这个lora是关于KAFUKA系列的游戏中风格的SD1.5模型，训练集中的80张图片来自于游戏内的截图，全部经过细致的处理，为了尽可能最大程度的保留卡芙卡的特征。
训练基础模型：MeinamixV11(推荐使用)
推荐权重：0.8-1
注意事项：这个lora固化了卡芙卡的特征，因此不需要特别的提示词，可以根据我范图中的提示词生成，可以获得很好的结果（过多的提示词会导致不好的结果），也可以结合openpose使用。由于该lora模型是固化特征，所以你无法通过提示词的改变很好地改变KAFUKA的特征。后续我会陆续上传各种用途的KAFUKA系列模型。
This Lora is an SD1.5 model about the style of the KAFUKA series of games. The 80 images in the training set are taken from screenshots within the game, all of which have been carefully processed to preserve the features of the Kafka as much as possible.
Training basic model: MeinamixV11 (recommended)
Recommended weight: 0.8-1
Attention: This LORA solidifies the characteristics of Kafka, so there is no need for special prompt words. It can be generated based on the prompt words in my diagram and can achieve good results (too many prompt words can lead to bad results).It can also be used in conjunction with openpose. Due to the fact that the Lora model is a fixed feature, you cannot effectively change the features of KAFUKA by changing the prompt words. I will upload various KAFUKA series models for various purposes in the future.
Description
3.5版本微调改善了上一版本中颜色过紫的问题，通过加入了光源的tag纠正解决了部分游戏截图存在的泛光问题，同时更换了训练模型获得更好的lora模型适配度~主要注意以下几点：
1.想获得最好的效果请使用realcartoon3D这个base model，因为我是在该模型上训练出来的lora,其他模型可能存在特征不明显的问题。我牺牲了泛化能力换取了最高的还原能力，因为这是我的初衷，并且这也是我日常工作的内容（该行业的从业人员）
2.使用时请根据我示例图中的提示词去改变人物的方向、背光还是面光、视角(fromside/frombehind/frombelow/fromabove)等情况~
3.如果你想生成不同服装的KAFUKA，我的模型并不能带来好的效果，目前的几个版本还原度和出图效果是我追求的最终目标
4.如果你需要其他的模型可以评论联系我，能为我提供训练集也是对我最大的帮助！
希望使用愉快！
The fine-tuning of version 3.5 has improved the issue of excessive purple color in the previous version. By adding a light source tag to correct the problem of flooding in some game screenshots, the training model has been replaced to achieve better Lora model adaptation. The main points to note are as follows:
1. For the best results, please use the realcaroon3D base model because I trained Lora on this model, and other models may have issues with unclear features. I sacrificed my generalization ability in exchange for the highest restoration ability, because this is my original intention, and it is also the content of my daily work (professionals in the industry)
2. When using, please use the prompts in my example image to change the direction, backlight or sidelight, and perspective of the character~
3. If you want to generate KAFUKA for different clothing, my model cannot bring good results. The current several versions have restoration and graphic effects as my ultimate goal
If you need any other models, please feel free to comment and contact me. Providing me with a training set would also be the greatest help!