Nunchaku Model Merger
Reassemble a split Nunchaku model into one file
- status
This one isn't a model-blending node - don't let "merger" fool you. NunchakuModelMerger takes a Nunchaku model that's been distributed as a folder of shards and reassembles it into a single file. It's a housekeeping utility, not a creative one, and it exists because some Nunchaku models ship split across multiple files and a single .safetensors is simpler to manage and load.
If you downloaded a Nunchaku model and ended up with a directory full of pieces instead of one neat file, this is the node that fixes that.
How it works
Large model weights are often published as multiple shards - partly for hosting limits, partly for parallel downloads. This node points at the folder holding those shards, stitches them back into one consolidated file, and writes it out under the name you give. It's the load-from-folder-vs-load-from-single-file distinction the pack's loaders mention: some workflows want the merged single file, and this produces it. Purely a file operation; it doesn't change the weights or the quality.
Because it writes a file rather than passing tensors downstream, it's an operation you run once after downloading, not something that sits in your generation graph every run.
The inputs and outputs that matter
Two inputs, both simple:
model_folder- the directory containing the split model shards. Pick the folder you downloaded.save_name(STRING) - the filename for the merged output. Name it something you'll recognize in your model dropdowns later.
Output is a status string - the result of the merge (success, path written, or an error message). You read it to confirm it worked; nothing downstream consumes it.
How to install it
Comes with the ComfyUI-nunchaku pack. ComfyUI Manager → search "ComfyUI-nunchaku" → install, or
cd ComfyUI/custom_nodes
git clone https://github.com/mit-han-lab/ComfyUI-nunchaku
then restart. Like the rest of the pack it expects the backend wheel installed (via NunchakuWheelInstaller), and it operates on Nunchaku model files you've already downloaded into your models folder.
Common issues & troubleshooting
Empty model_folder dropdown. No sharded model folder where it's looking. Download a Nunchaku model that ships as shards, or point it at the correct directory.
Merge fails or the file's unusable. Usually an incomplete download - one shard missing or truncated. Re-download the full set and try again; a partial folder can't be reassembled.
"Do I even need this?" Only if your model came as multiple files and your loader wants a single one. Plenty of Nunchaku models are already single files - in that case, skip this node entirely.
Disk space. Merging writes a fresh full-size copy, so make sure you've got room for the combined file before running it. Once it's written and loading fine, you can delete the shard folder.
Inputs (2)
| Name | Type | Default | Description |
|---|---|---|---|
| model_folder | COMBO | Nunchaku FLUX.1 model folder. | |
| save_name | STRING | Filename to save the merged model as. |
Outputs (1)
| Name | Type | Description |
|---|---|---|
| status | STRING | — |