🐳LoRA Loader

Read the name, reach for it expecting the built-in LoraLoader, and you will be confused. This is not that node. 🐳LoRA Loader (LoRALoaderSimple) from Pond Nodes loads a LoRA and multiplies every weight by a strength value, then hands you the scaled raw weights - it does not patch a checkpoint and it won't plug into a KSampler path. Its real job is the first half of a two-node flow: load a LoRA, bake a strength into it, and save a permanent copy with the pack's LoRA Saver.

That's genuinely useful, and it solves a specific pain: some LoRAs are only ever good at a strength you don't want to type every time. Say a style LoRA is always too strong at 1.0 and perfect at 0.6. Rather than remembering "0.6" forever, you load it here, set strength to 0.6, and save a pre-scaled .safetensors that you then load with the normal LoraLoader at 1.0. One-time tuning, permanent copy.

How it works

Under the hood it's honest and simple: it reads the LoRA from ComfyUI's models/loras folder (handles .safetensors via safetensors and .pt via torch.load), then loops over every tensor in the file and multiplies it by strength. That's the whole mechanism. The output is a dict of scaled tensors tagged with the custom LORA type.

Inputs and output

• lora_name - dropdown of the LoRA files in your models/loras folder.
• strength - 0.0 to 10.0, default 1.0. This is the multiplier baked into the weights.

Output is a single lora (LORA). Here's the part to internalize: that output is only consumable by nodes that understand this pack's LORA type - in practice, LoRASaver. The built-in ComfyUI LoraLoader won't accept it, and neither will a sampler. If you wire this straight into a normal LoRA position in a workflow, you'll get a type error, not an image. If you just want to apply a LoRA for generation, use the built-in loader; this node is for producing adjusted LoRA files.

A note on the math

Scaling a LoRA's weights by a constant is a shortcut that works reasonably for most SD/SDXL LoRAs (a linear "strength" has a predictable-ish effect), but it's not identical to how the model applies a LoRA weight at inference, and it can overshoot on weights with very different scales. For most LoRAs 0.5–0.8 in, 1.0 out will land close to what you expect. If you're baking in an aggressive scale, eyeball the result before trusting the file.

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. It needs torch (which you already have) and safetensors (bundled with ComfyUI). No model downloads. The pack's usual footnote: the README warns of a console-spam conflict with comfyui_HiDream-Sampler.