<ul><li>该模型基于 Animagine XL V3.0 Base 底模训练</li><li>用法：<ul><li>无需提示词触发。</li><li>推荐大模型：Animagine XL V3.1。</li><li>不使用 VAE。</li><li>采样器：Euler a，采样步数 28 步。</li><li>CFG Scale：6-7。</li><li>T2I 分辨率：不低于 768，推荐1280*。</li><li>Lora 权重：1.0。</li><li>开启 ADetailer 可以更好地修复手和脸。</li></ul>注意事项：<ul><li>衣服在训练时作为训练内容，可能难以替换。</li><li>如果出现动作promote难以推理，请更换或者使用相似的promote。</li></ul>该 Lora 模型的训练数据集截取自番剧《GIRLS BAND CRY》第二集&amp;第三集，采用标准优化器 AdamW 训练。不推荐使用其他模型推理，个人推测可能是因为训练集画风一致造成，此次训练未使用正则去规范画风，旨在还原动漫角色。</li></ul><ul><li>This model is based on the Animagine XL V3.0 Base model.</li><li>Usage:<ul><li>No prompt required.</li><li>Recommended base model: Animagine XL V3.1.</li><li>No VAE.</li><li>Sampler: Euler a, 28 sampling steps.</li><li>CFG Scale: 6-7.</li><li>T2I resolution: Not less than 768,1280* is recommended..</li><li>Lora weight: 1.0.</li><li>Enabling ADetailer can improve the generation of hands and faces.</li></ul></li><li>Notes:<ul><li>Clothes may be difficult to replace as they were part of the training data.</li><li>If the action is difficult to infer, please replace it or use a similar action.</li></ul></li></ul>The training dataset of this Lora model is intercepted from the second &amp; third episodes of the anime "GIRLS BAND CRY" and trained with the standard optimizer AdamW. It is not recommended to use other models for inference, which I personally speculate may be because the training set has a consistent style, and no regularization was used to normalize the style during this training, which aims to restore anime characters.

Files with SHA256: 63c1e7a991b2a42d43ea233b475a93b29599fa9b6b3d23006ef9ec9023d31890