<hr /><h2 id="heading-2737">v2b</h2>添加了圣诞皮肤和泳装皮肤作为数据集制作 3in1 模型。基本触发词：<ul><li>原始皮肤：uniform, cloak, thigh boots</li><li>圣诞皮肤：antlers, dress, christmas</li><li>泳装：swimsuit</li></ul>该版本特征丢失较为严重，故建议在使用中对照生成结果额外补充以下提示词：<ul><li>圣诞皮肤：white thighhighs, ribbon on breasts, very long hair 等</li><li>泳装：red frills 等</li></ul>同时，建议在使用某一装扮时将另外两个装扮的触发词加入负词缀。比如，在生成圣诞装扮时，将 uniform, cloak 加入负词缀。已知问题：在提示词较少时经常画出默认皮肤的手杖。Add Christmas suit and swimsuit into dataset, now this lora is 3in1.Base Trigger:<ul><li>Origin: uniform, cloak, thigh boots</li><li>Christmas: antlers, dress, christmas</li><li>Swimsuit: Just "swimsuit"</li></ul>Feature loss in this version is quite severe. Therefore, I recommended to supplement the following prompt words by comparing the generated results during use:<ul><li>Christmas: white thighhighs, ribbon on breasts, very long hair, etc</li><li>Swimsuit: red frills, etc</li></ul>Meanwhile, I recommended to add the trigger words of other two outfits into negative when you only using one outfit. For example, when generating Christmas, add ‘uniform' and 'cloak' into negative.Known issue: Often drawing a cane with default skin when there are few prompt words.<hr /><h2 id="heading-2738">v1</h2>[Note] Prompt of example grid:1girl, solo, princeton, full body, standing, dress, legwear, &lt;lora:Princeton_v1-000008:0.8&gt;,No negative.一个简单的《碧蓝航线》普林斯顿角色LoRA，仅使用单张图片训练。<ul><li>触发词：princeton, dress, legwear</li><li>推荐权重：0.5<ul><li>此权重下可方便地为角色更换服装和姿态。而需要完整还原角色，您可以适当提高权重。</li><li>高权重下建议将 purple background 添加到负词缀中。</li></ul></li></ul>Simply a character LoRA of Princeton from mobile game Azurlane. Trained from single image.<ul><li>Trigger: princeton, dress. legwear</li><li>Weight: 0.5<ul><li>On that weight you can easily draw other clothing for Princeton. Increase it when you want to revival her entirely.</li><li>Negative "purple background" when you on a high weight.</li></ul></li></ul>

Files with SHA256: 462bdcc20cda3ec696b0d4a9abaf3d962711f1fa172ec569db458b2be214e8fe