vigee_MIX

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

2024.02.24

这是个与众不同的版本，是我用另类的方法融合出来的。同时修改了模型本身的一些内置参数，比如对比度、亮度、细节，同时微调了色彩（并非内置lora的方法）。值得尝试一下，尤其是对红色和蓝色的敏感程度较高。在绝大多数情况下，画面的表现力会高于我之前发布的版本。当然了，这个色调不可能是所有大众都喜欢的。欢迎提出宝贵意见。

画面整体的色彩表现，我觉得在1.5版本的模型里已经很难再进一步提升了。毕竟才2G的容量，想让它做到万能的满意是不可能的。

想要更精细调整对比度、亮度、锐化、噪点、色彩、色温等参数的话，建议下载到本地，使用 https://github.com/portu-sim/sd-webui-bmab 这个sdwebui的插件进行微调。这种方法需要每次绘图时都打开插件并设置好参数。

DPM++2M Karras     steps： 38     512X1024   cfg：6

This is a unique version that I have fused using alternative methods. At the same time, some built-in parameters of the model itself were modified, such as contrast, brightness, and details, while fine-tuning the color (not using the built-in Lora method). It's worth trying, especially with high sensitivity to red and blue. In the vast majority of cases, the expressive power of the screen will be higher than the version I previously released. Of course, this color tone cannot be favored by all the public. Welcome to provide valuable feedback.

I think it is difficult to further improve the overall color representation of the picture in the 1.5 version model. After all, with only 2GB of capacity, it is impossible for it to achieve universal satisfaction.

If you want to adjust parameters such as contrast, brightness, sharpening, noise, color, and color temperature more finely, it is recommended to download and use https://github.com/portu-sim/sd-webui-bmab

Fine tune the plugin of this sdwebui. This method requires opening the plugin and setting the parameters every time you draw.

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

2024.02.20

ver 1.2,

新的分层融合方法,测试下效果吧.这个版本和我在国内网站发布的不完全相同.

更多的例图参考(超过1000张)

https://www.liblib.art/userpage/1e3c4c0fc33545f89056e8b46523d3da/publish/image

New layered fusion method, test the effect. This version is not exactly the same as what I published on domestic websites

More pic reference examples (over 1000s)

本地的参数：DPM++2M Karras     迭代步数：25    cfg：15  分辨率：512X912

开启插件：AD修脸、Dynamic Thresholding (CFG Scale Fix)

无高分辨率修复

parameters: DPM++2M Karras steps: 25 cfg: 15 Resolution: 512X912

Enable plugins: ADetailer, Dynamic Thresholding (CFG Scale Fix)

No high-resolution fix

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

V1.1:第一版的细节不够，所以添加了细节和阴影的效果。然后就出来两种风格。看个人喜好下载。

第一次融合模型，其实是我自己用的，拿来分享一下吧。这是我喜欢的风格。

试了一下，提示词的位置很影响NFSW的几率。要想出NFSW，要把其关键字尽量放前面，比如跟在1 girl的后面，这样的格式：1girl,((nude,nsfw))

建议采样方法：DPM++ 2M Karras，迭代步数25-30，面部修复采用ADetailer插件，同时ADetailer model这里选择 mediapipe_face_full。

当然你也可以尝试选择高清修复来生成更大尺寸的照片，因为我的显卡只有4G显存，就没有进行尝试了。

样图添加了 pureerosface_v1,ulzzang-6500-v1.1两个文本嵌入模型。可以让生成的脸型更符合亚洲的审美。没有添加脸型的lora。VAE配合使用的是clearVAE。

以上所需的文件如果你没有，可以联系我，我再上传。

正向提示词抽卡即可得到不错的效果：

best quality,masterpiece,realistic,photorealistic,1girl,pureerosface_v1,ulzzang-6500-v1.1

我的通用负面提示词：

(bad-artist,bad-hands-5,bad-image-v2-39000,bad_prompt_version2,EasyNegative,ng_deepnegative_v1_75t),advertisement,paintings,sketches,(worst quality:2),(low quality:2),(normal quality:2),

融合的模型：

majicmixRealistic_v6 https://civitai.com/models/43331

+realdosmix https://civitai.com/models/6925

+真人真实脸模 https://civitai.com/models/93300

以上三个模型的作者如果觉得本模型有侵权行为的话，请留言，我会下架本模型。

请勿用作商业用途，请勿用作真实公众人物的面部替换，如发生任何纠纷，请自行解决。

如有兴趣，欢迎本模型和任意模型进行融合，但请不要联系本人做技术咨询，因为我也是新手。- -

以下翻译来自deepl：

Translated with www.DeepL.com/Translator (free version)

V1.1:The first version of the details are not enough, so I add the details and shadow effects. Then it came out in two styles. You can download according to your personal preference.

The first merged model, in fact, I use it myself, to share it. This is the style I like.

Tried it, the position of the cue words very much affect the chances of NFSW. To come up with NFSW, try to put its keyword in front, such as following 1 girl, such a format: 1girl, ((nude,nsfw))

Suggested sampling method: DPM++ 2M Karras, Sampling steps: 25-30, facial repair using ADetailer plug-in, while ADetailer model here choose mediapipe_face_full.

Of course, you can also try to choose hires.fix repair to generate larger size and clearer photos, because my graphics card only has 4G video memory, so I did not try.

Two text embedding models, pureerosface_v1 and ulzzang-6500-v1.1, were added to the sample image. This allows the generated face shape to be more in line with Asian aesthetics. VAE is used in conjunction with clearVAE.

If you don't have the above required files, you can contact me and I will upload them.

U can get good results from simple prompt words:

best quality,masterpiece,realistic,photorealistic,1girl,pureerosface_v1,ulzzang-6500-v1.1

My generic negative words:

(bad-artist,bad-hands-5,bad-image-v2-39000,bad_prompt_version2,EasyNegative,ng_deepnegative_v1_75t),advertisements,paintings. sketches,(worst quality:2),(low quality:2),(normal quality:2).

The merged model:

majicmixRealistic_v6 https://civitai.com/models/43331

+realdosmix https://civitai.com/models/6925

+realistic face model https://civitai.com/models/93300

The authors of the above three models will take down this model if they feel that this model is infringing, please leave a message.

Please do not use it for commercial use, please do not use it as a facial replacement for real public figures, if any dispute occurs, please solve it by yourself.

If you are interested, you are welcome to merge this model with any model, but please do not contact me for technical advice, because I am also a novice.