Krea2 Merge • Save LoRA
Save your merged Krea 2 LoRA without nuking the file you already had
- merged_model
- saved_path
Three inputs, one job: write the merged state dict from Merge LoRAs to disk, then tell you where it went. Krea2 Merge • Save LoRA is the last node in the pack's pipeline, and it's the one with the friendliest guardrails - the author clearly got burned by overwriting files at some point and decided nobody else should be.
How it works
By default it writes to its own folder, ComfyUI/models/loras/krea2-merged-loras, which it creates if needed - a deliberate choice so it never collides with the output folder of the original comfyui-merge extension. If you give it an absolute path instead of a filename, it writes exactly there. Files ending in .safetensors are saved with safetensors; anything else falls through to torch.save.
The overwrite protection is the star. With allow_overwrite set to no (the default), an existing file with the same name raises a clear error - your previous merge survives. Set it to yes and the node first moves the old file to a timestamped backup, then saves, then deletes the backup on success. If the save fails, it restores the original automatically. On Windows there's one extra failure mode worth knowing: if the destination is open or memory-mapped - say, because something in your workflow has that LoRA loaded right now - you'll get a permission error. Use a different filename or restart ComfyUI rather than fighting it.
The inputs that matter
- merged_model - the state dict from Krea2 Merge • Merge LoRAs. Nothing else goes here.
- modeloutput - a filename, defaulting to
krea2_merged_lora.safetensors. Keep the.safetensorsextension. - allow_overwrite - no by default; flip to yes for repeatable runs that replace the same output.
The node is also registered as an output node, which means it runs even with nothing attached downstream - you don't need a display node to force the save. It still returns one saved_path string so you can hook it to a Show Text node (the bundled example workflow uses pythongosssss's ComfyUI-Custom-Scripts) to see exactly where the file landed.
Install
Same story as the rest of the pack - install once:
cd ComfyUI/custom_nodes
git clone https://github.com/cyberdeliaAI/comfyui-krea2-merge
or search Krea2 Merge in ComfyUI Manager, then restart. safetensors (already bundled) is the only dependency; no base model is ever downloaded or modified - this node only ever writes LoRA files.
After the save
That merged .safetensors is now a normal LoRA sitting in your lora folder. Load it with a stock Load LoRA node, or with this pack's Apply LoRA node, at whatever strength you like - and there's the upside of the whole pipeline: the merge is baked, but you still get to choose how hard to lean on the result at runtime. One note if you're coming from the older comfyui-merge: old workflows that saved to the original folder won't point here, and vice versa, so check the saved path the first time rather than assuming.
Inputs (3)
| Name | Type | Default | Description |
|---|---|---|---|
| merged_model | MODEL | — | |
| modeloutput | STRING | krea2_merged_lora.safetensors | — |
| allow_overwrite | COMBO | no | 2 options: no, yes |
Outputs (1)
| Name | Type | Description |
|---|---|---|
| saved_path | STRING | — |