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: 0.8-1Attention: 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.

Honkai Star Rail:KAFUKA for game

1.在V1.5版本的基础上加入了正则化数据集和金字塔噪声，提高出图质量2.把训练集的透明背景换为了纯白背景避免出现背景崩坏。3.比1.5版本有更高的泛化性，削弱了来自于游戏截图内的3D模型感觉，并且KAFUKA的特征仍能实现优秀的保留4.训练集中新增了官方立绘5.更方便的改变动作和姿势（没有nsfw），和之前的版本一样没有做过多的提示词描述，提高模型的易调用性和特征固化能力，缺点就是你没法在高权重的情况下去实现衣服和饰品的更换，因为那就是我训练这个模型的初衷~1. On the basis of V1.5 version, regularized datasets and pyramid noise have been added to improve the image quality2. Change the transparent background of the training set to a pure white background to avoid background collapse.3. It has higher generalization than version 1.5, which weakens the feeling of 3D models from game screenshots, and KAFUKA's features can still achieve excellent retention4. Official vertical painting has been added to the training set5. It is more convenient to change actions and postures (without NSFW), and like previous versions, there is no excessive description of prompt words, improving the model's ease of recall and feature solidification1.V 1.5バージョンに加えて正規化データセットとピラミッドノイズを加え、描画品質を向上2.訓練集の透明な背景を真っ白な背景に変えて背景の崩壊を避ける。3.バージョン1.5よりも高い汎化性を持ち、ゲームスクリーンショット内からの3 Dモデル感覚を弱め、しかもKAFUKAの特徴は依然として優れた保持を実現することができる4.訓練集中に公式立ち絵を追加5.より便利な動作と姿勢の変更（nsfwはありません）、前のバージョンと同じように多くの提示語の記述をしていない、モデルの呼び出しやすさと特徴の硬化能力を高めて、欠点はあなたが高重み付けの状況で服とアクセサリーの交換を実現することができない、それは私がこのモデルを訓練する初心だからです~

Files with SHA256: c1a99bba149a0e21b2f65ddd08a292bcf8760a752456dd2f803cf07678770092