How to Fix ComfyUI Plugin Import Failed Errors

If ComfyUI starts but your custom nodes do not load, your environment is already telling you something important: the plugin was downloaded, but Python could not import it safely.

That usually means one of four things:

• A required Python package is missing
• A plugin installed an incompatible dependency
• Torch or another core package was downgraded
• The plugin repo is stale, moved, or only partially installed

The good news is that plugin import failed does not always mean you must reinstall ComfyUI. In most cases, you can classify the failure, repair the environment, and keep your existing workflows and models.

What "plugin import failed" usually looks like

You will usually see one or more of these symptoms:

• IMPORT FAILED in the startup log
• A custom node folder exists, but none of its nodes appear in ComfyUI
• The workflow opens with red nodes after restart
• A node worked before, then stopped loading after another plugin install
• ComfyUI Desktop or a portable build becomes unstable after adding custom nodes

Typical log lines look like this:

IMPORT FAILED: ComfyUI-ExampleNode
ModuleNotFoundError: No module named 'somepackage'

or:

IMPORT FAILED: ComfyUI-AnotherNode
ImportError: DLL load failed while importing cv2

or:

IMPORT FAILED: ComfyUI-SomePlugin
AttributeError: module 'torch' has no attribute '...'

Step 1: Confirm which plugin actually failed

Do not start by installing random packages.

First, answer two questions:

1. Which plugin failed to import?
2. What exact error type stopped the import?

Check these places in order:

Startup log

Look for IMPORT FAILED and copy the full traceback. That tells you which plugin broke and what Python object could not be imported.

ComfyUI import failure API

If ComfyUI still launches far enough to open the UI, visit:

http://127.0.0.1:8188/v2/customnode/import_fail_info_bulk

This returns a JSON list of custom node import failures.

Workflow symptoms

If the workflow shows red nodes, note whether:

• only one plugin family is missing
• multiple unrelated plugins are missing
• the same plugin worked before another install changed the environment

That pattern helps you decide whether the problem is isolated or global.

Step 2: Classify the failure before you fix it

Most plugin import failures fall into one of these buckets.

Missing package

Example:

ModuleNotFoundError: No module named 'onnxruntime'

This is usually the simplest case. The plugin expects a package that is not installed.

Broken binary package

Example:

ImportError: DLL load failed while importing cv2

The package may be installed, but the binary wheel does not match your Python, CUDA, or Windows runtime.

Core dependency drift

Example:

AttributeError: module 'torch' has no attribute '...'

This often means another plugin install changed a core package like torch, numpy, pillow, or opencv-python.

Repo or source problem

Example:

• The repo moved
• The plugin was only partially cloned
• Git LFS left placeholder files behind
• The plugin is too old for your current ComfyUI version

Step 3: Repair the environment manually

Once you know the failure class, use the narrowest fix possible.

If a package is missing

Install the missing package, but do it carefully:

pip install somepackage

Then restart ComfyUI and test again.

If the package is known to be sensitive or heavy, prefer a pinned install or a wheel recommended by the plugin author.

If you suspect dependency drift

Run:

pip check

Look for conflicts around:

• torch
• torchvision
• torchaudio
• numpy
• pillow
• opencv-python
• transformers
• diffusers

If one plugin install changed one of those, repair the core package first, then test the failing plugin again.

For a broader walkthrough, see ComfyUI Dependency Conflicts.

If the plugin itself is outdated or partially broken

Go to the plugin folder under custom_nodes/ and verify:

• the repo is the right one
• the source files are real code, not Git LFS pointers
• the plugin still supports your ComfyUI version
• the plugin README does not require extra manual packages

If necessary:

cd custom_nodes/<plugin-name>
git pull

Then reinstall only that plugin's requirements:

pip install -r requirements.txt

Before doing that on a fragile environment, read the requirements file and watch for core package downgrades.

Step 4: Know when the command-line loop is making things worse

Manual repair is valid, but it gets expensive once you are doing this repeatedly:

• install one package
• restart ComfyUI
• hit a new import failure
• install another package
• break Torch
• repair Torch
• lose track of what changed

That is exactly where many users end up reinstalling even when the environment was technically recoverable.

A safer recovery path with Wonderful Launcher

Wonderful Launcher is built for the stage where ComfyUI is already messy.

Instead of treating the problem as "install a plugin and hope", the launcher is meant to help you:

• take over an existing Desktop or portable install
• recover from plugin import failures with less command-line hopping
• reduce reinstall risk when workflows, models, and custom nodes already exist
• move from self-recovery to expert help without changing tools

When to escalate to expert help

Use expert help if:

• multiple unrelated plugins fail at once
• startup failures appear after plugin repair attempts
• Torch, CUDA, or OpenCV were already changed by previous fixes
• you cannot tell whether the problem is the repo, the environment, or the workflow

That is where live recovery is usually faster than another evening of trial and error.

Related guides

• ComfyUI Startup Failed? How to Diagnose and Recover Faster
• ComfyUI Dependency Conflicts
• ComfyUI Plugin Management
• Troubleshooting Decision Tree

Need the fastest path?

If your environment is already unstable, download Wonderful Launcher first. If the environment is still blocked after that, use expert help as the second step.