Yuan

<p>描述词请从1girl开始尝试，负面词照搬封面即可，顺便推荐我的lora（封面图用的mflame）</p><p>。。。全身请开adetailer，不要开面部修复</p><p>大家可以参照参考图进行测试</p><p>1.从游戏爱好者/服装收集控的角度出发，优化人物整体动作僵硬问题吗，可以尝试各种角色风格</p><p>2.优化提示词理解（模型装傻）</p><p>3.优化构图结构，解决人物整体动作僵硬问题，</p><p>4.优化手部姿势，对动作&增添画面元素的提示词有更好的响应，较好的lora兼容性</p><p>（强烈推荐老六提示词插件：随机灵感很好用）</p><p>（1）角色概念图</p><p>角色：吸血鬼，德古拉风格</p><p>正面：multi-view,front view,side view,rear view,concept art,character painting,((1boy,male)),dracula costume,black cape,fangs,vampire-themed event,(masterpiece, best quality:1.2),two-dimensional culture,multiple views,reference sheet,gradient background,full body,</p><p>负面：easynegative,badhandv4,(worst quality, low quality:1.4),greyscale,border,artist name,out of frame,holding,zombie,watermark,text,logo,</p><p>（2）角色插画</p><p>吸血鬼，德古拉风格</p><p>正面：((1boy,male)),dracula costume,black cape,fangs,vampire-themed event,(masterpiece, best quality:1.2),</p><p>负面：easynegative,badhandv4,(worst quality, low quality:1.4),greyscale,border,artist name,out of</p><p>frame,holding,zombie,</p><p>美人鱼</p><p>正面：(1girl:0.9),(masterpiece, best quality:1.2),mermaid costume,iridescent scales,seashell bra,underwater-themed event,</p><p>负面：easynegative,badhandv4,(worst quality, low quality:1.4),greyscale,border,artist name,out of frame,holding,zombie,watermark,text,logo,</p><p>公主服, 花瓣裙, 王冠, 皇家舞会</p><p>正面：(1girl:0.9),(masterpiece, best quality:1.2),princess costume,ballgown,tiara,royal ball</p><p>负面：easynegative,badhandv4,(worst quality, low quality:1.4),greyscale,border,artist name,out of frame,holding,zombie,watermark,text,logo,</p><p>海盗</p><p>正面：(1girl:0.9),(masterpiece, best quality:1.2),,eye patch,tricorn hat,pirate-themed party,</p><p>负面：easynegative,badhandv4,(worst quality, low quality:1.4),greyscale,border,artist name,out of frame,holding,zombie,watermark,text,logo,</p><p>q版，可爱风格</p><p>正面((((no_humans)))),(((chibi_cat))),blindbox,cute,3d+你的描述词</p><p>此处为cinematic film still awardwinning photo,rust,epic,viking knotwork,wearing armor and shield,full body . shallow depth of field,vignette,highly detailed,high budget,bokeh . shallow depth of field,vignette,highly detailed,high budget,bokeh,cinemascope,moody,epic,gorgeous,film grain,grainy,</p><p>负面：nsfw,(worst quality:1.4),(low quality:1),(normal quality:2),badhandv4,easynegative,(worst quality, low quality:1.4),[:(badhandv4:1.5):27],(nsfw:1.3),flower,facing viewer,cleavage,</p><p>古装</p><p>（3）单一角色用法（ad这里识别错了），负面描述不变，可以作为参照思路</p><p>(easynegative),badhandv4,(worst quality, low quality:1.4),greyscale,border,artist name,out of frame,holding,zombie,watermark,text,logo,</p><p>负面：easynegative,badhandv4,(worst quality, low quality:1.4),greyscale,border,artist name,out of frame,holding,zombie,watermark,text,logo,</p><p>增加:黑色铠甲+发饰+红色头发</p><p>(1girl:0.9),(masterpiece, best quality:1.2),holding weapon,holding_sword,black armor,hairpin,red hair</p><p>负面：easynegative,badhandv4,(worst quality, low quality:1.4),greyscale,border,artist name,out of frame,holding,zombie,watermark,text,logo,</p><p>改变姿势：战斗姿态</p><p>(1girl:0.9),(masterpiece, best quality:1.2),holding weapon,holding_sword,black armor,hairpin,red hair,fighting_stance,</p><p>负面：easynegative,badhandv4,(worst quality, low quality:1.4),greyscale,border,artist name,out of frame,holding,zombie,watermark,text,logo,</p><p>参考例图结合自己想要的场景构思来增加/删减描述词达到自己想要的效果，不喜欢背景可以用simple_background</p><p>四、推荐参数（把例图丢进去）</p><p>推荐丢图：简单粗暴</p><p>参数设置：（要不咱还是丢图吧，除了ad基本一样的）</p><p>采样步数: 20,</p><p>采样器: euler a,</p><p>cfg scale: 7, （我比较懒所以默认的7，随机性强一些，如果响应不好的话适当调高，推荐7到13）</p><p>推荐分辨率: 768x512, 512x768，1024*512</p><p>vae: vae-ft-mse-840000-ema-pruned（不选会发灰）</p><p>clip skip: 2,</p><p>adetailer model: face_<a target="_blank" rel="ugc" href="http://yolov8n.pt">yolov8n.pt</a>, （不喜欢这个脸可以不开，在ad中套用其他lora，或者自己描述）</p><p>放大倍数：hires upscale: 2.5,</p><p>hires upscaler: 8x_nmkd-superscale_150000_g,</p><p>五、注意事项：</p><p>1.绘制近景，一般情况下没必要用adetailer（ad）</p><p>2.负面描述可以看自己需求增减，推荐负面emb：badhanv4</p><p>3.重绘幅度不是固定的，根据你观察到生图过程中的整体构图和最终成图进行对比，如果生图过程中出现了你想要的构图，但是最终差距很大，请锁定种子，适当降低重绘幅度，如果清晰度较低，请适当提高重绘幅度。</p>