No module named 'nunchaku' in ComfyUI: Nunchaku Fix

If you searched for NunchakuFluxLoraLoader, NunchakuFluxLoraStack, NunchakuFluxDiTLoader, or No module named 'nunchaku', the short answer is this:

do not install the unrelated PyPI package named nunchaku. First confirm whether the ComfyUI-nunchaku plugin imported, then install the official Nunchaku wheel that matches your Python, PyTorch, CUDA, and GPU.

Error message

When ComfyUI starts or runs a workflow, the terminal log shows:

ModuleNotFoundError: No module named 'nunchaku'

or:

ImportError: cannot import name 'SVDQW4A4Linear'

Nunchaku FLUX workflows may also show red or missing nodes such as:

Node import failed: NunchakuFluxLoraLoader
Node import failed: NunchakuFluxLoraStack
Node import failed: NunchakuFluxDiTLoader

Those node names are the visible workflow symptom. The root problem is usually that the ComfyUI-nunchaku plugin could not load its Nunchaku backend.

What Nunchaku is

Nunchaku is a 4-bit quantized inference engine developed by the Nunchaku team around the SVDQuant method. In ComfyUI, it is usually used through the ComfyUI-nunchaku plugin to run quantized FLUX, Qwen-Image, SANA, PixArt, and related model workflows on NVIDIA GPUs with lower VRAM usage.

What happens if you skip the install

ComfyUI itself can run fine without Nunchaku. Nunchaku is only needed when:

• your workflow uses nodes from the ComfyUI-nunchaku plugin
• you load SVDQuant/Nunchaku quantized model files
• the workflow explicitly asks for Nunchaku loader or sampler nodes

If your workflow does not involve those nodes or quantized models, you can ignore this error and avoid adding a heavy native dependency.

Which Nunchaku error do you have?

If the workflow only shows red node names, do not search for a package named after the node class. Fix the plugin import and backend wheel first.

First: avoid the PyPI name trap

The missing Python module is named nunchaku, but the public PyPI package named nunchaku is an unrelated scientific package for piece-wise linear segmentation. Installing it will not fix ComfyUI Nunchaku workflows.

Nunchaku's official docs recommend installing a prebuilt Nunchaku wheel from their release sources, or using the ComfyUI-nunchaku install workflow when available.

Hardware and software requirements

Nunchaku has strict compatibility constraints. Verify the following before installing:

GPU requirements

Nunchaku is primarily for NVIDIA GPUs. Official docs list support by GPU architecture, CUDA, PyTorch, and wheel availability. If you are on AMD, Intel, Apple Silicon, or CPU-only hardware, use a different workflow path such as GGUF quantization or a smaller model.

CUDA and PyTorch requirements

Nunchaku wheels are tied to your environment. Check:

python --version
python -c "import torch; print(torch.__version__, torch.version.cuda)"

On the official Windows portable package, run from the portable package root:

.\python_embeded\python.exe -s --version
.\python_embeded\python.exe -s -c "import torch; print(torch.__version__, torch.version.cuda)"

Match the wheel to:

• Python tag, such as cp311 or cp312
• PyTorch version
• CUDA version
• Windows architecture
• GPU generation when the wheel or docs mention it

Installation methods

Method 1: Use the ComfyUI-nunchaku installer workflow

If you already have a recent ComfyUI-nunchaku plugin installed, use its install_wheel.json workflow:

1. Load the install_wheel.json workflow in ComfyUI.
2. Run the installer node in update node mode to fetch available versions.
3. Select the wheel that matches your Python, PyTorch, CUDA, and GPU.
4. Run the node in install mode.
5. Fully restart ComfyUI.

This method is safest because the plugin can guide wheel selection for the active environment.

After the install, restart ComfyUI completely. A browser refresh is not enough because custom nodes are registered during backend startup.

Method 2: Manually install a matching wheel

Download a wheel from Nunchaku's official release sources. The official docs currently link these sources, and GitHub may redirect between project organization names:

• GitHub Releases: https://github.com/nunchaku-tech/nunchaku/releases
• Hugging Face organization: https://huggingface.co/nunchaku-tech
• ModelScope organization: https://modelscope.cn/organization/nunchaku-tech

Then install it with the Python that starts ComfyUI:

python -m pip install <path-or-url-to-matching-nunchaku-wheel.whl>

For the official Windows portable package:

.\python_embeded\python.exe -s -m pip install <path-or-url-to-matching-nunchaku-wheel.whl>

Do not copy an example wheel URL blindly. A wheel for Python 3.11 and PyTorch 2.7 will not repair a Python 3.12 / different PyTorch environment.

Method 3: Build from source only if you know you need it

Building Nunchaku on Windows requires a compatible CUDA toolkit, Visual Studio/MSVC, and the correct development environment. For most ComfyUI users, a prebuilt wheel or the installer workflow is the practical path.

Common installation failures

"No matching distribution found"

Your Python/PyTorch/CUDA combination does not have a matching wheel available from the source you used. Check:

python --version
python -c "import torch; print(torch.__version__, torch.version.cuda)"

Then pick a wheel built for that exact combination, or adjust the base environment before installing Nunchaku.

Wrong package installed

If you already installed the unrelated PyPI package named nunchaku, it will not repair ComfyUI Nunchaku. Uninstall it before installing the correct wheel:

python -m pip uninstall nunchaku -y

Then use the official Nunchaku wheel/install workflow.

Local directory conflict

If there is a folder named nunchaku in the current working directory, Python may load that folder instead of the installed package. Make sure there is no conflicting folder under the ComfyUI root or the plugin folder.

Node still red after the wheel imports

If import nunchaku works but NunchakuFluxLoraLoader, NunchakuFluxLoraStack, or NunchakuFluxDiTLoader is still missing, the backend may be installed while the custom node plugin is stale or failed during startup.

Check:

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

Then look for ComfyUI-nunchaku in the response or startup log. If another dependency conflict stopped the plugin before it registered nodes, repair that earliest import failure first. The broader dependency conflicts guide explains how to avoid breaking Torch, NumPy, OpenCV, or Transformers while fixing one plugin.

RTX 20 series special configuration

Older Turing GPUs have different datatype/attention constraints than newer Ada or Blackwell GPUs. Follow the Nunchaku docs for the exact node settings for your GPU generation instead of copying settings from a 40-series or 50-series workflow.

Verify the installation

Run:

python -c "import nunchaku; print(getattr(nunchaku, '__version__', 'installed'))"

Then fully restart ComfyUI and check whether the ComfyUI-nunchaku nodes register.

Alternatives

If your hardware does not meet Nunchaku's requirements, consider:

• Full-precision models: if you have enough VRAM
• GGUF quantization: often easier to run across more setups
• Smaller models: SDXL or SD 1.5 workflows can run on less VRAM
• Cloud GPU: when local hardware is not suitable

Still not resolved?

If you still have issues, collect:

• full terminal error log
• output of python --version
• output of python -c "import torch; print(torch.__version__, torch.version.cuda)"
• your GPU model
• the exact wheel filename or installer workflow output

Then use expert help if you need the fastest recovery path.

Source References

• ComfyUI-nunchaku GitHub repository
• Nunchaku wheel repository on Hugging Face
• Nunchaku GitHub releases
• ComfyUI custom node troubleshooting guide