🐳LoRA Saver

LoRA Saver is the tail end of a two-node trick that the Pond Nodes pack plays: its sibling LoRA Loader loads a LoRA, scales every weight by a strength, and this node writes the result back out as a fresh .safetensors file. Load a LoRA that's always too strong, bake in a strength you actually like, and walk away with a permanent pre-scaled copy you can load anywhere like a normal LoRA.

It's an output node, so it doesn't hand anything forward - its job is finished the moment the file is on disk. If you've ever wished a particular LoRA just came at the right strength, this is the workflow that gets you there.

How it works

The mechanism is a thin wrapper around safetensors' save_file. It takes the LORA dict it's given, ensures every tensor is contiguous and savable, appends .safetensors if you forgot the extension, and writes to ComfyUI's output directory. It even prints the saved path to the console, which is more feedback than most ComfyUI output nodes bother with.

Inputs

• lora - the LORA input. In practice this comes from the pack's LoRALoaderSimple, which is the only common source of this custom LORA type.
• filename - the output name, default adjusted_lora. It lands in ComfyUI/output/adjusted_lora.safetensors unless you change it.

That's it. No outputs beyond the file.

The important caveat

The lora input is the pack's custom LORA type, and that limits what you can feed it. You can't, for instance, grab the patched-LoRA reference from ComfyUI's built-in LoraLoader and dump it here - the built-in loader doesn't output this type. This node is designed to sit directly after LoRALoaderSimple, forming a load → scale → save loop. Wire it any other way and you'll get a type mismatch.

One more honest note: baking a strength in by multiplying raw weights is a shortcut, not the same thing as the sampler's LoRA application math. It's usually close for SD/SDXL LoRAs at moderate strengths, but if you're scaling aggressively, verify the output actually looks like you expect before you delete the original file. Keep the source LoRA around - the whole point is a convenience copy, not a replacement.

Install

Standard Pond Nodes install:

ComfyUI Manager → search "Pond Nodes" (comfy_Pond_Nodes)

or:

cd ComfyUI/custom_nodes
git clone https://github.com/Pondowner857/comfy_Pond_Nodes
cd comfy_Pond_Nodes
pip install -r requirements.txt

Restart after. Torch and safetensors cover it - both already in your ComfyUI environment. The pack's usual asterisk: the README flags a console-spam conflict with comfyui_HiDream-Sampler if you have both.