<h2 id="matsubara-kanon-f.1-dev-lora-model-vb92gn5y4">Matsubara Kanon F.1 Dev LoRA Model</h2><p>A LoRA Model based on the base model <strong>FLUX.1 Dev.</strong></p><p>I only use the transparent images for training. So the background will be single-color, not flexible, I think.</p><h2 id="the-usage-lgceharpf">The Usage</h2><p>Well, The LoRA can be triggered by the single word 'Matsubara Kanon', However, I still recommend u to add '1girl, solo, purple eyes,  <strong>long hair(optional)</strong>, blue hair, <strong>light blue hair(Optional)</strong>' after the trigger word.</p><p>Also, I only tested the case that the weight of the LoRA is 1.25, and I don't really know if lower value can still make the output normal. <s>(You can test it yourself and tell me in the comment area)</s></p><h2 id="the-introduction-of-the-character-h78dli2ax">The Introduction of the Character</h2><p><a target="_blank" rel="ugc" href="https://danbooru.donmai.us/posts?tags=matsubara_kanon">Danbooru Posts</a></p><p><a target="_blank" rel="ugc" href="https://en.wikipedia.org/wiki/List_of_BanG_Dream!_characters#Kanon_Matsubara">Wikipedia</a></p><hr /><h2 id="-u3327w6qa">一些基本的训练参数</h2><p>素材以人物卡面为主，并且由于我存储的卡面都是透明图像，所以最终的输出的背景基本是单色的，方便后期抠图（确信</p><p>训练脚本是直接采用sd-scripts的sd3分支，并受到fluxgym写script的启发，自行手搓脚本和数据设置而成</p><p>考虑到本人笔电配置不佳，所以每次训练的效率不高，并且迄今为止找不到办法有效降低avr_loss的办法 <s>(除了拉高steps，但那样更费时间)</s></p><h2 id="lora-z16r6jark">怎么用这个LoRA</h2><p>跟正常使用SD 1.5的LoRA一样，用FLUX.1 Dev的底模，然后使用触发词'matsubara kanon'触发LoRA，目前测试的权重只有1.25，其它更低或更高的值没有测试。当然，为了保证角色特征是正确的，推荐加上"1girl,solo,purple eyes,long hair,blue hair,light blue hair"这些特征词，并且尽可能只使用DeepBooru词库的词语。</p>

Files with SHA256: 79d4e6a56c3733e12f42c57e516f7ad150b76179abefde72a240604c1a1c34b6