I found a new AI tool Shakker, a best image to image tool. You can try it via <a target="_blank" rel="ugc" href="https://www.shakker.ai">https://www.shakker.ai</a> SDXL 的 LCM2 的基础模型，做了一些调整和优化，所以不适用LCM采样器，对SDE类别的采样器十分友好。<hr />如果您不阅读繁琐的文档，那么这里推荐的配置如下：采样器：DPM++ SDE KarrasCFG：1.5-2.5步数：5<hr /><h3 id="heading-1525">采样器推荐</h3>针对SDE采样器比较友好，推荐采样器如下<ol><li>DPM++ SDE Karras - 速度慢，细节更完美（5步+）</li><li>DPM++ 2M SDE Karras - 速度快（5步+）</li><li>DPM++ SDE - 适合高步数(8步+)，速度慢，细节更完美</li></ol><hr /><h3 id="heading-1526">CFG 推荐</h3>CFG通常设置为1即可，可以调整为1-3，值越大，收拢速度越快，也就是可以使用更少的步数完成图片的生成。<hr /><h3 id="heading-140">测试用例</h3>Prompt: (Well-bred's works:1.13),a beautiful girl, 24yo, skinny, pretty, [k-pop:detailed eyes, detailed beautiful face:0.5],Negative Prompt: ng_deepnegative_v1_75t,negative_hand,disfigured,ugly,deformed,worst quality,low quality,extra limbs,extra legs,extra arms,deformed hands,deformed anatomy,umbrella,采样：DPM++ 2M SDE Karras 或者 DPM++ SDE Karras步骤：5+尺寸：768 x 1024 以上<hr /><h3 id="heading-1527">采样器测试</h3>以5步为例，不同采样器效果如下<img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/b52f601c-4fae-4773-a658-51fc55225c3f/width=525/b52f601c-4fae-4773-a658-51fc55225c3f.jpeg" />以10步为例，不同采样器效果如下<img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/faf4d532-4b92-45bc-962d-e05468fc431f/width=525/faf4d532-4b92-45bc-962d-e05468fc431f.jpeg" />以20步为例，不同采样器效果如下<img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/ff1ec001-10a0-4e2d-aec5-b618fc434126/width=525/ff1ec001-10a0-4e2d-aec5-b618fc434126.jpeg" /><hr /><h3 id="heading-1528">DPM++ 2M SDE Karras 步数测试</h3>DPM++ 2M SDE Karras比较适合快速抽卡，从1步到8步的效果如下<img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/f7562aa5-756e-4aae-80eb-1f1289fe9844/width=525/f7562aa5-756e-4aae-80eb-1f1289fe9844.jpeg" />第4步就形成了可用的图像，随着步数的增加，细节逐渐完善。更多测试报告:<a target="_blank" rel="ugc" href="https://www.yuque.com/mengxiangyuan-7fhri/eo7oa2/gtifyimxamueeixg?singleDoc#">https://www.yuque.com/mengxiangyuan-7fhri/eo7oa2/gtifyimxamueeixg?singleDoc#</a> 《SDXL-LCM采样器测试报告》<hr />My Github:<a target="_blank" rel="ugc" href="https://github.com/SoftMeng/ComfyUI_Mexx_Styler">https://github.com/SoftMeng/ComfyUI_Mexx_Styler</a><a target="_blank" rel="ugc" href="https://github.com/SoftMeng/ComfyUI_Mexx_Poster">https://github.com/SoftMeng/ComfyUI_Mexx_Poster</a><a target="_blank" rel="ugc" href="https://github.com/SoftMeng/ComfyUI_Mexx_Image_Mask">https://github.com/SoftMeng/ComfyUI_Mexx_Image_Mask</a><a target="_blank" rel="ugc" href="https://github.com/SoftMeng/stable-diffusion-prompt-pai">https://github.com/SoftMeng/stable-diffusion-prompt-pai</a><a target="_blank" rel="ugc" href="https://github.com/SoftMeng/Fooocus-zh">https://github.com/SoftMeng/Fooocus-zh</a>Or My LibLib:<a target="_blank" rel="ugc" href="https://www.liblib.art/userpage/0f37258e861044879b8728309c0ac3b0/publish">https://www.liblib.art/userpage/0f37258e861044879b8728309c0ac3b0/publish</a>

MexxL_LCM2

mexxlLCM2_yy.safetensors

v1.0

mexXL_lcm2

The YY version is a fine-tuned version (recommended).

Files with SHA256: 1f6206dd7a50859b64ed3f8ed8488a861f794a9a718d586aac494e47be98d8d7