<img src="https://imagecache.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/24f988d5-1887-44ce-ebb9-d7504410ac00/width=525/24f988d5-1887-44ce-ebb9-d7504410ac00" />::: 该页面为中国传统纹样loha的使用介绍，关于详细的训练过程，分层调试方法，可以到我们的wiki上去查看。<a target="_blank" rel="ugc" href="https://aigc.ioclab.com/sd-showcase/chinese-ornament.html">https://aigc.ioclab.com/sd-showcase/chinese-ornament.html</a> :::中国传统纹样lora by IoC Lab训练及数据：@shichen @ciaochaos版本 介绍 2023.4.1 上传lora，介绍内容（本篇） 2023.4.4 更新训练方法（wiki待上传）在图片中你可以看到有大约25个以上的版本，我挑选了其中最有特点的几个版本，你可以在version中挑选下载，或者前往huggingface下载全部的版本<a target="_blank" rel="ugc" href="https://huggingface.co/ioclab/ioc-lab-models/tree/main">https://huggingface.co/ioclab/ioc-lab-models/tree/main</a><img src="https://imagecache.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/a6458c3d-eeec-48ef-c929-f2e96db0cf00/width=525/a6458c3d-eeec-48ef-c929-f2e96db0cf00" /><h2>使用方法：</h2>在训练集的标注中，分为以下五个语段：<pre><code>traditional chinese pattern
</code></pre><pre><code>with circular symmetrical composition
with square symmetrical composition
with special-shaped symmetrical composition
</code></pre><pre><code>red and blue color scheme
</code></pre><pre><code>featuring horses and clouds
featuring flowers and branches
</code></pre><pre><code>a flower in the center
</code></pre>所以你可以围绕着这五种固定句式进行扩充，如需要作为背景，可以调低权重，增加 1 girl<h2>分层拆解：</h2>在分层图片中可见，在高权重下，通过仅开启 midd，outd 等 lora 层的权重，可以做到相比于直接降低权重，更能保留纹样风格的人物背景图。<h2>推荐模型版本介绍：</h2>这里是一些效果比较特色的模型版本，但出图效果与参数，种子等都有关系，你可以尽情尝试，期待你的评论<h2>特色模型版本</h2>以下是一些特色模型版本：<ul><li>4：特点是纹样更卡通化，色彩更缓和</li><li>7：特点是对关键词的还原更好，更符合数据集的倾向</li><li>10：特点是更对称，同时构图倾向会偏向于非中心对称构图</li><li>13：特点是相比10的颜色会更缓和，高权重下会出现第三种散布构图</li><li>15：特点是高饱和度和高横向对称构图</li><li>16：特点是高饱和度+散布构图</li></ul><h2>人物配合版本</h2>以下是容易和人物配合的版本：<ul><li>loha01_old3: 更容易出卡通光影，构图和花纹装饰性强</li><li>loha04_old3: 构图容易出全身，光影保留效果好</li><li>loha06_old1: 对构图限制小，人物动作自然</li><li>loha02_old2: 背景层次更多，对称性强</li></ul>数据集标注及loha训练方法，见wiki链接<a target="_blank" rel="ugc" href="https://aigc.ioclab.com/sd-showcase/chinese-ornament.html">https://aigc.ioclab.com/sd-showcase/chinese-ornament.html</a><h1>Chinese Ornament Wiki in English</h1>::: This page is an introduction to the use of traditional Chinese pattern Lora, for detailed training processes and layer debugging methods, please refer to our wiki. <a target="_blank" rel="ugc" href="https://aigc.ioclab.com/sd-showcase/chinese-ornament.html">https://aigc.ioclab.com/sd-showcase/chinese-ornament.html</a> :::Chinese traditional pattern Lora by IoC LabTraining and Data: @shichen @ciaochaosVersion Description 2023.4.1 Uploaded Lora, Introduction Content (This Article) 2023.4.4 Updated Training Method (Wiki To Be Uploaded)In the pictures, you can see more than 25 versions. I selected the most distinctive versions for you to download, or you can go to Huggingface to download all versions.<a target="_blank" rel="ugc" href="https://huggingface.co/ioclab/IocLab_Models/tree/main/Lora_Chinese_traditional_pattern">ioclab/IocLab_Models at main (</a><a target="_blank" rel="ugc" href="http://huggingface.co">huggingface.co</a><a target="_blank" rel="ugc" href="https://huggingface.co/ioclab/IocLab_Models/tree/main/Lora_Chinese_traditional_pattern">)</a><h2>How to use:</h2>In the annotation of the training set, the following five language segments are divided:<pre><code>traditional chinese pattern

</code></pre><pre><code>with circular symmetrical composition
with square symmetrical composition
with special-shaped symmetrical composition

</code></pre><pre><code>red and blue color scheme

</code></pre><pre><code>featuring horses and clouds
featuring flowers and branches

</code></pre><pre><code>a flower in the center

</code></pre>So you can expand around these five fixed sentence patterns. If you need to use it as a background, you can reduce the weight and add one girl.<h2>Layered Decomposition:</h2>As shown in the layered images, under high weight, by only turning on the weights of midd, outd, and other Lora layers, you can achieve a person-background image with a style more retained than directly reducing the weight.<h2>Recommended Model Version Introduction:</h2>Here are some model versions with distinctive effects, but the output effect is related to parameters, seeds, etc. You can try them out and we look forward to your comments.<h2>Distinctive Model Versions</h2>Here are some distinctive model versions:<ul><li>4: The characteristic is that the pattern is more cartoonish and the color is more gentle.</li><li>7: The characteristic is that it better restores the keywords and is more in line with the tendency of the dataset.</li><li>10: The characteristic is that it is more symmetrical, and the composition tends to be non-centrally symmetrical.</li><li>13: The characteristic is that compared to 10, the color will be more gentle, and the third scattered composition will appear under high weight.</li><li>15: The characteristic is high saturation and high horizontal symmetry composition.</li><li>16: The characteristic is high saturation + scattered composition.</li></ul><h2>Character Matching Versions</h2>The following are versions that are easy to match with characters:<ul><li>loha01_old3: More likely to produce cartoon light and shadow, with strong composition and decorative patterns.</li><li>loha04_old3: The composition is easy to produce the whole body, and the effect of retaining light and shadow is good.</li><li>loha06_old1: The restriction on composition is small, and the character's movements are natural.</li><li>loha02_old2: More layers in the background, strong symmetry.</li></ul>For dataset annotation and Lora training methods, please refer to the wiki link.<a target="_blank" rel="ugc" href="https://aigc.ioclab.com/sd-showcase/chinese-ornament.html">https://aigc.ioclab.com/sd-showcase/chinese-ornament.html</a><img src="https://imagecache.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/278491c1-e44b-45bc-130c-c8cfad77a000/width=525/278491c1-e44b-45bc-130c-c8cfad77a000" />

Chinese traditional pattern

<pre><code>特点是纹样更卡通化，色彩更缓和</code></pre>

Files with SHA256: b722522072e6dcd8e70176b3eda0a42d03a4a65f154759fa3d334409deacc6d2