用感觉来理解未知的事物，这疑似有点太原始了，但是你有更好的方法吗。Understanding the unknown through sensation, this method may be somewhat primitive, but do you have a better approach.<hr />一次完整的模型处理需要13个流程，这些流程分为3种，总之我会尽量完整的解释给你们A complete model processing requires 13 processes, which are divided into 3 types. In short, I will try my best to explain it to you thoroughly流程1拆解模型 A ：(被提取模型) 模型 B ：Extract component B 模型 C : Extract component Cα= -1Process 1 BreakdownModel a：( extracted model ) Model b ：Extract component B Model c : Extract component Cα= -1流程2阶段提取模型 A ：流程1提取物 1 模型 B ：(被提取模型) 模型 模型 C ：流程1提取物 1α= 2进行这一步混合后会得到 流程1提取物2，之后你需要不断更换模型A和模型C 用于得到 流程1提取物4Phase 2 extractionModel A: Process 1 Extract 1 Model B: (Extracted Model) Model C: Process 1 Extract 1α = 2After this step of mixing, you will obtain Process 1 Extract 2. Then, you need to continuously switch between Model A and Model C to obtain Process 1 Extract 4我简单介绍一下获取的模型类型Let me briefly introduce the types of models obtained流程1提取物 1：通过提取模型以及混合算法，去除了被提取模型的基础部分，而留下来的则是这个模型的画风。这部分我称作画风1.0Process 1 Extract 1：By extracting the model and using a mixed algorithm, the foundational parts of the extracted model have been removed, leaving behind what I refer to as the style of the model. This part I call Style 1.0流程1提取物 2：基础1.0 、流程1提取物 3：画风2.0 、流程1提取物 4：基础2.0Process 1 Extract 2：Basic 1.0 Process 1 Extract 3：Style 2.0 Process 1 Extract 4：Basic 2.0流程3提纯 流程3需要结合之前的流程使用模型 A ：画风2.0 模型 B ：画风2.0拆解 流程1提取物 3 模型 C ：画风2.0α = -1现在这个模型就回到了未拆解的阶段，并成功的进行了活性化基础2.0的混合路线也是如此Phase 3 refine Process 3 needs to be used in conjunction with the previous processesModel A ：Style 2.0 模型 B ：Style 2.0 Breakdown Process 1 Extract 3 模型 C ：Style 2.0α = -1The model has now returned to the unassembled stage and has successfully been ActivationThe mixed route of Basic 2.0 is the same.这种方式也可以增加提取物信息的深度，不过需要进行一些调整This method can also enhance the depth of the extract information, but some adjustments are needed我承认我并不是一个专业的研究者，我所做的一切或许毫无意义，又或者我发现的都是错误的，总之，我现在已经感到了疲劳，显然我已经失去了热情。我把大量的时间用在模型混合上，但始终没有得到突破式的发展，或许从最开始我的方向便是错误的。模型训练是AI的正确路线，模型混合不是。模型版本升级是正确路线，模型混合不是。所以你有什么想法吗。

KM_Model_splicing_material_demonstration

XL版本的模型确实很大，我想我的硬盘存储不了几个，pony模型效果确实让人感到惊艳。但可惜的是我的所有素材都是SD1.5，而且我还有一些古怪的样本需要测试。The model of version xl is indeed very large, and I think my hard drive cannot store several of them. The pony model's effect is truly astonishing. However, it is unfortunate that all my materials are sd1.5, and I also have some peculiar samples that need testing使用LORA对模型进行污染，然后进行模型拆分。效果是不需要提示词就能表现出像素风格，嗯等同于直接使用LORA。(这不是没有任何意义吗！)好了理论上也可以适用其他风格。Use lora to contaminate the model, and then perform model splitting. The effect is that it can exhibit pixel style without the need for prompts, equivalent to directly using lora. (Isn't this meaningless!) Well, theoretically it can also be applied to other styles虽然没有达到预期数据，但是我想也不会在提升了，而且我也不希望长时间停留在该阶段。总之明天上传提取用的模型。Although the expected data has not been achieved, I do not think it will improve further, and I also do not wish to remain in this stage for an extended period. In summary, I will upload the model for extraction tomorrow

Files with SHA256: 55266bbe19e155f00671cbfd37377c09f86b6c2c5578100544ccb06ea66b0147