<h3 id="netalumina2"><strong>基于NetaLumina2训练</strong></h3><h3 id="based-on-netalumina2"><strong>Based on NetaLumina2</strong></h3><h1 id="v2:"><strong>V2:</strong></h1><pre><code>
触发词：&lt;style&gt;Blue Archive CG&lt;/style&gt;
Trigger Word: &lt;style&gt;Blue Archive CG&lt;/style&gt;</code></pre><h1 id="v1:">V1:</h1><pre><code>
触发词：&lt;style&gt;An image in the style of Blue Archive CG&lt;/style&gt;
Trigger Word: &lt;style&gt;An image in the style of Blue Archive CG&lt;/style&gt;</code></pre><p><strong><em>在部分情况下，v1的表现不一定强于v0.5</em></strong></p><p><strong><em>v1 is not always an improvement over v0.5</em></strong></p><p>推荐权重为1.0，若饱和度过高可酌情提高权重</p><p>Recommended weight is 1.0. If you find the image is over-saturated, you can try increasing the weight.</p><h1 id="v0.5">v0.5：</h1><p>推荐权重为2.0，此版本为训练时的第一个epoch，在使用2倍权重时会有较好的效果，但饱和度会比较低，饱和度过低时，可酌情降低权重</p><p>Recommended weight is 2.0. This is the first epoch from the training process. It produces good results at a weight of 2.0, but with lower saturation. If the saturation is too low, you can try decreasing the weight.</p>

Files with SHA256: a75797e201b8b754727d3347f12dd3bbbf7dd991ac441ddd576cba0781335f4b