LoRA Save 🍌

This is a terminal node - it doesn't pass anything downstream, it just writes a LoRA to disk. It's the last step in cgem156's lora_merger folder, which handles combining multiple LoRAs into one file entirely inside ComfyUI rather than requiring a separate CLI tool or script.

Where this fits

The pack's README lists two features under this umbrella: lora_merger for combining LoRAs and multiple_lora_loader for applying several at once at inference time. This node belongs to the merger side - after whatever merge or SVD-rank operation you've run elsewhere in the pack (LoRA SVD Rank being one example in this same folder), LoRA Save is what actually persists the result as a file you can reuse later, share, or load normally in any other workflow without re-running the merge every time.

Inputs and outputs

Required:

• lora - a LoRA value, typically the output of a merge step upstream in this pack's lora_merger nodes.
• file_name - what to name the saved file, default merged.
• extension - a combo choosing the file format to save in (the standard choices in this space are .safetensors and .pt-style formats).

is_output_node is true and there are no outputs - this node's entire job is the side effect of writing a file.

Installing it

ComfyUI Manager, search "cgem156-ComfyUI". Manual clone:

cd ComfyUI/custom_nodes
git clone https://github.com/laksjdjf/cgem156-ComfyUI

Restart ComfyUI. No model download tied to this node specifically - it's a file writer operating on a LoRA you've already produced elsewhere in the graph.

Common issues

The most common surprise with save nodes generally is not knowing where the file landed - check your ComfyUI output directory conventions (or whatever path this pack's other nodes default to) if a merge seems to succeed but you can't find the result afterward. Beyond that, since this is purely a file-writing step with no transformation of its own, most real problems trace back to whatever fed it: if the saved LoRA doesn't behave as expected when reloaded, the bug is almost certainly upstream in the merge or SVD-rank step, not in this node. Worth double-checking extension matches what you actually intend to load it with later - a mismatch between the format you saved and the format your loader expects is an easy, avoidable mistake.

If you're building a merge pipeline for the first time in this pack, it's worth reading the pieces in order: merge your source LoRAs with whatever node handles that step in lora_merger, optionally run LoRA SVD Rank if you want to compress the result down to a smaller rank first, and only then hit this node to write the final file. Trying to plug a raw, unmerged LoRA straight into LoRA Save skips the actual point of the pipeline - this node saves whatever you hand it, it doesn't merge anything itself.