<h1 id="model-info">Model info</h1>v1.1 has the highest fidelity.This LoRA is fine-tuned on a dataset of 300+ images. <span style="background-color:rgb(26, 27, 30);color:rgb(193, 194, 197);font-family:-apple-system, BlinkMacSystemFont, &quot;Segoe UI&quot;, Roboto, Helvetica, Arial, sans-serif, &quot;Apple Color Emoji&quot;, &quot;Segoe UI Emoji&quot;;font-size:16px">It can recreate Yaoguang using the trigger word <code>Yaogaung</code><span style="background-color:rgb(26, 27, 30);color:rgb(193, 194, 197);font-family:-apple-system, BlinkMacSystemFont, &quot;Segoe UI&quot;, Roboto, Helvetica, Arial, sans-serif, &quot;Apple Color Emoji&quot;, &quot;Segoe UI Emoji&quot;;font-size:16px"> alone, while also providing detachable clothing tags.In addition to LoRA based on IllustriousXL, I have also trained versions based on NoobAI, Anima, AnimagineXL, PonyXL, and WAI-IllustriousXL.<h1 id="param-settings">Param Settings</h1><ul><li>Recommended weight: 0.8 - 1.0</li><li>example prompt :(Please adjust the prompts according to model card of the checkpoint you are using.)</li></ul><pre><code>masterpiece, best quality, very aesthetic, newest, hires, Yaoguang, 1girl, solo, looking at viewer, medium breasts, hair ornament, bracelet, dress, smile, upper body, romantic, love, detailed light, spotlight, simple background,</code></pre><ul><li>full body:</li></ul><pre><code>masterpiece, best quality, very aesthetic, newest, hires, Yaoguang, 1girl, solo, looking at viewer, hair ornament, bracelet, dress, high heels, smile, full body, detailed light, spotlight, simple background,</code></pre>For full-body shots, you can try adding <code>full body</code> or <code>high heels</code>.<h1 id="tips">Tips</h1><ul><li>Official design is bound to the following tags: <span style="background-color:rgba(135, 131, 120, 0.15);color:rgb(235, 87, 87);font-family:SFMono-Regular, Menlo, Consolas, &quot;PT Mono&quot;, &quot;Liberation Mono&quot;, Courier, monospace;font-size:85%;line-height:normal">dress, bracelet, earrings, single fur trim, jewelry, high heels</li><li>If you find the official outfit elements are not decoupling when using your own checkpoint, add cloth tags to the negative prompt. For example, if you want an bare-shoulder look, remove <span style="background-color:rgba(135, 131, 120, 0.15);color:rgb(235, 87, 87);font-family:SFMono-Regular, Menlo, Consolas, &quot;PT Mono&quot;, &quot;Liberation Mono&quot;, Courier, monospace;font-size:85%;line-height:normal">single fur trim or add it to the negative prompt.</li></ul><h1 id="others">Others</h1><span style="background-color:rgb(26, 27, 30);color:rgb(193, 194, 197);font-family:-apple-system, BlinkMacSystemFont, &quot;Segoe UI&quot;, Roboto, Helvetica, Arial, sans-serif, &quot;Apple Color Emoji&quot;, &quot;Segoe UI Emoji&quot;;font-size:16px">Models are automatically deployed via AI Agent. Please report any parameter errors or version issues in the comments.While I am very satisfied with the accuracy of Yaoguang's features, <span style="background-color:rgb(26, 27, 30);color:rgb(193, 194, 197);font-family:-apple-system, BlinkMacSystemFont, &quot;Segoe UI&quot;, Roboto, Helvetica, Arial, sans-serif, &quot;Apple Color Emoji&quot;, &quot;Segoe UI Emoji&quot;;font-size:16px">please feel free to leave your generations or feedback. Thank you for your support!<hr /><h2>模型简介</h2>v1.1有最高还原度300+ 张图片的小数据集上微调得到。此LoRA使用触发词<code>Yaoguang</code>即可还原人物，也提供了可拆卸的服装提示词，方便各类二创发挥。除了基于IllustriousXL的版本，我还做了基于NoobAI, Anima, AnimagineXL, PonyXL, WAI-IllustriousXL 版本。<h2>推荐设置</h2><ul><li>推荐权重： 0.8 - 1.0</li><li>（请根据你实际使用的Checkpoint模型的官方文档，适当调整提示词策略。）</li><li>上本身prompt:</li></ul><pre><code>masterpiece, best quality, very aesthetic, newest, hires, Yaoguang, 1girl, solo, looking at viewer, medium breasts, hair ornament, bracelet, dress, smile, upper body, romantic, love, detailed light, spotlight, simple background,</code></pre><ul><li>全身prompt:</li></ul><pre><code>masterpiece, best quality, very aesthetic, newest, hires, Yaoguang, 1girl, solo, looking at viewer, hair ornament, bracelet, dress, high heels, smile, full body, detailed light, spotlight, simple background,</code></pre>全身可尝试添加<code>full body</code>或<code>high heels</code> 。<h2 id="tips">Tips</h2><ul><li>官方服设绑定在以下标签：<span style="background-color:rgba(135, 131, 120, 0.15);color:rgb(235, 87, 87);font-family:SFMono-Regular, Menlo, Consolas, &quot;PT Mono&quot;, &quot;Liberation Mono&quot;, Courier, monospace;font-size:85%;line-height:normal">dress, bracelet, earrings, single fur trim, jewelry, high heels</li><li>如果你用你的checkpoint推理的时候服装元素不能正确解耦，把上述服装tag加到负面提示词。举例子，如果想要光肩膀的效果，移除 <span style="background-color:rgba(135, 131, 120, 0.15);color:rgb(235, 87, 87);font-family:SFMono-Regular, Menlo, Consolas, &quot;PT Mono&quot;, &quot;Liberation Mono&quot;, Courier, monospace;font-size:85%;line-height:normal">single fur trim或把它加进负面提示词。</li></ul><h2>小记</h2>模型由AI Agent全自动发布，若参数、模型版本有问题，欢迎评论反馈。虽然我已经对爻光的特征还原程度很满意，如果有任何反馈意见欢迎评论，以便我做出调整，谢谢。我会定期在社交媒体发布图片，可在对应平台取原图。Bilibili账号：<a target="_blank" rel="ugc" href="https://space.bilibili.com/12595237?spm_id_from=333.1007.0.0">https://space.bilibili.com/12595237?spm_id_from=333.1007.0.0</a>

Yaoguang | 爻光 (detailed 细节增强）Honkai:Star Rail 崩坏：星穹铁道

Yaoguang_illusXL_bs1_noTE_trig-Yaoguang.safetensors

v1.0

v1.1

v1.0 Zuki

<h1 id="model-description">Model Description</h1>This LoRA model is a character enhancement model for Yaoguang (爻光) from Honkai: Star Rail. Since currently only one official illustration is available, this model was fine-tuned on a small dataset of 200+ images.In addition to the version based on Illustrious XL, I have also trained versions based on WAI-IllustriousXL and ZukiAnimeIll (derivative models of Illustrious XL), which will be released soon. However, the version trained on the standard Illustrious XL is the most stable choice to avoid feature conflicts.<h1 id="usage-tips">Usage Tips</h1><ul><li>Clothing: Special optimization has been applied to address the "hard-baked" clothing issue. Yaoguang's official outfit is now detachable. Her official design is primarily bound to the following tags: <code>dress</code>, <code>bracelet</code>, <code>hair ornament</code>.</li><li>Changing Outfits: In my testing with common checkpoint models, removing the official clothing tags allowed for adding any preferred clothing elements. If residual elements persist when using your specific checkpoint for inference, consider adding the aforementioned tags (e.g., <code>dress</code>, <code>hair ornament</code>, etc.) to the negative prompt.</li></ul><h1 id="recommended-settings">Recommended Settings</h1><ul><li>Recommended weight: 0.7 - 1.0</li><li>(Please adjust the prompts according to the official documentation of the checkpoint model you are using.)</li><li>upper-body prompt :</li></ul><pre><code>(masterpiece, best quality, very aesthetic, newest, hires), Yaoguang, 1girl, solo, medium breasts, hair ornament, bracelet, dress, bare shoulders, looking at viewer, smile, upper body, romantic, love, detailed light, epic, dramatic,

</code></pre><ul><li>cowboy-shot prompt :</li></ul><pre><code>masterpiece, best quality, newest, absurdres, highres, 1girl, solo, Constance, hat, dress, gloves, high heels, tail, standing, medium breasts, detailed light, epic, dramatic,
</code></pre>For full-body shots, you can try adding <code>full body</code> and <code>high heels</code>.<h1 id="others">Others</h1>While I am very satisfied with the accuracy of Yaoguang's features, I cannot guarantee that every detail is perfectly stable yet. I plan to further optimize the details after the official 3D model is released.If you have any feedback, please feel free to leave a comment.Thank you for your support!<hr /><h2>模型简介</h2>这个LoRA模型是角色爻光的角色增强模型，因为目前素材仅有官方发布的一张立绘，我只能在一个 200+ 张图片的小数据集上微调。 除了基于Illustrious XL的版本，我还训练了基于WAI-IllustriousXL，ZukiAnimeIll（Illustrious XL的衍生模型）的版本，稍后会发布，但基于标准IllustriousXL训练的版本肯定是最稳妥的选择，因为不会有特征冲突。<h2>使用建议</h2><ul><li>服装与配饰 ： 针对服装 hard-baked问题进行了专门优化。现在爻光的官方外设是可拆卸的，官方服设主要绑定在绑定了以下标签：<code>dress, bracelet, hair oenament</code></li><li>更换服装： 实测下来我在我常用的checkpoint模型上在去掉上述官方服装提示词后，可以自由添加任意喜欢的服装元素。如果你在使用其他checkpoint模型推理时仍发现有官方服装残留，建议将上述标签（如 <code>dress</code>, <code>hair ornament</code> 等）加入到负面提示词中。</li></ul><h2>推荐设置</h2><ul><li>推荐权重： 0.7 - 1.0</li><li>（请根据你实际使用的Checkpoint模型的官方文档，适当调整提示词策略。）</li><li>上本身prompt:</li></ul><pre><code>(masterpiece, best quality, very aesthetic, newest, hires), Yaoguang, 1girl, solo, medium breasts, hair ornament, bracelet, dress, bare shoulders, looking at viewer, smile, upper body, romantic, love, detailed light, epic, dramatic,

</code></pre><ul><li>cowboy-shot prompt:</li></ul><pre><code>(masterpiece, best quality, very aesthetic, newest, hires), Yaoguang, 1girl, solo, medium breasts, hair ornament, bracelet, dress, bare shoulders, looking at viewer, smile, confident, sideboob, cowboy shot, romantic, love, detailed light, epic, dramatic, mibir,

</code></pre>全身可尝试添加<code>full body</code>, <code>high heels</code> 等。<h2>小记</h2>虽然我已经对爻光的特征还原程度很满意，但不能保证所有细节稳定，我会在官方3d模型发布后针对细节优化。如果有任何反馈意见欢迎评论，以便我做出调整，谢谢。我会定期在社交媒体发布图片，可在对应平台取原图。Bilibili账号：<a target="_blank" rel="ugc" href="https://space.bilibili.com/12595237?spm_id_from=333.1007.0.0">https://space.bilibili.com/12595237?spm_id_from=333.1007.0.0</a>小红书账号：<a target="_blank" rel="ugc" href="https://www.xiaohongshu.com/user/profile/670412f2000000001d0209cd">https://www.xiaohongshu.com/user/profile/670412f2000000001d0209cd</a>

Files with SHA256: c1a6fa1554d42305b72fb5ee4275f10e62c078a7c894fdf93b3c96c9017e26b8