此模型训练了三种常见的兽装风格(kemono fursuit,toony fursuit,realistic fursuit)。lora 强度：1~2注意：受限于模型的基础知识和本地设备的训练难度，此综合模型只添加了非常少量的其它方面训练，不如<a target="_blank" rel="ugc" href="https://civarchive.com/models/1531352/xl-realistic-fursuit">SDXL版</a>效果强烈和风格多样。但klein胜在可以进行多角色生成和复杂语义理解。使用turbo版时触发词不太可靠，可能需要抽奖尝试，将触发词融入进提示词中或许会改善。可以使用中英混合提示，但是直接写中文"兽装"不稳定，中文训练数据占比低。使用"cute,kawaii",可以接近和改善kemono风格。使用"cartoon"可以接近和改善toony风格。即使是turbo版也可以使用&gt;1的cfg来干预输出结果。base版学的像，但画质上turbo版表现好。base模型好于使用加速lora，加速lora好于turbo版。加速lora容易造成过饱和。你可以使用提示词和颜色矫正节点进行改善。This model has been trained with three common styles of fursuit(kemono fursuit,toony fursuit,realistic fursuit).lora strength: 1 - 2Note:Due to the limitations of the model's basic knowledge and the training difficulty of the local device, this comprehensive model only adds a very small amount of training for other aspects, and its effect and style are not as strong and diverse as the <a target="_blank" rel="ugc" href="https://civarchive.com/models/1531352/xl-realistic-fursuit">SDXL version</a>. However, the advantage of the klein model lies in its ability to generate multiple characters and understand complex semantics.When using the turbo version, the trigger words are not very reliable. It may be necessary to try a lottery to integrate the trigger words into the prompt. This might improve the situation.Using "cute , kawaii" can bring closer connection and improvement to the kemono style.Using "cartoon" can bring closer connection and improvement to the toony style.The base version learns well, but the turbo version performs better in terms of image quality. The base model is better than using accelerated Lora, and Lora is better than the turbo version. Accelerated Lora is prone to over-saturation. You can improve it by using prompt words and color correction nodes.

Files with SHA256: a49c5736d51db13d92933e24ec50211aaa47648be5788ca135b38012f204ec99