No trigger word needed.Recommended settings:<ul><li>LoRA strength: 0.75–1.0</li><li>Step: ≥16</li><li>CFG: 3–5</li><li>Sampler: Euler/Heun</li><li>Scheduler: Simple</li></ul>The training set is taken from OrangeO recent miscellaneous image collection, with a portion of the bad-hand images removed; resolution 1344 px.For tagging I used the official recommended JSON format (long, medium and short descriptions included; on top of that I also marked the censor positions with bboxes — maybe that helps prevent the model from overfitting to the censored parts?).See: <a target="_blank" rel="ugc" href="https://github.com/OpenMOSS/MOSS-Image/blob/main/USERGUIDE.md">https://github.com/OpenMOSS/MOSS-Image/blob/main/USERGUIDE.md</a>Follow OrangeO, and thanks for the follow.<a target="_blank" rel="ugc" href="https://www.pixiv.net/users/128894200">https://www.pixiv.net/users/128894200</a><hr />不需要触发词。推荐设置：<ul><li>LoRA strength: 0.75–1.0</li><li>Step: ≥16</li><li>CFG: 3–5</li><li>Sampler: Euler/Heun</li><li>Scheduler: Simple</li></ul>训练集取自橘佬近期杂图集，剔除了一部分坏手的图片，分辨率 1344 P。打标方面我使用了官方推荐的 JSON 格式（包括长中短三种描述，除此之外我还用 bbox 标出了 censor 的位置，也许能防止 censor 部分过拟合？）详见：<a target="_blank" rel="ugc" href="https://github.com/OpenMOSS/MOSS-Image/blob/main/USERGUIDE.md">https://github.com/OpenMOSS/MOSS-Image/blob/main/USERGUIDE.md</a>关注橘佬喵，关注橘佬谢谢喵。<a target="_blank" rel="ugc" href="https://www.pixiv.net/users/128894200">https://www.pixiv.net/users/128894200</a>

Files with SHA256: 167c5149184b9c16396271653202f1f224344bba6272a6afd7c325d1e8c2bd5d