Low concentration model example

混合使用插件： sd-webui-supermerger

Mixed use plugin：sd-webui-supermerger

混合过程：差额叠加 ->trainDifference->模型 A Base model 模型 B 任何模型 模型 C Base model

mixing process：Add difference-> trainDifference->model A Base model model B Anything model , model C Base model

本次演示期间我统一使用这个参数 α=0.5

最近的研究目标是提纯模型特征，也就是高纯度，但进度缓慢。总之现在让我演示低浓度这一特征。

The recent research goal is to purify the model features, that is, high purity, but the progress is slow. In short, now let me demonstrate the low concentration feature.

模型介绍：模型融合的一个现象，很合理，但是要做到这一点并不容易。这个模型在混合时可以很好的获得其他模型的特征，仅限于2.5D及以上特征的模型，番剧类模型只能获得线条特征。据我推测它应该是优先获取了该模型纯度最高的特点

Model Introduction:

This is a phenomenon of model fusion, which is very reasonable. However, it is not easy to achieve this. This model can obtain the features of other models well when mixed, limited to models with 2.5D features and above, and animation models can at most obtain line features. According to my speculation, it should prioritize obtaining the characteristic of the highest purity of the model.

我在前几天混合的模型，之后进行了优化。在处理好了现实生活的琐事后我发现我忘记了如何混合模型，或许我应该留下一些文字记录。

Recently, I optimized the mixed model I worked on a few days ago. After dealing with the trivial matters of daily life, I found that I had forgotten how to mix the model. Perhaps I should leave some written records.

提炼特点的关键是找到关键的特点提炼，真好笑。

目前而言我仅提炼成功了一个高纯度特征模型，这种模型仅限融合使用，因为过于追求纯粹导致了画面崩坏因此需要其他模型混合稀释。

Currently, I have only successfully extracted a high-purity feature model. This model is only suitable for fusion use, as the excessive pursuit of purity leads to image distortion, thus requiring the blending and dilution of other models.

以下信息仅属于个人猜测，本人根据自己融合时发现的现象进行的推论猜想。事实上我也很怀疑自己的猜想。

The following information is only personal speculation. I made conjectures based on the phenomena I discovered during my own integration. In fact, I also doubt my own conjectures.

假设模型是一个图书馆，每次生成图片便会随机抽出一些图书信息，而这其中的图书类型则代表了整体模型画风。在这种猜想下，每次融合相当于图书交换，随着融合的加深整体的画风就会开始偏移。纯度代表图书馆内占比最大的类型书籍，浓度应该是不重复书籍的数量。好难，或许我该吃个脑子了
Assuming the model is a library, each time an image is generated, some book information will be randomly selected. The book types represent the overall style of the model. Under this assumption, each fusion is equivalent to a book exchange, and as the fusion deepens, the overall style will begin to shift. Purity represents the proportion of the most dominant type of books in the library, while concentration should be the number of unique books.

融合时的提示词可以类比为索引，LORA则是给图书馆临时展示的书籍，按照这种推论融合可以存在这以下方向：强化提示词效果、调整画风。好吧我该怎么做

The prompt word during fusion can be likened to an index, while Lora is for the temporary display of books in the library. According to this kind of inference, fusion can exist in the following directions: strengthening the effect of prompt words, and adjusting the art style. Well, what should I do?