Fix Custom Node requirements.txt Not Installed in ComfyUI

Fix ComfyUI custom nodes that fail to load because their requirements.txt dependencies were never installed in the correct Python environment.

Community Knowledge

This page is based on common ComfyUI troubleshooting patterns and has not been fully tested across all environments. Back up your environment before changing packages.

A ComfyUI custom node shows IMPORT FAILED in the startup log with a ModuleNotFoundError most often because the node's requirements.txt was never installed. Cloning or downloading a node into custom_nodes/ only copies the code; its Python packages must be installed separately, into the Python that starts ComfyUI.

IMPORT FAILED: ComfyUI-ExampleNode
Traceback (most recent call last):
  File "...\custom_nodes\ComfyUI-ExampleNode\__init__.py", line 4, in <module>
    import somepackage
ModuleNotFoundError: No module named 'somepackage'

Triage

Missing module	Common workflow area	First thing to verify
`triton`	Acceleration, SageAttention, video, large-model workflows	Decide whether it is optional or fatal before installing
`sageattention`	Wan/video acceleration and forced attention flags	Remove forced flags unless the workflow truly requires it
`onnx` / `onnxruntime`	Face, pose, DWPose, ReActor, IPAdapter, video helpers	Prefer CPU `onnxruntime` first unless GPU ONNX is required
`llama_cpp`	GGUF prompt helpers, local LLM/VLM nodes	Package is usually `llama-cpp-python`, not `llama_cpp`
`nunchaku`	Nunchaku-backed FLUX/Qwen/SANA workflows	Do not install the unrelated PyPI `nunchaku` package
`sox` / `decord`	Audio and video helper nodes	Identify the node pack first; do not batch-install every video package
`groundingdino` / `segment_anything`	Detection, grounding, segmentation, masking	Use the SAM/GroundingDINO import guide, then confirm model assets
`dlib` / `nvvfx`	Face/video effect workflows	Native wheels and platform support matter; read the plugin README first

If a module is not in the table, the same rule applies: find the failing plugin first, then install the smallest needed dependency into the Python that starts ComfyUI.

Why it happens

Manual git clone without pip install — the repo was cloned but requirements were never installed
ComfyUI Manager partial install — the repo cloned but the requirements step failed silently
Environment was rebuilt — reinstalling or updating the portable package wiped previously installed packages
Wrong Python used — requirements went into system Python, not ComfyUI's embedded Python
No requirements.txt — some nodes list dependencies only in their README

Step 1: Find the failing node

Check the startup log for the IMPORT FAILED line. It names the custom node folder and the missing module.

Step 2: Check whether requirements.txt exists

dir ComfyUI\custom_nodes\<node-name>\requirements.txt

If it exists, continue. If not, check the node's README on GitHub for manual dependency instructions.

Step 3: Read the requirements file before installing

type ComfyUI\custom_nodes\<node-name>\requirements.txt

Watch for red flags before running it:

What to watch for	Risk
`torch`, `torchvision`, `torchaudio`	May replace your CUDA-enabled PyTorch with a CPU version
Unpinned versions (no `==` or `>=`)	May install incompatible versions
`opencv-python` when you have `opencv-python-headless`	Can cause DLL conflicts
Very old or very new version pins	May conflict with other nodes

Step 4: Install the requirements

For Windows portable:

.\python_embeded\python.exe -s -m pip install -r ComfyUI\custom_nodes\<node-name>\requirements.txt

For a manual venv:

python -m pip install -r custom_nodes/<node-name>/requirements.txt

If the file would replace core packages like torch, install only the specific missing package instead:

.\python_embeded\python.exe -s -m pip install somepackage

Step 5: Verify

Restart ComfyUI and confirm the IMPORT FAILED line is gone:

.\python_embeded\python.exe -s ComfyUI\main.py --windows-standalone-build

When not to run pip install blindly

Always read requirements.txt first — some authors pin aggressive torch versions.
If pip check shows conflicts after installing, fix the conflict before installing more.
Very old pins may not work with your current Python or PyTorch.
Install one node at a time, checking for errors between installs, rather than batch-installing every node's requirements.

Source References

Start free with Wonderful Launcher if this affects your real ComfyUI environment. It keeps launcher-native repair, task logs, and runtime checks in one place; credits are only for image generation and metered tools.

Download Wonderful Launcher See credit packages

Did this fix your issue?

Your answer helps prioritize verified ComfyUI repairs.

Community Knowledge

This page is based on common ComfyUI troubleshooting patterns and has not been fully tested across all environments. Back up your environment before changing packages.

IMPORT FAILED: ComfyUI-ExampleNode
Traceback (most recent call last):
  File "...\custom_nodes\ComfyUI-ExampleNode\__init__.py", line 4, in <module>
    import somepackage
ModuleNotFoundError: No module named 'somepackage'

Triage

Missing module	Common workflow area	First thing to verify
`triton`	Acceleration, SageAttention, video, large-model workflows	Decide whether it is optional or fatal before installing
`sageattention`	Wan/video acceleration and forced attention flags	Remove forced flags unless the workflow truly requires it
`onnx` / `onnxruntime`	Face, pose, DWPose, ReActor, IPAdapter, video helpers	Prefer CPU `onnxruntime` first unless GPU ONNX is required
`llama_cpp`	GGUF prompt helpers, local LLM/VLM nodes	Package is usually `llama-cpp-python`, not `llama_cpp`
`nunchaku`	Nunchaku-backed FLUX/Qwen/SANA workflows	Do not install the unrelated PyPI `nunchaku` package
`sox` / `decord`	Audio and video helper nodes	Identify the node pack first; do not batch-install every video package
`groundingdino` / `segment_anything`	Detection, grounding, segmentation, masking	Use the SAM/GroundingDINO import guide, then confirm model assets
`dlib` / `nvvfx`	Face/video effect workflows	Native wheels and platform support matter; read the plugin README first

If a module is not in the table, the same rule applies: find the failing plugin first, then install the smallest needed dependency into the Python that starts ComfyUI.

Why it happens

Manual git clone without pip install — the repo was cloned but requirements were never installed
ComfyUI Manager partial install — the repo cloned but the requirements step failed silently
Environment was rebuilt — reinstalling or updating the portable package wiped previously installed packages
Wrong Python used — requirements went into system Python, not ComfyUI's embedded Python
No requirements.txt — some nodes list dependencies only in their README

Step 1: Find the failing node

Check the startup log for the IMPORT FAILED line. It names the custom node folder and the missing module.

Step 2: Check whether requirements.txt exists

dir ComfyUI\custom_nodes\<node-name>\requirements.txt

If it exists, continue. If not, check the node's README on GitHub for manual dependency instructions.

Step 3: Read the requirements file before installing

type ComfyUI\custom_nodes\<node-name>\requirements.txt

Watch for red flags before running it:

What to watch for	Risk
`torch`, `torchvision`, `torchaudio`	May replace your CUDA-enabled PyTorch with a CPU version
Unpinned versions (no `==` or `>=`)	May install incompatible versions
`opencv-python` when you have `opencv-python-headless`	Can cause DLL conflicts
Very old or very new version pins	May conflict with other nodes

Step 4: Install the requirements

For Windows portable:

.\python_embeded\python.exe -s -m pip install -r ComfyUI\custom_nodes\<node-name>\requirements.txt

For a manual venv:

python -m pip install -r custom_nodes/<node-name>/requirements.txt

If the file would replace core packages like torch, install only the specific missing package instead:

.\python_embeded\python.exe -s -m pip install somepackage

Step 5: Verify

Restart ComfyUI and confirm the IMPORT FAILED line is gone:

.\python_embeded\python.exe -s ComfyUI\main.py --windows-standalone-build

When not to run pip install blindly

Always read requirements.txt first — some authors pin aggressive torch versions.
If pip check shows conflicts after installing, fix the conflict before installing more.
Very old pins may not work with your current Python or PyTorch.
Install one node at a time, checking for errors between installs, rather than batch-installing every node's requirements.

Source References

Download Wonderful Launcher See credit packages

Did this fix your issue?

Your answer helps prioritize verified ComfyUI repairs.

Fix Custom Node requirements.txt Not Installed in ComfyUI

Triage

Why it happens

Step 1: Find the failing node

Step 2: Check whether requirements.txt exists

Step 3: Read the requirements file before installing

Step 4: Install the requirements

Step 5: Verify

When not to run pip install blindly

Source References

Table of Contents

Fix Custom Node requirements.txt Not Installed in ComfyUI

Triage

Why it happens

Step 1: Find the failing node

Step 2: Check whether requirements.txt exists

Step 3: Read the requirements file before installing

Step 4: Install the requirements

Step 5: Verify

When not to run pip install blindly

Source References

Table of Contents