强度：0.9当与角色lora共同使用时，最好将角色lora的强度降至0.8。已知问题：<ul><li>这不是一个很好的lora，在某些模型上会表现出明显的过拟合（尤其是在偏3d风格的模型上），如果您遇到这种情况，尝试使用一些画师串或者风格lora，会有较大的改善。因为我只有五张正面图和两张背面图来训练，而且都很模糊。我首先尝试ai放大这些图像，但是图像过于模糊导致视觉模型都识别不出裙子上的珍珠链子，放大后总是变成奇怪的蕾丝甚至铠甲，所以还是直接用原图进行了训练。</li><li>如果不明确写出视角提示词，生成的人物很可能是侧身的。</li><li>生成全身图像时，推荐使用放大和面部修复。</li><li>当角色胸部过大或使用俯视角时，胸口的珍珠链子很可能会消失。</li></ul>---Strength: 0.9When using this LoRA with character LoRAs, it's best to lower character strength to 0.8.Known issues:<ul><li>This is not a very good LoRA. It may show clear overfitting on some models (especially on those with a more 3D style). If this happens, try using some artist combinations or style LoRAs for major improvement,it's helpful.Because I only had five front-view images and two back-view images for training, all of which were blurry. I first tried AI upscaling, but the images were too fuzzy—even visual models couldn't recognize the pearl chains on the dress. After upscaling, they often turned into strange lace or even armor, so I proceeded with training using the original images.</li><li>If you don't include a perspective prompt (e.g., from front/view from behind), the generated character will likely be facing sideways.</li><li>For generating full-body images, it's recommended to use hires. fix and face restoration.</li><li>The pearl chains on the chest may disappear if the character's chest is too large or when using a top-down view angle.</li></ul>

Files with SHA256: 14d4ff79bfff5e14b6e63d2b6a89e1da79d25438031d156c5820099e376eae02