This article mainly lists several ComfyUI plugins that I commonly use and recommend.
These plugins are all very powerful, so it is difficult to cover every single function they offer. Here, I mainly focus on the core nodes that are commonly used, while nodes with overlapping functions or those that are not used often are not covered in detail. If you are interested, you can explore them further on your own.
1.ComfyUI-Easy-Use
This plugin integrates many functions and can make your workflow much cleaner. Below are several nodes that I commonly use
EasyLoader (Full)
This directly integrates the model, clip, and vae into a single node for loading, and also provides several commonly used sizes for quick switching.
It also supports connecting to lora stack and controlnet stack for quick use.EasyKSampler (Full)
EasyKSampler (Full) is paired with EasyLoader (Full), and you can simply use it as a regular KSampler.Text Index Switch
It allows you to quickly switch between multiple text inputs. It is recommended to use it together with the Text String node from the was-ns plugin for fast character switching. Many models usually come with a large number of built-in characters, so this makes switching much faster.
2.ComfyUI LoRA Manager
This plugin is used by many people. I have now switched my LORA management to this plugin, as it allows me to quickly look up a LORA’s Civitai page and trigger words, making it very convenient to use. The recommended node connection method is shown in the image below.
Separate the LORA into Style, Character, Clothes, and Action, then connect them together. Organizing the LORA this way makes them much easier to manage. For the conditioning, use the “Conditioning (Combine)” node to connect them.
If it's the portable version, remember to enable portable mode so it will be easier to back up.
Also, using the portable version is more convenient for users like me who run two different versions of ComfyUI at the same time. I use one for running WAN2.2 and the other for running the Anima model.
It is recommended to place the downloaded image data in a separate folder and compress that entire folder when making a backup. When restoring the backup later, transferring a compressed archive will be faster, since transferring a large number of small files is usually much slower.
3.ComfyUI-RMBG
This is the best choice for removing image backgrounds. For the best results, you should use RMBG 2.0 as the background removal model.
4.ComfyUI-WildPromptor
Using WildCard in an integrated way makes it easier to combine multiple wildcards together. These wildcards are usually used for switching things like camera angle, brightness, actions, and characters, while also helping structure the prompt.
Please place your wildcard folder in the following directory.
Please name the txt files in the folder according to its naming convention.
...\ComfyUI_windows_portable\ComfyUI\custom_nodes\ComfyUI-WildPromptor\dataPlease edit the config.json file in the following plugin directory and keep only the folders you need.
5.RES4LYF
Beginners are generally recommended to install this plugin. After installing it, you will have many more Sampler and Scheduler options available to use.
Please look for the " intro to clownsampling " workflow in the plugin directory. It contains the author's detailed feature introduction.
...\custom_nodes\RES4LYF\example_workflows6.ComfyUI-KJNodes
A must-have for beginners, as it includes a large number of very useful features.
Below are several commonly used nodes.
Resize Image v2
Patch Sage Attention KJ
GGUFLoaderKJ
Lazy Switch KJ
Load Images From Folder (KJ)
7.rgthree-comfy
It is mainly used with the “Fast Groups Bypasser (rgthree)” node to quickly toggle node groups on and off.
The "Image Comparer (rgthree)" node allows you to compare changes in images more clearly.
8.ComfyUI-Impact-Pack
This is the plugin used for fixing eyes and hands in images. Below, I list the parameters I use with images.
Eye correction
If you want to fix the face, remember to use “DWPose” to detect facial expressions, and then use it together with openpose to prevent the facial expression from drifting.
“tiled encode” and “tiled decode” do not need to be enabled.
Hand correction
Remember to use it together with ControlNet’s openpose to prevent the pose from shifting. Also, because the hand area is usually relatively large, the mask is made larger to avoid inaccurate hand detection.
“tiled encode” and “tiled decode” do not need to be enabled.
9.ComfyUI-Impact-Subpack
An extension pack for ComfyUI-Impact-Pack
10.ComfyUI-Manager
A plugin for managing ComfyUI plugins, and it is a must-have.
11.ComfyUI_LayerStyle
A very powerful image editing and compositing tool, suitable for users with a professional photo retouching background.
It can adjust color temperature, brightness, contrast, and saturation, and also create HDR effects, common blur effects, and image compositing,
as well as simulate film grain, vignette, and blurred edges.
There is one more node worth specifically mentioning here: “LayerUtility: Purge VRAM.” It is used to clear VRAM. This is useful because ComfyUI currently tends to keep models loaded in VRAM after a workflow finishes, making repeated runs faster by avoiding reloading the model.
However, when using an LLM, or in workflows that use a lot of VRAM such as generating a video after generating an image, keeping the model loaded can actually become a disadvantage. It may keep occupying VRAM and trigger memory swapping, which can make generation much slower.
Please open Task Manager and check your GPU’s idle VRAM usage. In my case, even with two different versions of ComfyUI open at the same time, neither of them running, and a YouTube video playing in the background, the VRAM usage is only about 1.7GB. Please check whether your VRAM is actually being released after execution finishes.
12.ComfyUI-WD14-Tagger
Used for analyzing image tags.
13.comfyui_controlnet_aux
A must-have plugin for using ControlNet, as it includes all the nodes required for ControlNet processing.
14.ComfyUI_UltimateSDUpscale
A plugin used for image upscaling and repair, usually for refining and enhancing image details.
This is usually used together with ControlNet Tile, because repairing image details requires reverse inference on the image, and ControlNet Tile is the most suitable choice for that purpose. At the same time, it can also help prevent strange artifacts from appearing after upscaling. Below, I will show how I use it with images.
Please note that when performing upscaling repair, you need to pass all prompts and LoRAs into Ultimate SD Upscale so that it knows all the parameters used when the image was generated. This will produce the best results.
Please use getNode to load the model or POS_CON into the upscaler to ensure they are loaded correctly. When using non-native nodes, strange bugs can easily occur.
Please note that if you notice overlapping artifacts after execution, change the “seam fix mode” value in the Ultimate SD Upscale node to None.
Also, increase the mask blur and try again until the overlapping artifacts are gone.
Upscaling workflow reference:
Workflow For Upscaling with Upscale LoRA
LoRA and models used:
WAI-illustrious-Rectified-4Steps
15.comfy-image-saver(Deprecated)
Used for saving image parameters.
Save the image parameters so they can be used as a reference in the future.
Use "Save Image (LoraManager)" in ComfyUI-Lora-Manager to save images.
16.comfyui-prompt-reader-node
Read the saved image information. Please note that only images saved with comfy-image-saver will contain this data.
17.ComfyUI-WanVideoWrapper
This is a required installation for generating videos with Wan models, and all the necessary functions are already included in it.
18.ComfyUI-GIMM-VFI
Used for frame interpolation of the AI-generated videos.
19.ComfyUI-VideoHelperSuite
Used to combine sequential images generated for a video into a video.
postscript:
I use ComfyUI mainly just to generate images and make some simple videos for my own enjoyment, and I do not do highly professional workflows. So I use very few nodes, and I do not know how to use overly advanced features.
I also wrote this in the hope that it can help some beginners who are just starting out. Honestly, it would have been great if someone had told me from the beginning what I needed to install. After all, once you get deeper into ComfyUI and install a lot of plugins, the code dependencies inside them can start conflicting with each other.
Especially since ComfyUI is constantly being updated, conflicts with older plugins become more likely over time.
So after installing the plugins and confirming that everything works properly, it is best to lock the versions of ComfyUI and ComfyUI Frontend to avoid bugs. At the same time, export the versions of the Python packages currently in use as well, so it will be easier to track down bugs later.
Finally, I’d also like to briefly promote the Anima assistant. You only need to provide a composition request or upload a composition reference image, and it can generate a prompt for you. At the moment, it does not yet include Danbooru character data, and I wrote the system prompt in the article “Use ComfyUI-QwenVL Notes.” Below is the ChatGPT GPT link. It is currently accessible only via direct link and has not been made public yet.
https://chatgpt.com/g/g-69be3e48b0c081919dd24962ac75fec4-anima-ti-shi-ci-zhu-shou