Update Notes(for more details check the "About this version" on the left):更新公告（更多详情请查看左侧"About this version"栏）v17.7(TV series version):Changes:-included more TV series Elaina into the training set, deleted all non TV series Elaina-performed more manual tagging, now you should see more characteristics of Elaina(i.e. her pink hair bow, braid...)-在训练图集里加入了更多剧场版的伊蕾娜，删除了非剧场版的图集-在训练图集提示里进行了更多的手动标记，现在您应该看到伊蕾娜的更多特征（如她的粉色发带、辫子...）Tips:the "Trigger Words"("elaina \(majo no tabitabi\)") is optional, but it might make things better(?)可以在提示中输入“触发词”("elaina \(majo no tabitabi\)")，非必要，但是可能会让图片更好（？） This model is made by <a target="_blank" rel="ugc" href="https://github.com/bmaltais/kohya_ss">kohya_ss gui</a>模型由<a target="_blank" rel="ugc" href="https://github.com/bmaltais/kohya_ss">kohya_ss gui</a>制作 Disclaimer: It is important to note that the usage of these artistic creations does not imply endorsement or affiliation with the original creators. The ownership and copyright of the artistic works used in the model remain with their respective creators, and this project does not claim any endorsement or association with them. The artistic creations are solely used for training and fine-tuning the LoRA model and are not being reproduced, modified, or transformed. No commercial use, distribution, or exploitation of the included artistic creations is allowed. All users of the model are expected to comply with applicable laws and respect the rights of the original creators. If there is any infringement, please notify us promptly, and we will promptly remove the relevant content.声明（如有出入请以英文为准）： 请注意，使用这些艺术创作并不意味着原创者对其进行认可或有任何关联。在该模型中使用的艺术作品的所有权和版权仍归原创者所有，本项目不声称任何认可或关联。这些艺术创作仅用于训练和优化 LoRA 模型，不会进行复制、修改或变换。严禁对包含的艺术创作进行任何商业用途、分发或利用。所有使用该模型的用户需遵守适用法律，并尊重原创者的权利。如有侵权，请及时通知，我们将及时删除相关内容。 Respective Contributors and Artists(listed in no particular order) 模型使用的作品来源（排名不分先后顺序）:Hefe: <a target="_blank" rel="ugc" href="https://www.pixiv.net/users/6458819">https://www.pixiv.net/users/6458819</a>nyoro: <a target="_blank" rel="ugc" href="https://www.pixiv.net/users/1337257">https://www.pixiv.net/users/1337257</a>RaVerzoArt-: <a target="_blank" rel="ugc" href="https://www.pixiv.net/users/49364520">https://www.pixiv.net/users/49364520</a>ダマスカス: <a target="_blank" rel="ugc" href="https://www.pixiv.net/users/620557">https://www.pixiv.net/users/620557</a>夏末: <a target="_blank" rel="ugc" href="https://www.pixiv.net/users/33947729">https://www.pixiv.net/users/33947729</a>サインこす: <a target="_blank" rel="ugc" href="https://www.pixiv.net/users/11229342">https://www.pixiv.net/users/11229342</a>エリス・グレイラット <a target="_blank" rel="ugc" href="https://www.pixiv.net/users/68823782">https://www.pixiv.net/users/68823782</a>nyoro <a target="_blank" rel="ugc" href="https://www.pixiv.net/users/1337257">https://www.pixiv.net/users/1337257</a>シャビークイズ <a target="_blank" rel="ugc" href="https://www.pixiv.net/users/50126589">https://www.pixiv.net/users/50126589</a>big radish <a target="_blank" rel="ugc" href="https://www.pixiv.net/users/22858438">https://www.pixiv.net/users/22858438</a>あずーる <a target="_blank" rel="ugc" href="https://www.pixiv.net/users/5838770">https://www.pixiv.net/users/5838770</a>しんりゅう <a target="_blank" rel="ugc" href="https://www.pixiv.net/users/2055332">https://www.pixiv.net/users/2055332</a>Wandering Witch: The Journey of Elaina(Anime television series) | 魔女の旅々动画版Please refer to their respective websites for more information about their work.请参考作家们的网页以了解更多关于他们的作品的信息。Again, thanks to these content creators' work.再次感谢以上作家们做出的贡献。 希望各位佬能指出我的不足之处，我超虚心接受意见的~

魔女の旅々(Majo no Tabitabi) | Wandering Witch: The Journey of Elaina | 魔女之旅 - イレイナ/Elaina/伊蕾娜

-redesigned prompts on training images-let some "more important" images repeat more times during the training process(I got the idea from <a target="_blank" rel="ugc" href="https://civitai.com/user/demoran/models">demoran</a>'s article "How to make a LoRA", check it out!)-reduced epoch(back to 1) on training and increase repeat times(20 times for general and 40 times "more important" images) on each image-more fine-tunings on training hyperparameters-重新设计了训练图集的提示词-让一些“”更重要“”的图片在训练是能重复更多次（参考了<a target="_blank" rel="ugc" href="https://civitai.com/user/demoran/models">demoran的文章：“How to make a LoRA”</a>，有兴趣请点进去查看）-降epoch降低至1，且提高了训练图片的重复次数（普通的图片重复20次，而更重要的图片会重复40次）-更多的参数微调

Files with SHA256: 84cee05464ea1139a403a85b1cc4d58aefcc9e1fd3865476bcd251ffbe2074c8