LoRA Loader Weight Only 🍌

This node is easy to misread at first: it doesn't take a MODEL input and it doesn't output one either. That's on purpose. cgem156's LoRA family splits "which LoRA, at what weight" from "apply it to a model" into two separate steps, and this node is only the first half - it packages a LoRA choice and its strengths into a LoRA object you pass along to something else.

Why split it this way

Once a LoRA's name and weight are captured as a standalone object, you can do more with it than a normal loader lets you: feed it into LoraMerger to combine with a second LoRA, or into LoraLoaderFromWeight to finally apply it to a model. It's the same LoRA definition, reusable across whichever downstream node needs it, instead of re-picking the LoRA and re-typing the strength every time.

Inputs and outputs that matter

• lora_name - pick from your installed LoRAs, standard dropdown.
• strength_model (default 1, range −20 to 20) and strength_clip (default 1, same range) - the usual two strength dials, separate for the UNet and the text encoder, same convention as the stock LoRA loader.
• lbw - a text field for a LoRA Block Weight preset/ratio string, left blank by default. This follows the convention from hako-mikan's original sd-webui-lora-block-weight extension for A1111, where a preset string (things like ALL, or a comma-separated per-block ratio) lets you apply a LoRA more strongly at some U-Net depths than others instead of one uniform strength. The pack's own docs don't spell out the exact syntax this field expects, so if you want to go past "leave it blank," look up hako-mikan's block-weight preset conventions as your reference.
• Output: LoRA - a definition object, not a patched model. Wire it into LoraLoaderFromWeight (to apply it) or LoraMerger (to combine it with another LoRA).

Installing it

ComfyUI Manager → search cgem156-ComfyUI🍌 → install → restart. Or:

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

then restart ComfyUI. No extra downloads for this node.

Common issues & troubleshooting

Nothing happens - no model changes. That's correct behavior, not a bug. This node only defines the LoRA and its weight; it has no model input to patch. If you want the effect on your image, you still need LoraLoaderFromWeight (or LoraMerger) downstream consuming its LoRA output.

Confusing this with the standard LoRA loader. If all you want is "load one LoRA onto one model," the built-in LoraLoader node is simpler and does it in one step. Reach for this split version specifically when you want to merge LoRAs together first or reuse the same LoRA definition in more than one place.

lbw field doing nothing visible. Leave it blank unless you specifically intend to use block-weight ratios - an malformed or unsupported string here is more likely to be silently ignored or applied unevenly than to error clearly, so don't add it "just in case."

Strength out of the useful range. ±20 is the slider's technical limit, but useful values for most LoRAs sit much closer to 0–1.5; cranking either strength toward the extreme ends is a fast way to get artifacts, not a stronger effect.