
请勿将本模型及生成的图片用于任何商业用途!
请勿利用本模型生成及传播不恰当的图片!!
这次采用了77张脸部的图片整合成了数据集,水叔的题材和画风都是有极强的特色,作为一个Lora的制作者,在此声明一下,不要使用本模型生成的图片进行不恰当的传播!谢谢配合,
【海报中包含了多组组内相似的图片,每组的第一张是未使用ADetailer的,剩余组内图片为使用ADetailer的,】
这次会上传两个不同Epoch的版本:
ASYLJ77-2-000020_MainlyForFace主要用来搭配ADetailer使用,参数0.4-0.5,重绘幅度0.5- 0.7(有时也可以考虑参数0.7左右,重绘0.4-0.5),因为训练轮数更多,拟合程度更高,在只针对脸部进行重绘时效果更好,值得注意的时,因为训练集有一些图片包含毛绒绒的领子,为了避免干扰,在重绘时最好让蒙版在Y轴方向上移一些,同时,在Adetailer的Positive中加入(bare_neck), 在Negative中加入choker,
ASYLJ77-2-000014_MainlyForBody则可以用来绘制全身图和半身图,虽然只投入了脸部的图片,但是在权重较低的情况下,绘制全身和半身图依然时可行的,参数同样推荐为0.4左右,高分辨率修复 (Hires. fix)的重绘推荐在0.4左右,
需要注意的是,这次的两个版本是可以交换使用的,020可以用来画身子,014同样也可以用来画脸,并没有非常非常明确的区分,同时,参数方面也可以根据自己的需求调整,作者在绘制展示海报时也是根据需要进行了多种参数的尝试,
两者的触发词是一致的,均为ASYLJ77,
Please do not use this model and generated images for any commercial purposes!
Please do not use this model to generate and distribute inappropriate images!!
This time, 77 images of faces were used to integrate the dataset. The style of the pictures of As109 are extremely distinctive. As a Lora maker, I would like to make a statement that Do Not use the images generated by this model for inappropriate distribution! Thanks for the cooperation.
[The poster contains multiple groups of similar images, the first in each group is the one that does not use ADetailer. The remaining in-group images are those using ADetailer.]
Two different Epoch versions will be uploaded this time:
ASYLJ77-2-000020_MainlyForFace is mainly used with ADetailer. Parameter 0.4-0.5, strength 0.5- 0.7 (sometimes you can also consider parameter around 0.7, strength 0.4-0.5). Because of the higher number of training rounds, the fit of this version might be higher and works better when redrawing only the face. It is worth noting that since some of the images in the training set contain fluffy collars, it is better to have the mask moved up a bit in the Y-direction when redrawing in order to avoid interference. At the same time, add (bare_neck) to Adetailer's Positive and add choker to Negative might also work.
ASYLJ77-2-000014_MainlyForBody can be used to draw full and half body images. Although only the face image is input, it is still feasible to draw full and half body images with low weights. Again, a parameter of about 0.4 is recommended. For high resolution fixes (Hires. fix) a strength around 0.4 might be enough.
Note that the two versions are interchangeable in different situations. 020 can be used to paint the body, and 014 can likewise be used to paint the face. There is no very, very clear distinction. At the same time, the parameters can be adjusted according to your own needs, and the author has tried a variety of parameters as needed when drawing the display poster.
The trigger word for both is the same, which is ASYLJ77.
描述:
-
Note: XL版本的范例图中表情较为平淡的为未使用ADetailer的图片。一般为每组第三张。 (The XL version of the poster images with the more bland expressions are those that do not use ADetailer. Usually the third one in each group.)
-
XL版本依然推荐主要作为ADetailer修复面部的脸模使用。
-
The XL version is still recommended to be used primarily as a face mold for ADetailer repair faces.
-
如果直接用来生成完整图片,推荐权重0.9(不开高清修复情况)。如果进行脸部修复时,针对较小脸部区域,推荐权重0.95,重绘幅度0.6-0.7。如果脸部区域较大则推荐重绘幅度不高于0.6。具体情况请根据实际需要决定。
-
If used directly to generate a complete picture, the recommended weight is 0.9 (without Hires. fix). For face restoration, the recommended weight is 0.95 and the recommended strength is 0.6-0.7 for small face areas, or no higher than 0.6 for larger face areas, depending on the actual needs.
Note: XL版本的范例图中表情较为平淡的为未使用ADetailer的图片。一般为每组第三张。 (The XL version of the poster images with the more bland expressions are those that do not use ADetailer. Usually the third one in each group.)
XL版本依然推荐主要作为ADetailer修复面部的脸模使用。
The XL version is still recommended to be used primarily as a face mold for ADetailer repair faces.
如果直接用来生成完整图片,推荐权重0.9(不开高清修复情况)。如果进行脸部修复时,针对较小脸部区域,推荐权重0.95,重绘幅度0.6-0.7。如果脸部区域较大则推荐重绘幅度不高于0.6。具体情况请根据实际需要决定。
If used directly to generate a complete picture, the recommended weight is 0.9 (without Hires. fix). For face restoration, the recommended weight is 0.95 and the recommended strength is 0.6-0.7 for small face areas, or no higher than 0.6 for larger face areas, depending on the actual needs.
[1] XL版支持瞳孔和眼睛颜色分别调色。基本支持控制多色瞳孔颜色。基本支持异色瞳。
[1] The XL version supports separate color for pupils and eyes. To some extent support for controlling multi-colored eyes colors. To some extent support for heterochromatic pupils.
[2] XL版支持更详细的表情控制 (XL version supports more detailed expression control):
(1) 较为有效的负面表情词 (Relatively Effective Negative Emotion words):
expressionless, contempt, disdain, annoy, angry, tears, tearing up, shouting, scared, endured face, despair, pain, sad, wince
(2) 较为有效的中性表情词 (Relatively Effective Neutral Emotion words):
clenched teeth, fang, fangs, thinking, naughty face, confused, thinking, evil smile, :t, :o,
(3) 较为有效的中性表情词 (Relatively Effective Positive Emotion words):
+_+, :d, smile, laughing, smirk, and so on
训练词语: ASYLJ77
名称: ASYLJ77-XL-MainlyForFace.safetensors
大小 (KB): 232350
类型: Model
Pickle 扫描结果: Success
Pickle 扫描信息: No Pickle imports
病毒扫描结果: Success