<h1 id="heading-296">先看说明！！！</h1>鹅群：墨幽炼丹阁-太清阁：858495398墨幽炼丹阁-上清阁：764481772也请多多关注我的小红书和抖音账号：yy469370733<h3 id="heading-297">墨幽：主美，全部素材的收集、筛选、精挑，所有轮次训练集的制作，训练后模型的融合微调（所有的融合素材模型均为基于sdxl1.0的训练模型）及第4、5、6轮的大模型训练，最终微调及所有轮次的测试。</h3><h1 id="heading-298">特别鸣谢：</h1><h2 id="heading-299">天选打工人赵图图：第1、2、3轮的大模型训练及详细的技术指导。</h2><h2 id="heading-300"><a target="_blank" rel="ugc" href="https://civarchive.com/user/tututututuz/models">tututututuz Creator Profile | Civitai</a></h2><h2 id="heading-301">致谢：</h2>LEOSAM：对sdxl真实人物模型的可行性验证。 <a target="_blank" rel="ugc" href="https://civarchive.com/user/LEOSAM/models">LEOSAM Creator Profile | Civitai</a>凯登工作室（ArienTOP）：理论支持。<a target="_blank" rel="ugc" href="https://civarchive.com/user/ArienTOP/models">ArienTOP Creator Profile | Civitai</a>简介：xl模型在人像真实度方面有着质的飞跃，有着无与伦比的光影视觉冲击，丰富且富有感染力的情绪表达，不论是在影棚的时尚大片、真实感的户外摄影、自然的手机随拍、或是艺术写真方面均有良好的表现。<h2 id="heading-302">生成建议：（还请多参考例图tag）</h2><h2 id="heading-303">推荐tag：（不支持固定的简单tag稳定出图，建议多参考例图描述）</h2>正：chinese woman,moyou,Beautiful Chinese Women, （chinese woman触发亚洲人特征）<h3 id="heading-304">（正面触发tag选择性添加，建议参考例图描述，或入群自取原图包参考）</h3>负：(worst quality, low quality, cgi, bad eye, worst eye, illustration, cartoon),deformed,distorted,disfigured,poorly drawn,bad anatomy,wrong anatomy,open mouth,<h3 id="heading-305">（负面tag较为通用，可用作人物固定负面tag）</h3><h2 id="heading-306">基本可以无脑开启 After Detailer 修脸插件，开启时把正负面tag完全复制到 After Detailer 的正负tag对应位置</h2><h2 id="heading-307">推荐参数：</h2>CLIP 终止层数：2vae：已内置（选无）迭代步数：30-50采样方法：DPM++ 2M Karras推荐分辨率（横竖皆可）：1024×1024、1024×1280、768×1280、888×1280（分辨率亦可在1024×1024附近范围自行调节）提示词引导系数（CFG Scale）：7<h3 id="heading-308">不要开启高分辨率修复！</h3><h3 id="heading-309">不要开启面部修复！</h3>1、解决了ai摄影的低分辨率问题，1280×1024的直出分辨率拥有更清晰更丰富的细节，还拥有更细腻的质感。给直出后通过其他软件进行的二次放大加工提供了更高的上限。2、远超sd1.5的光照表现力，无与伦比的情绪感染力、更加自然真实的表情、更加细致丰富的细节刻画、更加夸张的镜头语言传递3、极其微妙的镜头语言，即使大头照也可有明显的大光圈效果。<h2 id="heading-310">还请多多返图支持！感谢！！！</h2>Let's take a look at the instructions first!!!鹅群：墨幽炼丹阁-太清阁：858495398墨幽炼丹阁-上清阁：764481772Please also pay more attention to my Little Red Book and Tiktok account: yy469370733Mo You: The main focus is on beauty, including the collection, screening, and refinement of all materials, the production of all training sets, the fusion fine-tuning of trained models (all fusion material models are based on SDxl1.0 training models), as well as the 4, 5, and 6 rounds of large model training, and the final fine-tuning and testing of all rounds.Special thanks:Zhao Tu, a migrant worker from Tianxuan, showed the first, second, and third rounds of large-scale model training and detailed technical guidance.<h2 id="heading-205"><a target="_blank" rel="ugc" href="https://civarchive.com/user/tututututuz/models">tututututuz Creator Profile | Civitai</a></h2>Thank you:LEOSAM: Feasibility verification of the sdxl real person model. <a target="_blank" rel="ugc" href="https://civarchive.com/user/LEOSAM/models">LEOSAM Creator Profile | Civitai</a>ArienTOP: Theoretical support.<a target="_blank" rel="ugc" href="https://civarchive.com/user/ArienTOP/models">ArienTOP Creator Profile | Civitai</a>Introduction: The xl model has made a qualitative leap in the authenticity of human portraits, with unparalleled visual impact of light and shadow, rich and infectious emotional expression. Whether it is in fashion blockbusters in the studio, realistic outdoor photography, natural mobile photography, or artistic photography, it has performed well.Generate suggestions: (Please also refer to the example tag)Recommended tag: (does not support stable plotting of fixed simple tags, it is recommended to refer to example diagrams for more descriptions)True: Chinese woman, moyou, Beautiful Chinese Women, (Chinese woman triggers Asian characteristics)(Selective addition of positive trigger tags, it is recommended to refer to the example image description, or self select the original image package for reference when joining the group)Negative:,(Negative tags are more general and can be used as fixed negative tags for characters)It is basically possible to open the After Detailer facelift plugin without thinking, and when enabled, copy the positive and negative tags completely to the corresponding positions of the positive and negative tags in the After DetailerRecommended parameters:CLIP termination layers: 2Vae: Built in (select none)Iteration steps: 30-50Sampling method: DPM++2M KarrasRecommended resolution (both horizontal and vertical): 1024 × 1024, 1024 × 1280, 768 × 1280, 888 × one thousand two hundred and eighty(Resolution can also be within 1024 × Self adjustable range near 1024)CFG Scale: 7Do not enable high-resolution repair!Do not activate facial repair!1. Solved the low resolution problem of AI photography, 1280 × The 1024 resolution provides clearer and richer details, as well as a finer texture. This provides a higher upper limit for secondary amplification processing through other software after straight out.2. Far superior to SD1.5 in terms of light expression, unparalleled emotional appeal, more natural and realistic expressions, more detailed and rich details, and more exaggerated lens language transmission3. Extremely subtle lens language, even large head shots can have a significant large aperture effect.Please return more pictures for support! Thank you!!!

Files with SHA256: a1c3c7e4ea42abcfc127b70bf4f4c464c9f5ee61fb03b9ee0ce586f11422d5ab