<h2 id="introduction">Introduction</h2><a target="_blank" rel="ugc" href="https://virtualyoutuber.fandom.com/wiki/Hanazono_Serena">Hanazono Serena</a> is a Japanese female indie VTuber known for her cat-girl appearance.<h2 id="usage">Usage</h2>Recommended LoRA strength: 1.0-1.2Trigger word:<pre><code>hanazono serena</code></pre>Default outfit (Yes there are so many):<pre><code>maid headdress, black x hairclip, blue fish hairclip, pink star hairclip, navy side hair bows, navy cropped jacket, lavender sailor collar, gold star sleeve ornaments, white frilled blouse, light blue neck bow, blue oval brooch, black lace-up corset, layered lavender scalloped skirt, white apron with purple flower and four-pointed star trim, long aqua-to-lavender gradient back bow tails, white ankle cuffs, navy ribbon shoes</code></pre><h2 id="training-details">Training Details</h2><ul><li>Base model: Anima Base 1.0</li><li>Training tool: <code>sorryhyun/anima_lora v1.17.1</code></li><li>Dataset: 21 manually captioned training images</li><li>LoRA rank / alpha: 32 / 128</li><li>Learning rate: <code>2e-5</code></li><li>Optimizer: AdamW</li><li>Learning rate schedule: Cosine decay with 5% warmup</li><li>Training duration: 40 epochs / 840 steps</li><li>Precision: BF16</li><li>Resolution: Aspect-ratio buckets, approximately 1 megapixel</li><li>GPU: NVIDIA GeForce RTX 3090</li></ul><h2>简介</h2><a target="_blank" rel="ugc" href="https://virtualyoutuber.fandom.com/wiki/Hanazono_Serena">花園セレナ</a> 是一位日本女性独立 VTuber，以猫娘形象活动。<h2>使用</h2>推荐权重：1.0激活词：<pre><code>hanazono serena</code></pre>常服（是的没弄错，就是有这么多 tags）：<pre><code>maid headdress, black x hairclip, blue fish hairclip, pink star hairclip, navy side hair bows, navy cropped jacket, lavender sailor collar, gold star sleeve ornaments, white frilled blouse, light blue neck bow, blue oval brooch, black lace-up corset, layered lavender scalloped skirt, white apron with purple flower and four-pointed star trim, long aqua-to-lavender gradient back bow tails, white ankle cuffs, navy ribbon shoes</code></pre><h2>训练参数</h2><ul><li>基础模型：Anima Base 1.0</li><li>训练工具：<code>sorryhyun/anima_lora v1.17.1</code></li><li>训练数据：21 张手工标注训练图片</li><li>LoRA rank / alpha：32 / 128</li><li>学习率：<code>2e-5</code></li><li>优化器：AdamW</li><li>学习率调度：余弦衰减，预热比例 5%</li><li>训练量：40 epochs / 840 steps</li><li>训练精度：BF16</li><li>训练分辨率：按长宽比分桶，约 1 M 像素</li><li>训练显卡：NVIDIA GeForce RTX 3090</li></ul>

Files with SHA256: 348a8f2c58f379ee210a5af36817ce772dd0ffd73f342b94f98c9c05b7edd4d1