<s>为什么我感觉我训练的人物lora都是一些比较冷门的角色……</s><s>算了这不重要。</s>这个Lora模型是基于final-prune.ckpt（【旧短哈希1d4a34af】新短哈希22fa233c2d），使用了768x768的数据集进行训练。推荐权重为0.6-0.8.如果人物无法生成你想要的动作，请增加动作的权重或者降低Lora的权重。如果生成的人物图像面部不佳，推荐您勾选高清修复。这个模型可以生成nsfw的图片。相比于1.0版本，1.5版本能够提供更多的细节，但也更难操控。测试了kotosmix以及AbyssOrangeMix2，AbyssOrangeMix3，以及anyhentai和Grapefruit，在使用高清修复时基本上都工作良好，1.5版本在不使用高清修复时可能会导致人物面部不佳。英语是用Google翻译的，如果有造成误解之处非常抱歉。<s>Why do I feel that the lora characters I trained are all relatively unpopular characters...</s><s>Forget it is not important.</s>This Lora model is based on final-prune.ckpt ([old short hash 1d4a34af] new short hash 22fa233c2d), using a 768x768 dataset for training.The recommended weight is 0.6-0.8.If the character cannot generate the motion you want, please increase the weight of the motion or reduce the weight of Lora.If the face of the generated character image is not good, it is recommended that you check HD Repair.This model can generate pictures of nsfw.Compared with version 1.0, version 1.5 can provide more details, but it is also more difficult to control.I tested kotosmix, AbyssOrangeMix2, AbyssOrangeMix3, and anyhentai and Grapefruit, and they basically work well when using high-definition restoration. Version 1.5 may cause bad faces when not using high-definition restoration.English is translated by Google, sorry if there is any misunderstanding.

Files with SHA256: 715933ac06447356294a8907fa0903792db2f70ec3ee02b3673eb5fd49955a5f