由于网站奇怪的审核机制，很多很正常的图片被夹，所以能发出来的只有一小部分了。<h3 id="heading-98">严禁使用该lora生成任何NSFW内容</h3><ul><li>可以适配大多常见人物类checkpoint，示意图使用 dreamlikefiffusion10 生成。总体上说，真人模型效果會比二次元模型要好。</li><li>一定要开启Hires. fix！！，推荐使用R-ESRGAN 4x+ Anime6B，Denoising 0.5-0.75</li><li>lora 权重推荐0.8~1.0。Steps 推荐15-20</li><li>sampling method 推荐 Euler a</li><li>trigger words：chibi figure，提高权重，人物越“Q”，降低权重，人物越接近真人比例</li><li>与大多数衣服，人脸类lora都有较好的适配性，但是由于是Q版形象，不会显示过多细节。</li><li>该LoRa在训练的过程中，未使用任何真人数据。只有当checkpoint中有相应的人物数据时，才能生成如示意图中的Johnny Depp， Rihanna等的图片。</li><li>如果喜欢请点个❤️，并且分享您使用后的作品和建议</li></ul><h3 id="heading-99">Using this LoRa for generating any NSFW content is strictly prohibited.</h3><ul><li>It can be adapted to most common character checkpoints.The illustrative images above are generated using dreamlikefiffusion10. Overall, the results of photorealistic models are better than those of anime models.</li><li>Be sure to enable Hires. fix!! It is recommended to use R-ESRGAN 4x+ Anime6B and Denoising 0.5-0.75.</li><li>Recommended weight for LoRa: 0.8 to 1.0. Recommended steps: 15-20.</li><li>Recommended sampling method: Euler a.</li><li>Trigger words: chibi figure. Increase the weight for more "chibi" characters, decrease the weight for characters closer to realistic proportions.</li><li>It has good compatibility with most clothing and facial LoRa models, but due to the chibi style, it won't show excessive details.</li><li>This LoRa was trained without using any real human data. It can only generate images of specific individuals like Johnny Depp and Rihanna when the checkpoint contains corresponding character data, as shown in the illustrative images.</li><li>If you like it, please give a ❤️ and share your works and suggestions after using it.</li></ul>

Files with SHA256: db590696623556d699376c5502aef7eb4faada5ff0e588934bfdebb6ec323b17