Daily versatility-easy get incredible ComfyUI workflow — v2.1(Llama 3.1 Polish up)

V2版本的变体,仅仅替换了V2工作流的超级提示词和图像反推模型。

现在可以通过Comfyui自带的ollama节点,使用语言大模型(LLM)生成及润色提示词。推荐使用最新的llama3.1,非常强悍,已经接近chatgpt4o,对提示词的拓展描写非常可观。

图像反推功能选择llava模型,比原来的Florence2模型更优,描述更加具体,细节也更多。

ollama是独立于Comfyui运行的,需要到 https://ollama.com/ 下载ollama。安装很简单,安装完毕后,请在键盘输入win+R打开运行,在运行里输入CMD打开命令提示符窗口,之后输入ollama run llama3.1,会自动下载模型。你也可以去上面的网址尝试下载其他模型。模型默认会下载在C盘,需要更改路径,请在环境变量中的系统变量里添加OLLAMA_MODELS,自行设置路径。llava模型同理。

为中国用户准备:工作流相关的部分模型(不包含LLM模型),将models文件夹覆盖ComfyUI文件夹内models文件即可。其他一些模型需要运行工作流自动下载,直接放置模型不管用。

链接: https://pan.baidu.com/s/1qNTKAYIHmyfMf1RsqffitQ?pwd=n4lr

提取码:n4lr

V2

The previous workflow was reconstructed using the [easy use] node to make it more concise and portable in terms of daily versatility. The workflow still includes common functions such as basic textual pictures (intelligent completion of prompt words, compatible with Chinese and English), picture reverse reasoning, controlnet, IPAdapter, detail repair, high-definition repair, and background replacement. It also adds node functions such as portrait master prompt, IPAdapter enhancement, instantID, and PAG attention guidance enrich the use of prompt words and the adjustment of prompt levels. The combination of instantID, controlnet and IPAdapter can change a variety of gameplays. In addition, the option of kolors polishing has been added to the HD repair. Although I don't think the effect is much different from normal HD repair, you can try it.

The overall idea of this workflow is lazy design, which basically covers all aspects of daily drawing requirements, but does not need to keep switching various workflows for certain specific functions. It is suitable for lazy players who have certain hardware configurations and are not satisfied with sd-webui.

This is a super stitching monster. Most of the work is attributed to the omnipotent netizens. I just put it together. There is a lot of use in [easy use], a highly integrated node group. Its author is from China (Bilibili:@ Luananah AI). You can also go to his github homepage: https://github.com/yolain/ComfyUI-Easy-Use .

He even provided a workflow sharing page : https://github.com/yolain/ComfyUI-Yolain-Workflows , from which I benefited a lot.

I heard that he is recently rewriting the [easy use] node group to adapt to the new version of comfyui, and I am very looking forward to it.

Each function of the workflow has node groups with annotations. Right click in the blank space and select [easy use] to enable or ignore node groups and enable them as needed. Before you use it, be sure to go to each management group to check the detailed function description. Some nodes will automatically download models at runtime, while others require you to download and place them manually.

The overall operating logic of the workflow is still divided into a Wensheng diagram process and a background replacement process. If you want to switch from the Wensheng diagram process to the background replacement process, you need to obtain the [Change background_basis] in the node from the HD repair management group. Or [Change background_repair] and connect it to the [picture] of the pre-detailed repair node.

To use the normal Text generates image process, please turn off the relevant node groups for background replacement.

Acquisition nodes are set for the images generated in each link. The first image generated by the [Wensheng Picture] node group is [base image], the image after [Detail Repair] is [base image2], the image after [HD Repair] is [HD image], the image retouched by [kolors Zoom in] is [HD image2], the image after [Background Replacement] is [Change background_basis], and the image after [Background Replacement: Repair] is [Change background_repair]. Please select the correct picture option on the [Get] node according to the switching status of the actual node group and the logic of workflow operation.

When using workflows, please do not open too many node groups at once, which may easily burst video memory. Please operate according to your actual hardware level. If you really need to complete so many functions at one time, you can choose a relay method, complete the previous few steps first, then close some node groups that will not affect subsequent operations according to the actual situation, and copy and paste the pictures into subsequent links to continue completing the task.

————————————————————————————————————————

使用【easy use】节点重构了上一个工作流,使其在日常多功能的特点上,更加简洁、便携。工作流仍然包含了基础文生图(提示词智能补全,兼容中英文)、图片反向推理、controlnet、IPAdapter、细节修复、高清修复、背景替换等常用功能,同时还增加了肖像大师提示、IPAdapter增强、instantID、PAG注意力引导等节点功能,丰富了提示词的使用方式和提示程度的调节,instantID和controlnet及IPAdapter结合,可以变化出多种玩法。此外高清修复那里还增添了kolors润色的选项,虽然我觉得效果与正常高清修复并无太多不同,但你可以尝试。

这个工作流的整体思路是懒人设计,基本上涵盖了日常生图需求的方方面面,但又不需为了某些特定功能,不停去切换各种工作流。它适合那些有一定硬件配置又不满足于sd-webui的懒人玩家。

这是个超级缝合怪,大多数的工作要归功于万能的网友,我只是将其组合在一起。其中大量使用了【easy use】,这是一个高度集成的节点组,它的作者是来自中国的(Bilibili:@乱乱呀AI),你也可以去他的github主页看看: https://github.com/yolain/ComfyUI-Easy-Use 。

他甚至提供了一个工作流的分享页面

https://github.com/yolain/ComfyUI-Yolain-Workflows ,

我从中获益不浅。听说他最近在重写【easy use】节点组,以适应comfyui的新版本,我十分期待。

工作流的各项功能都做了节点组,附带注释。右键空白处,选择【管理组(easy use】,可以启用或忽略节点组,按需启用。在你使用前,请务必去每个管理组旁查阅详细的功能描述。有些节点会在运行时自动下载模型,有些则需要你自行手动下载、放置。

工作流的整体运行逻辑还是分为文生图流程和背景替换流程,如果你要从文生图流程切换到背景替换流程,你需要在高清修复管理组那里,将获取节点内的【Change background_basis】或【Change background_repair】接入预细节修复节点的【图片】。使用正常文生图流程,请关闭背景替换的相关节点组。

每个环节生成的图片都设置了获取节点,【文生图】节点组生成的第一张图片是【base image】,经过【细节修复】的图像是【base image2】,【高清修复】后的图像是【HD image】,通过【kolors放大】润色的图像则是【HD image2】,【背景替换】后的图像是【Change background_basis】,【背景替换:修复】后的图像是【Change background_repair】。请根据实际节点组的开关情况,按照工作流运行的逻辑,在【获取】节点选择正确的图片选项。

使用工作流时,请不要一次性打开过多的节点组,容易爆显存,请按照你的实际硬件水平操作。如果你确实需要一次性完成这么多功能,可以选择接力的方式,先完成前面的几个步骤,再根据实际情况关闭部分不影响后续运行的节点组,将图片复制粘贴到后续环节继续完成任务。

为中国用户准备:工作流相关的部分模型,将models文件夹覆盖ComfyUI文件夹内models文件即可。其他一些模型需要运行工作流自动下载,直接放置模型不管用。

链接: https://pan.baidu.com/s/1qNTKAYIHmyfMf1RsqffitQ?pwd=n4lr

提取码:n4lr

V1

This is a multi-functional daily workflow for your own use, including intelligent completion generation of prompts (compatible with Chinese and English), image reverse reasoning, controlnet, IPAdapter, face repair, hand repair, HD repair, SUPIR magnification and other common functions.

It is actually a super suture monster, most of the work comes from omnipotent netizens, I just put it together.

The overall idea is lazy design, prompt words simple input, can be used to achieve good results, but also leave enough room for users who like to get "inspiration" from other pictures.If the firepower is full, you can achieve the stunning ultra-high definition picture.

All the functions are made into node groups, with comments. Right-click in the blank, and select 【Management group(easy use)】. You can enable or ignore node groups, and enable them as needed.

In addition to the normal process of creating a picture, the workflow is accompanied by a process of changing the background. If you need to change the background, please go to the high-definition amplification node group and switch [get note] from BASE_IMAGE to BASE_IMAGE2.Also switch the [get note] of the color node of the picture.

You can first use the normal process to generate satisfactory images, and then fix the threshold value to start the background change process.

————————————————————————————————

这是自用的多功能日常工作流,包含了提示词智能补全生成(兼容中英文)、图片反向推理、controlnet、IPAdapter、脸部修复、手部修复、高清修复、SUPIR放大等常用功能。

它其实是个超级缝合怪,大多数的工作来自于万能的网友,我只是将其组合在一起。

整体思路是懒人设计,提示词简单输入,即可达成不俗的效果,同时也给喜欢从其他图片获取“灵感”的用户留足了空间。火力全开的话,可以达到极致惊艳的超高清大图。

各项功能都做了节点组,附带注释。右键空白处,选择【管理组(easy use】,可以启用或忽略节点组,按需启用。

在正常文生图的流程之外,工作流还附有更换背景的流程,如果需要更换背景,请先前往高清放大节点组将【get note】从BASE_IMAGE切换为BASE_IMAGE2。再同样切换图片调色节点的【get note】。

可以先用正常流程生成满意的图片,再固定seed值后开启更换背景流程。