ComfyUI-Detail-Daemon细节增强
🌹大家好!欢迎来到破狼公众号。感谢大家的支持与鼓励。在AIGC探索道路上,我将与你一路同行。喜欢就星标关注破狼公众号或文末扫码加入交流群 !本人仅运营公众号平台,未经授权严禁CSDN等其他平台抄袭和转载!
ComfyUI-Detail-Daemon简介
ComfyUI-Detail-Daemon是一款来自SDWebUI插件sd-webui-Detail-Daemon
(https://github.com/muerrilla/sd-webui-detail-daemon/)的ComfyUI移植插件,它能够被应用于 调整 sigmas来增强图像细节,并去除不需要的背景模糊或光晕效果 。特别是在使用 Flux-DEV 模型时的模糊背景图像质量(同样适用于 SDXL、SD1.5,以及可能的其他模型)。另外作者提醒到:如果调整的值过于极端,可能会导致过度锐化和/或 HDR 效果。ComfyUI-Detail-Daemon插件一共包含了四个节点:Detail Daemon Sampler、Detail Daemon Graph Sigmas、Multiply Sigmas、Lying Sigma Sampler。分别对应功能如下:
• Detail Daemon Sampler :该节点允许使用 Detail Daemon 计划调整进行采样,它保持注入的噪声水平不变,同时减少每一步移除的噪声量,从而有效地增加了细节。
Detail_amounts
在 0 到 1.0 之间效果最佳。• Detail Daemon Graph Sigmas:节点允许将调整后的 sigmas 以图形方式展示,直观地看到不同参数对曲线的影响。必须与 Detail Daemon Sampler 节点分开。只需将值设置得与那个节点相同,或者在不同的原始节点上设置输入,这些输入既输入到 Detail Daemon Sampler 节点也输入到这个 Graph Sigmas 节点。注意:该节点实际上并不改变生成时使用的 sigmas,它只绘制图像报告。
• Multiply Sigmas:这是一个简单的节点,通过提供的因子乘以所有 sigmas(同时乘以添加的噪声水平和去噪的因子,这以某种方式通过小于 1 的因子增加了细节)。因子值在 0.95-0.99 之间效果最佳(默认值是 1.0)。它是无状态的,意味着它在每次队列计算 sigmas(其他乘以 sigmas 节点似乎在先前运行的 sigmas 上计算)。因为这个节点乘以所有步骤的 sigmas(没有开始或结束值),它将倾向于改变图像的整体构成,变化结构最大。
• Lying Sigma Sampler:Detail Daemon Sampler 的一个更简单的版本,只有量调整(-0.05 的dishonesty_factor相当于 Detail Daemon 中的 0.5 细节量),开始和结束值。dishonesty_factor值在 -0.1 和 -0.01 之间效果最佳。
ComfyUI-Detail-Daemon ComfyUI体验
需要更新ComfyUI并插件管理器搜索ComfyUI-Detail-Daemon和安装该插件。该插件无需下载其他模型导入工作流即可使用。工作流可LIBLIB平台下载地址:https://www.liblib.art/modelinfo/d182bb6df6fa4e17960852d7723be1d1?versionUuid=d73fb09a54a04e8c87d722bee9edf2d0
• ComfyUI-Detail-Daemon:https://github.com/Jonseed/ComfyUI-Detail-Daemon
Flux文生图工作流
Flux文生图感兴趣的同学可参考LIBLIB在线运行工作流:FLUX[续篇]:12B参数23G最大开源文生图模型,Dev版直出惊艳美图欣赏
ComfyUI-Detail-Daemon工作流
ComfyUI-Detail-Daemon工作流可LIBLIB平台下载地址:https://www.liblib.art/modelinfo/d182bb6df6fa4e17960852d7723be1d1?versionUuid=d73fb09a54a04e8c87d722bee9edf2d0
注意:工作流采用了在同一个提示和种子上比较ComfyUI-Detail-Daemon插件节点的不同的细节增强体验。
01.人物
pl, close-up, of an Asian woman seated in a train. The photograph features a young, East Asian woman with long, straight, dark brown hair and fair skin. She has a slender yet curvy physique with notably large breasts. She wears a tight-fitting, white, short-sleeved crop top that reveals her midriff, highlighting her ample cleavage. Her attire also includes light blue denim shorts. Her expression is neutral, with slightly parted lips and a subtle, almost wistful gaze towards the left, out of frame. The interior of the train is modern and sleek, with blue upholstered seats and large windows offering a view of an overcast day with distant bridges and railway tracks visible through the glass. The lighting is soft, diffused by the train's windows, creating a serene ambiance.
默认文生图 | Detail Daemon |
Lying Sigma | Multiply Sigmas |
Detail Daemon 有较大的背景变化(多了身后人物、窗外细节);Lying Sigma 较小的背景变化(主要为窗外的树);Multiply Sigmas则人物和背景都有较大的变化,类似重新生成。
02.艺术雕像
sfw, a digitally created artwork. The image features a stylized, three-dimensional sculpture of a woman's head and shoulders, with intricate, twisting branches of a tree emerging from her neck and flowing down her body. The sculpture is made of a dark, metallic material, likely bronze or copper, giving it a polished and slightly shiny texture. The branches are densely covered with lush, vibrant green leaves, creating a harmonious contrast with the metallic sculpture. The leaves are detailed and realistic, with varying shades of green, some glossy and others matte, adding depth and texture to the overall composition. The background is a stark, clean white, which emphasizes the sculpture and branches, making them stand out prominently. To the right side of the sculpture, seamlessly integrated into the artwork, enhancing the theme of nature and human connection. The overall style is a blend of realism and surrealism, with a focus on natural elements and human form.
默认文生图 | Detail Daemon |
Lying Sigma | Multiply Sigmas |
背景和主体变化同样情况,变化幅度Multiply Sigmas > Detail Daemon > Lying Sigma 。
最后,在实际使用中请根据实际需要细节变化幅度调节对应参数计划值,调整 sigmas来增强图像细节,并去除不需要的背景模糊或光晕效果。如果调整的值过于极端,可能会导致过度锐化和/或 HDR 效果。
更多推荐文章:
• [ComfyUI]PyramidFlow:快手开源视频模型,与Mochi比拼谁更强?8G可运行10秒768P与24帧视频生成
• [ComfyUI]Mochi:可商用!迄今最大开源视频生成模型,100亿参数&高保真动作&高提示遵循
• [ComfyUI]阿里EcomID:最强写真换脸?2百万淘宝图训练,结合PuLID和InstantID最新面部一致性框架
• [ComfyUI]Flux:商用级AIGC!工作流商用封装变现!0代码一键转换商用网站实战
• [ComfyUI]F.1-Lite:F.1轻量级开源模型发布!bf16全精度&80亿参数&节省7GRAM&提速23%
• [ComfyUI]Flux:太赞了!DGirl亚洲写实F.1再现,色彩调节与写实摄影
• [ComfyUI]Flux:超强反推Joy2!该升级你的反推插件了,多样风格和详细提示操作指南
• [ComfyUI]Flux:3倍加速神器!阿里开源F.1加速器Turbo,生态里程碑加速度
• [ComfyUI]Flux:太酷啦!F.1创意小红书写真贴纸艺术风,字节PuLID和创意模板无限创意
• [ComfyUI]Flux:太强了!任意扩图神器,小红书极致逼真风格出游打卡写实风
• 海螺AI重大里程碑升级-图生视频,Flux联手打造自媒体视频新篇章
• [ComfyUI]Flux:轻松一键搞定!最新PuLID-Flux和最强Joy反推实现私人写真自由
• [ComfyUI]Upscaler:超清分辨率!Flux专属高清放大CN模型,F1生态里程碑