<h2 id="heading-2398">v3</h2><h3 id="heading-2399">中文</h3><p>全新的训练数据，第一次尝试把两个服装融进一个lora里。训练时保留了一部分服装tag以获得更稳定的输出和更容易地换装，结果确实令人惊喜。推荐权重0.8，想要改变外观只需要保留daie white/daie black及头发相关提示词，并且降低权重到0.6左右即可。如果换装时发色也变了的话添加"blonde hair"即可。对于黑色外观，添加armored boots有很大概率是全身图，即使添加cowboy shot也不能获得相应的图像，需要去除armored boots。同时，有概率出现其他穿着铠甲的人，对于这个问题，解决办法是不要让她干巴巴的站着(standing)，添加一些其他的描述，比如背景，姿势，动作等。<span style="color:rgb(250, 82, 82)">训练素材为纯黑背景，一般需要添加背景描述词。</span><span style="color:rgb(18, 184, 134)">在真实模型上也能有一定程度的表现，如果你会分层使用lora，则可获得更好的图像。</span></p><h3 id="heading-2400">English</h3><p>Brand new training data, the first attempt to integrate two clothing into one lora. A part of the clothing tag was kept during training to obtain a more stable output and make it easier to change clothes. The result is indeed surprising. The recommended weight is 0.8. If you want to change the appearance, you only need to keep daie white/daie black and hair-related prompts, and reduce the weight to about 0.6. If the hair color also changes when changing clothes, just add "blonde hair".For the black appearance, adding armored boots has a high probability of being a full-body image, even if adding cowboy shot, the corresponding image cannot be obtained, so you need to remove armored boots. At the same time, there is a probability that other people wearing armor will appear. For this problem, the solution is not to let her stand lonely, but to add some other descriptions, such as background, posture, action, etc.<span style="color:rgb(250, 82, 82)">The training material is a pure black background, and background descriptors generally need to be added.</span><span style="color:rgb(18, 184, 134)">It can also have a certain degree of performance on the real model. If you can use lora layered, you can get a better image.</span></p><p><a target="_blank" rel="ugc" href="http://tusi.art">tusi.art</a>是一个免费在线图像生成网站，碍于电脑配置的推荐去试试~</p><p>V2</p><p>修复了哪里都是的头发。</p><p>fix the meaningless hair.</p><p>ps:试了下好像真人模型也是可以的~不保证效果，因为没有真人图片</p><p>After trying, I found that it may also be applicable to real models,but I cannot guarantee the effect.</p><p>新人第一次炼丹，出现错误请多多包涵~</p><p>This is my first lora,may have many errors.Forgive me.</p><h3 id="heading-2408">不允许商业用途</h3><h3 id="heading-2409">NO COMMERCIAL USAGE</h3>

Files with SHA256: 02cc2aea6097b5a231e0b093e3279b3e587de968231e3b85005e5c39891b921d