<h3 id="heading-1874">导言</h3>这是一个《Umamusume Pretty Deby》<s>《闪耀</s><s>优俊少女</s><s>》</s>的爱如往昔角色LyCORIS。<h3 id="heading-1875">使用方法</h3>建议以0.8左右的权重使用。在打标过程中，为隔离元素污染，采用了每个卡面分开设关键词的方法，以下是各个立绘的关键词：特雷森制服：tracen_unifrom决胜服：shobofuku<h3 id="heading-1876">训练细节</h3>以NAI Final作为底模训练，使用了约140张图片作为训练集，生成同等数量的类似图片作为回归训练集，训练20个epoch，batch size 取为3，最终步数为960步。以下为主要设置参数：network_dim=32network_alpha=16keep_tokens=1conv_dim=8conv_alpha=4lr="5e-6"unet_lr="5e-6"text_encoder_lr="5e-6"batch_size = 3optimizer = Lion<h3 id="heading-1877">局限性</h3><ul><li>人物立绘中的不对称设计的还原难度非常大，基本上只能依靠随机抽卡和手动辅助达到一定的还原效果。</li></ul><h3 id="heading-1878">Introduction</h3>This is a LyCORIS of Still in love from "Umamusum Pretty Deby".<h3 id="heading-1879">Usage</h3>Suggest using a weight of around 0.8.Teresan uniform: tracen_unifromShobofuku：shobofuku<h3 id="heading-1880">Training Details</h3>Using NAI Final as the base model for training, approximately 140 images were used as the training set, and an equal number of similar images were generated as the regression training set. 20 epochs were trained with a batch size of 3, resulting in a final step count of 960.The following are the main settings parameters:network_dim=32network_alpha=16keep_tokens=1conv_dim=8conv_alpha=4lr="5e-6"unet_lr="5e-6"text_encoder_lr="5e-6"batch_size = 3optimizer = Lion<h3 id="heading-1881">Limitations</h3><ol><li>It is very hard to draw the asymmetrical designs in character portraits. Maybe a lot of tries and manual assistance is a certain way to achieve good results.</li></ol>

Still in Love - Umamusume Pretty Deby

使用PonyXL6炼制了一个版本，在肢体稳定性和prompt响应性上相比1.5版本来得好。参数参考了：<a target="_blank" rel="ugc" href="https://civitai.com/articles/3879/sdxl-lora-lazy-training-guide-anime-sdxl">https://civitai.com/articles/3879/sdxl-lora-lazy-training-guide-anime-sdxl</a>训练集与原来相同。A version trained on PonyXL6, which showed better limb stability and prompt responsiveness compared to version 1.5.The parameters refer to: <a target="_blank" rel="ugc" href="https://civitai.com/articles/3879/sdxl-lora-lazy-training-guide-anime-sdxl">https://civitai.com/articles/3879/sdxl-lora-lazy-training-guide-anime-sdxl</a>The training set is the same as before

Files with SHA256: ac46ebafd515293f7faa493b88eff7286e3b52a5b133de7f48e87cb2897f97c1