ComfyUI GGUF: Loader, Model Folder, and No Module Named 'gguf'

If you searched for comfyui gguf not showing, unet loader gguf comfyui, where to put gguf in comfyui, or comfyui-gguf modulenotfounderror: no module named 'gguf', the practical answer is:

install or update ComfyUI-GGUF, put each .gguf file in the model folder its loader scans, and use a GGUF loader instead of the normal checkpoint or diffusion loader.

GGUF in ComfyUI is not just a file extension swap. The official ComfyUI issue tracker has repeated reports where users placed GGUF files in a model folder and expected the stock loader to see them. The fix is usually not moving the file again; it is using the custom loader that understands GGUF.

Quick Answer

Do not start with Load Checkpoint for a GGUF diffusion model. Do not assume Load Diffusion Model will load .gguf just because the file is under models/diffusion_models.

What the Official Issue Trail Shows

The current evidence points to one important distinction: ComfyUI can scan model folders, but core ComfyUI loaders are not the same as ComfyUI-GGUF loaders.

• In ComfyUI issue #6352, a user placed GGUF models in diffusion_models and expected Load Diffusion Model to list them. The answer was to use the loaders from city96/ComfyUI-GGUF.
• In ComfyUI issue #6098, a user expected core DualCLIPLoader to show a T5 GGUF file. A ComfyUI maintainer stated that core nodes did not support .gguf; the user needed ComfyUI-GGUF.
• The ComfyUI feature request #4598 for built-in GGUF support is still a useful signal: GGUF demand is real, but working local setups still commonly depend on the custom node.
• In ComfyUI-GGUF issue #348 and #206, the visible symptom was ModuleNotFoundError: No module named 'gguf'. In both patterns, the repair is installing requirements into the environment that actually runs ComfyUI, not the system Python you happen to have in PATH.

That is why this guide treats GitHub Issues as evidence of real failure modes, but uses the ComfyUI source and ComfyUI-GGUF README for the actual folder and loader behavior.

When Should You Use GGUF?

Use GGUF when the workflow or model page explicitly provides a .gguf model and a matching GGUF loader. It is most common for large transformer or DiT-style models where quantization can reduce VRAM pressure, such as Flux-style workflows, Wan video workflows, SD3.5 variants, and newer community quantized model releases.

GGUF is a good choice when:

• your GPU is short on VRAM
• the workflow was built around a GGUF diffusion model or GGUF text encoder
• the model author provides tested GGUF files and loader instructions
• you are willing to trade some setup simplicity for lower memory use

GGUF is not the right first fix when:

• you are running a normal .safetensors checkpoint workflow
• the workflow says Load Checkpoint and does not mention GGUF
• the GGUF custom node is not importing cleanly
• the model architecture is not supported by your GGUF loader yet

Do not rename .safetensors to .gguf, and do not put a random GGUF file into checkpoints hoping a normal checkpoint workflow will load it.

Fix No Module Named 'gguf'

That error means ComfyUI-GGUF tried to import its Python dependency and failed. The usual cause is that gguf was installed into the wrong Python environment.

For the official Windows portable package, the portable root is the folder that contains run_nvidia_gpu.bat, python_embeded, and the ComfyUI folder.

After installing requirements, restart ComfyUI and confirm the startup log no longer shows:

IMPORT FAILED
ModuleNotFoundError: No module named 'gguf'
Cannot import ... ComfyUI-GGUF

Install or Update ComfyUI-GGUF

For a manual install, open a terminal in your ComfyUI folder and clone the custom node:

cd ComfyUI/custom_nodes
git clone https://github.com/city96/ComfyUI-GGUF

Then install its requirements in the same Python environment used by ComfyUI:

cd ComfyUI-GGUF
python -m pip install -r requirements.txt

For a Windows portable package, run from the portable root:

git clone https://github.com/city96/ComfyUI-GGUF ComfyUI\custom_nodes\ComfyUI-GGUF
.\python_embeded\python.exe -s -m pip install -r .\ComfyUI\custom_nodes\ComfyUI-GGUF\requirements.txt

If the custom node already exists, update it from the custom node manager or run git pull inside ComfyUI/custom_nodes/ComfyUI-GGUF, then reinstall requirements if the README or logs suggest it.

Where to Put GGUF Files

Current ComfyUI source maps the newer folder names and legacy folder names together:

• diffusion_models scans both ComfyUI/models/diffusion_models and ComfyUI/models/unet
• text_encoders scans both ComfyUI/models/text_encoders and ComfyUI/models/clip

ComfyUI-GGUF also adds GGUF-specific file lists on top of those model categories. That is why older guides may say models/unet, while newer ComfyUI docs and templates usually point to models/diffusion_models.

Do not put a GGUF diffusion model here:

ComfyUI/models/checkpoints

That folder is for checkpoint loader workflows. A GGUF diffusion model is a split diffusion model file, not an all-in-one Stable Diffusion checkpoint.

For a broader folder map, see Where to Put Safetensors in ComfyUI.

Use the Right GGUF Loader

After installing ComfyUI-GGUF, search for nodes such as:

Unet Loader (GGUF)
CLIPLoader (GGUF)
DualCLIPLoader (GGUF)
TripleCLIPLoader (GGUF)

Use Unet Loader (GGUF) in place of Load Diffusion Model when the workflow expects a GGUF diffusion model. Use a GGUF CLIP loader when the workflow expects a GGUF T5 or text encoder.

If you have several similar loader nodes from different extensions, follow the workflow author's exact node names. If a workflow uses a loader from a different GGUF-related extension, installing only city96/ComfyUI-GGUF may not satisfy that workflow.

Do not spend twenty minutes moving model files around until you confirm the loader node itself exists. If the node is absent, the real problem is usually the custom node install or import.

Fast Diagnosis

Template and Download Popup Caveat

ComfyUI workflow templates can embed model names, URLs, and target directories. Official docs note that current missing-file detection checks the corresponding top-level directory, and that some model formats such as .gguf can be treated as unsafe for embedded template downloads.

Practical result: a GGUF workflow may still require manual model download even if a template or workflow knows which file it wants. Download from the model author's page, place the file in the matching top-level folder, restart or refresh ComfyUI, then choose it in the GGUF loader.

Common Problems

GGUF vs Checkpoint vs Safetensors

The file extension alone is not enough. The workflow and loader node decide where the file belongs.

How Wonderful Launcher Helps

GGUF setup can touch custom nodes, model folders, Python packages, and workflow loaders at the same time. Wonderful Launcher helps when you want to:

• keep a stable non-GGUF environment separate from a GGUF experiment
• preserve working workflows before adding custom nodes
• collect startup logs when loader nodes fail
• recover models and workflows before rebuilding ComfyUI

If GGUF install breaks other nodes, start with ComfyUI Dependency Conflicts.

Related Guides

• Where to Put Safetensors in ComfyUI
• How to Save a Workflow in ComfyUI
• Download Models
• Install ComfyUI Custom Nodes
• ComfyUI Plugin Import Failed
• ComfyUI No Module Named 'llama_cpp'
• ComfyUI Dependency Conflicts
• GPU Compatibility

Source References

• ComfyUI-GGUF README
• ComfyUI-GGUF node source
• ComfyUI folder path source
• Official ComfyUI model concepts
• Official ComfyUI CLIPLoader docs
• Official ComfyUI workflow template docs
• ComfyUI issue #4598: Support GGUF
• ComfyUI issue #6352: Load Diffusion Model does not list GGUF
• ComfyUI issue #6098: DualCLIPLoader does not see GGUF CLIP models
• ComfyUI-GGUF issue #348: No module named gguf import failure
• ComfyUI-GGUF issue #206: gguf installed in the wrong environment