 face close-up images. In addition to the general keyword "mignon30", additional keywords were used for the full-body images "mignonf13", body-only images "mignonb9", and face close-ups "mignonh9".

You can add both "mignon30" & "mignonh9" for normal generation process, especially for face drawing. For larger areas of skin, add "mignon30" & "mignonb9" & "mignonh9". Adjustments can be made depending on the situation and needs. Although the effect on the image will not be very significant, it will be more or less in line with the original style.

The recommended weight is 0.6.

This time, it is still the Z series, that is, according to the more traditional painting style extraction method for training. z-000032 and z-000036 are slightly different in generalization performance. The authors have formulated a total of 4 broad character outline and 3 kinds of shots, 12 sets of xyz drawings in total, to test the performance of Lora's outgoing drawings under purely random seed conditions. You can go to Gallery to view and select the corresponding version for download.

The model itself and the images generated by the model should not be used for any commercial or improper purposes!!

描述:

32 与 36 泛化表现略有差异 请查看32 页面的Gallery

训练词语: mignon30,mignonh9,mignonb9,mignonf13