<h1 id="clear-sky-yuan-or-moss-image-lora">Clear Sky 清透二次元画风 | MOSS-Image LoRA</h1>干净通透的二次元插画风：明亮清澈的大眼睛、柔和细腻的发丝、通透的天空和光感，配色清爽不脏。本 LoRA 为个人爱好训练，与原画师无关，非官方。请勿用于冒充画师本人、伪造约稿或商业用途。<h2>触发词</h2><ul><li><code>@sy4</code>（放在提示词最前面）</li></ul>训练时用的是 tag 写法，推荐按 <code>@sy4, general, 1girl, solo, ...</code> 的格式写。MOSS 的画风 LoRA 不写触发词也会生效，写上会更稳。<h2>推荐设置</h2>在 MOSS-Image SFT（int8）上测试：<ul><li>不加 DMD：40-50 步，CFG 3，euler / simple，shift 5（颜色更饱满、光影更丰富）</li><li>加 DMD：叠加 8 步 DMD 加速 LoRA，12 步，CFG 1，euler / simple（更干净明亮、出图快）</li><li>LoRA 强度：1.0（可在 0.8～1.2 之间调）</li><li>分辨率：约 1MP，例如 1024×1536、832×1216</li></ul><img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/aa395db1-487d-4f06-9743-08740d889cc6/original=true/aa395db1-487d-4f06-9743-08740d889cc6.jpeg" /><h2>提示词示例</h2><code>@sy4, general, 1girl, solo, white hair, long hair, fox ears, fox tail, animal ear fluff, oversized hoodie, sitting by a window, indoors, holding cup, looking at viewer, smile</code><code>@sy4, general, 1girl, solo, silver hair, long hair, wolf ears, winter coat, scarf, snow, night, street lights, breath, looking at viewer, smile</code>小提示：<ul><li>画面里想出现文字时，在提示词里写清楚文字内容；没写就出现的乱码字，是没指定文字造成的，不是 LoRA 的问题。</li><li>示例图的提示词和种子都写在图片信息里，可以直接复现。</li></ul><h2>训练信息</h2><ul><li>底模：MOSS-Image Base</li><li>数据：79 张，WD14 tag 标注</li><li>训练器：MonadForge-Kohaku</li><li>rank 32 / alpha 128，学习率 1e-4，每步 1 张，8000 步（每张图约 100 遍）</li></ul><hr /><h2 id="english">English</h2>Clear Sky anime style for MOSS-Image. Clean, airy anime illustration: bright clear eyes, soft fine hair, open skies and light, fresh colors. Great for kemonomimi girls, school uniforms and everyday scenes.Fan-made and unofficial; not affiliated with the original artist. Do not use it to impersonate the artist or for commercial purposes.Trigger: <code>@sy4</code> at the start of the prompt. Trained on tags, so write prompts like <code>@sy4, general, 1girl, solo, ...</code>. The style also shows without the trigger, but it is more stable with it.Settings (MOSS-Image SFT int8): 40 steps, CFG 3, euler / simple, shift 5; or with an 8-step DMD LoRA at 12 steps, CFG 1. LoRA weight 1.0 (0.8–1.2). Around 1MP, e.g. 1024×1536.Training: 79 images, MOSS-Image Base, MonadForge-Kohaku, rank 32 / alpha 128, lr 1e-4, 8000 steps at 1 image per step (~100 passes per image).

Files with SHA256: f18da1c5b92eaaa135a37a6da2b554a59027de837635180251e89e392962c96f