Must-Have ComfyUI Custom Nodes (2026): rgthree, VideoHelperSuite, KJNodes, Crystools and More

A fresh ComfyUI install can already run Qwen-Image, Wan 2.2, LTX-2.5 and MiniMax H3 with native nodes. But open almost any workflow from Civitai, Hugging Face or a YouTube tutorial and you will hit a wall of red “missing node” boxes. A small set of custom node packs shows up in nearly every serious workflow: rgthree-comfy, VideoHelperSuite, KJNodes, Crystools, ComfyUI-GGUF and a few more. This is my list of the must-have ComfyUI custom nodes for November 2026, what each one actually does, the exact install steps for Manager, Windows portable and Linux, and which old packs you can now skip.

How to Install Any Custom Node Pack

There are three ways, and all of them end with the pack in ComfyUI/custom_nodes/<pack-name>/.

Option 1: ComfyUI-Manager (easiest)

1. Click Manager (or the puzzle-piece icon) in ComfyUI and open Custom Nodes Manager.
2. Search the pack name, check the author matches the one in this guide, and click Install.
3. Restart ComfyUI when asked, then refresh the browser tab.

When you load a workflow with red nodes, Install Missing Custom Nodes finds and installs the packs it needs in one go.

Option 2: Windows portable, manual

Open a terminal in the folder that contains run_nvidia_gpu.bat:

cd ComfyUI\custom_nodes
git clone https://github.com/rgthree/rgthree-comfy
cd ..\..
.\python_embeded\python.exe -s -m pip install -r .\ComfyUI\custom_nodes\rgthree-comfy\requirements.txt

The important part is .\python_embeded\python.exe. Installing requirements with your system Python instead is the most common reason for “IMPORT FAILED” on portable installs. (rgthree has no heavy requirements, but the same pattern applies to every pack.)

Option 3: Linux, macOS or a manual venv

cd ComfyUI/custom_nodes
git clone https://github.com/kijai/ComfyUI-KJNodes
source ../venv/bin/activate
pip install -r ComfyUI-KJNodes/requirements.txt

Comfy Desktop users can use the built-in Manager, or open the Desktop’s terminal (it activates the right Python for you) and run the same git clone and pip install commands inside <Desktop base folder>/custom_nodes.

ComfyUI-Manager (Now Built In)

The old way was to git-clone ComfyUI-Manager as a custom node. The new ComfyUI-Manager is built into ComfyUI core but needs to be switched on. According to the official docs:

• Desktop: installed and enabled with every ComfyUI install. Settings live in the Manage panel under Startup Args.
• Windows portable:

  .\python_embeded\python.exe -m pip install -r ComfyUI\manager_requirements.txt
  .\python_embeded\python.exe -s ComfyUI\main.py --windows-standalone-build --enable-manager

  Add --enable-manager to your run_nvidia_gpu.bat so it is always on.

• Manual install: pip install -r manager_requirements.txt, then python main.py --enable-manager.
• Old interface: add --enable-manager-legacy-ui. To keep Manager’s background features but hide its UI: --disable-manager-ui.

If you still have an old git-cloned copy, it must sit at exactly ComfyUI/custom_nodes/comfyui-manager (not nested like ComfyUI-Manager/ComfyUI-Manager), and that copy cannot switch to the new UI. Manager also has a command line tool, cm-cli, that installs and updates packs without starting ComfyUI. More launch flags are in my ComfyUI command line arguments guide.

Quick List: Must-Have Packs

rgthree-comfy

If you install only one pack, make it this one. It is in a huge share of shared workflows, and it makes big graphs far easier to manage. The nodes I use every day:

• Power Lora Loader: load any number of LoRAs in one compact node with an on/off toggle and strength per LoRA. It replaces chains of LoraLoader nodes (the old Lora Loader Stack is deprecated). Essential for Wan 2.2 LoRA stacks, where you need one per expert.
• Fast Groups Bypasser / Fast Groups Muter: one panel that lists every group in the workflow with a toggle. Turn the “upscale” or “interpolate” group off while you test prompts, and on for the final render.
• Seed: random or fixed seed with buttons to randomise or reuse the last seed.
• Image Comparer: drag a slider between two images, perfect for before/after upscales and LoRA tests.
• Display Any: shows any value (text, numbers, prompt from an LLM node) after a run.
• Context / Context Switch, Any Switch, Power Primitive, Power Puter, Label, Bookmark: wiring and layout helpers.
• Extras: a progress bar along the top of the window, “Queue Selected Output Nodes” in the right-click menu, and a Link Fixer at http://127.0.0.1:8188/rgthree/link_fixer for repairing broken workflow links.

Install: Manager search “rgthree”, or git clone https://github.com/rgthree/rgthree-comfy.git in ComfyUI/custom_nodes. Settings live in rgthree_config.json (copy it from rgthree_config.json.default).

ComfyUI-VideoHelperSuite

ComfyUI now has native video load and save nodes, but VideoHelperSuite (VHS) is still what most video workflows use, because it has more control:

• Load Video (Upload / Path): turns a video into frames with force_rate, force_size, frame_load_cap, skip_first_frames and select_every_nth. Great for V2V, upscaling and interpolation.
• Load Image Sequence: loads a folder of frames.
• Video Combine: saves frames as H.264/H.265 MP4, WebM, ProRes, GIF and more, with optional audio, loop count, ping-pong and frame rate. Formats are defined by JSON files in its video_formats folder, so you can add your own ffmpeg settings.
• Load Audio, plus batch helpers: Split Batch, Merge Batch, Select Every Nth, Get Count, Duplicate Batch.

VHS uses ffmpeg for encoding. It finds a system ffmpeg on your PATH, or the imageio-ffmpeg Python package. If Video Combine shows only GIF/WebP formats, ffmpeg was not found: install it (on Windows, winget install ffmpeg) and restart. I use VHS in all my MiniMax H3 and Wan workflows, for example the H3 RIFE interpolation workflow.

ComfyUI-KJNodes

kijai’s KJNodes is a toolbox of small nodes that fill gaps in core ComfyUI. Highlights:

• Set / Get nodes: send a value anywhere in the graph without dragging a wire across the screen. They make large workflows readable.
• Resize Image v2, Image Concatenate, Color Match: resize to multiples of 16 or 32 for video models, place images side by side, and match colours between clips when you join them.
• Patch Sage Attention KJ and torch.compile nodes: speed up Wan, LTX and H3 sampling if you have SageAttention or Triton installed.
• Points Editor, mask and grow-mask tools, string and math helpers, VRAM debug nodes.

Many of kijai’s other packs (WanVideoWrapper, Florence2, MMAudio) and lots of community workflows depend on KJNodes, so it is worth installing even if you never add its nodes yourself.

ComfyUI-Crystools

Crystools puts a live resource monitor in the ComfyUI menu bar: CPU, GPU usage, RAM, VRAM, GPU temperature and disk space, with a configurable refresh rate. When you are tuning a 16 GB GPU for a video model, seeing VRAM climb in real time is far more useful than guessing. It also adds a progress bar with elapsed time, metadata nodes (read, compare and save workflow metadata), Show Any / Show Any to JSON debugging nodes, and Load Image with Metadata.

Notes: GPU metrics need an NVIDIA card (there is a separate AMD branch for Linux that should be cloned manually, not through Manager). Its dependencies are psutil, torch and pynvml. The last changelog entry I can see is 1.27.0 from August 2025, so it is mature rather than fast-moving, but it still works with current ComfyUI. With two GPUs, it shows each card, which is how I watch both RTX 5080s on my machine.

ComfyUI-GGUF

city96’s ComfyUI-GGUF lets ComfyUI load GGUF versions of diffusion models and text encoders, so 20-40 GB models fit on 8, 12 and 16 GB GPUs. It adds Unet Loader (GGUF) and GGUF CLIP loaders under the bootleg category. GGUF diffusion files go in ComfyUI/models/unet/ (or diffusion_models/) and GGUF text encoders in ComfyUI/models/text_encoders/. I have a complete ComfyUI-GGUF guide with fp8 vs GGUF advice and fixes.

Video Packs: Frame Interpolation, SeedVR2, WanVideoWrapper

• ComfyUI-Frame-Interpolation (Fannovel16): RIFE and FILM nodes that double or triple the frame rate. RIFE weights download automatically into ComfyUI/custom_nodes/ComfyUI-Frame-Interpolation/ckpts/rife/. Turns a 16-24 fps AI clip into smooth 48-72 fps.
• SeedVR2 Video Upscaler (numz): one-step diffusion upscaler for video and images. Models download to ComfyUI/models/SEEDVR2/. See my 1080p 60 fps RIFE + SeedVR2 workflow.
• ComfyUI-WanVideoWrapper (kijai): an alternative Wan pipeline with block swap, context windows and early support for new Wan research. Native Wan nodes are simpler for most people; use the wrapper when a workflow asks for it.
• ComfyUI-LTXVideo (Lightricks): official extra nodes for LTX-2.x; see my LTX-2.5 setup.

Image Packs: Impact, ControlNet Aux, RMBG, Ultimate SD Upscale

• ComfyUI-Impact-Pack (ltdrdata): Face Detailer, SEGS detection and masking, iterative upscale. Version 8.28.3 shipped in April 2026. Some detectors live in the separate Impact-Subpack; install both if a workflow needs UltralyticsDetectorProvider.
• comfyui_controlnet_aux (Fannovel16): depth, OpenPose/DWPose, canny, lineart, normal and more, plus an AIO preprocessor. Weights download on first use of each preprocessor. Also needed for Wan Fun Control and many pose workflows.
• ComfyUI-RMBG (1038lab): background removal with several models, including video matting. My RMBG 3.3 guide covers the latest version.
• ComfyUI_UltimateSDUpscale (ssitu): tiled upscaling for images that are too big to sample in one go.
• ComfyUI-Inpaint-Nodes (Acly): better inpainting, Fooocus inpaint and LaMa fill.

Workflow Helpers: Easy-Use, Custom-Scripts, MultiGPU

• ComfyUI-Easy-Use (yolain): bundled “easy” nodes that combine loader, prompt and sampler steps, plus handy switches and XY plots. Many Chinese and Civitai workflows use it.
• ComfyUI-Custom-Scripts (pysssss): autocomplete for embeddings and LoRAs in prompts, image feed, workflow snapping and other quality-of-life tweaks. Check which features are still needed, because the new frontend has absorbed some of them.
• ComfyUI-MultiGPU (pollockjj): put the text encoder or VAE on a second GPU, or spread a model’s layers over a second GPU and system RAM. Very useful on two-card machines like mine (2x 16 GB). Its GGUF loaders let a big video model use both cards.

Packs You Can Skip in 2026

• A git-cloned ComfyUI-Manager on new installs: use the built-in Manager with --enable-manager.
• WAS Node Suite: once everywhere, now archived. Only install it if an old workflow needs it.
• AnimateDiff-Evolved and IPAdapter Plus for new projects: both are tied to the SD 1.5/SDXL era. For animation use Wan, LTX or MiniMax H3; for reference images use the edit models’ native multi-image input, such as Qwen-Image 2.1 Edit.
• Separate “save video” or “preview video” packs: core ComfyUI and VHS cover this.
• Five different utility packs that do the same thing: every pack adds startup time and possible conflicts. If rgthree and KJNodes cover a job, skip the duplicate.

Custom Node Safety

A custom node is Python code that runs with your user’s full permissions. There have been malicious packs in the past. Simple rules:

• Install from the Manager / Comfy Registry or from the author’s real GitHub, and check the author name (for example kijai, rgthree, Kosinkadink, city96).
• Be wary of brand-new packs with no stars that a workflow suddenly “requires”.
• Keep Manager’s security level at the default and update ComfyUI regularly.
• Back up your custom_nodes folder (or use Manager’s snapshot feature) before big updates, so you can roll back.

Fix “Missing Nodes” and Import Failed

• Red nodes after loading a workflow: Manager → Install Missing Custom Nodes, then restart. If a node is still red, the pack may have renamed it; update the pack.
• “(IMPORT FAILED)” in the console: scroll up to the first error. It is almost always a missing Python package. Install the pack’s requirements.txt into ComfyUI’s Python (python_embeded\python.exe on portable).
• Two packs fight over a dependency: one pack pins an old version of a library (often numpy, opencv or transformers). Update both packs; if it persists, disable the less important one.
• ComfyUI will not start after installing a pack: rename the new folder in custom_nodes (add .disabled) and start again. Then fix or remove it.
• Stuck on “Reconnecting” after installing nodes: a pack crashed the server. See ComfyUI Reconnecting fixes. More errors: ComfyUI errors and fixes.

Frequently Asked Questions

What custom nodes should I install first in ComfyUI?

rgthree-comfy, ComfyUI-VideoHelperSuite, ComfyUI-KJNodes, ComfyUI-Crystools and ComfyUI-GGUF. Together they cover LoRA loading, video input and output, utility nodes, resource monitoring and low-VRAM models.

Where are custom nodes installed in ComfyUI?

Each pack goes in its own folder inside ComfyUI/custom_nodes/, for example ComfyUI/custom_nodes/rgthree-comfy/. On Comfy Desktop the folder is custom_nodes inside your Desktop base folder.

Is ComfyUI-Manager built into ComfyUI now?

Yes. The new ComfyUI-Manager is part of ComfyUI core. Desktop enables it automatically; on portable and manual installs, install manager_requirements.txt and start ComfyUI with –enable-manager.

What does rgthree-comfy do?

It adds quality-of-life nodes such as Power Lora Loader, Fast Groups Bypasser and Muter, Seed, Image Comparer, Display Any, context and switch nodes, plus a progress bar and a link fixer.

What is ComfyUI-VideoHelperSuite used for?

Loading videos and image sequences as frames and saving frames as MP4, WebM, GIF and other formats with audio, using ffmpeg. Most video workflows use its Load Video and Video Combine nodes.

Why does Crystools not show my GPU?

Crystools reads GPU stats through NVIDIA’s NVML, so it needs an NVIDIA GPU and the pynvml package. AMD users on Linux need its separate AMD branch.

How do I fix IMPORT FAILED for a custom node?

Read the first error above the IMPORT FAILED line, then install the pack’s requirements.txt with ComfyUI’s own Python, for example python_embeded\python.exe -m pip install -r requirements.txt on Windows portable, and restart.

Final Thoughts

You do not need fifty custom node packs. Enable the built-in Manager, install rgthree-comfy, VideoHelperSuite, KJNodes, Crystools and ComfyUI-GGUF, and add the video or image packs only when you start that kind of work. Your ComfyUI will start faster, break less after updates and still run almost every workflow you download.

Next: learn where every model file goes with the ComfyUI model folders cheat sheet, pick sampler settings with samplers and schedulers explained, or put the packs to work with my best Wan 2.2 LoRAs guide.

Sources

• ComfyUI docs: Install ComfyUI-Manager (built-in, –enable-manager)
• Comfy-Org/ComfyUI-Manager
• rgthree/rgthree-comfy
• Kosinkadink/ComfyUI-VideoHelperSuite
• kijai/ComfyUI-KJNodes
• crystian/ComfyUI-Crystools
• city96/ComfyUI-GGUF
• ltdrdata/ComfyUI-Impact-Pack, Fannovel16/comfyui_controlnet_aux, Fannovel16/ComfyUI-Frame-Interpolation
• ComfyUI Custom Nodes Guide 2026 (DEV Community)