Load LoRA

ComfyUI already ships a LoraLoader, so a new one has to justify itself. HYBS_LoadLoRA justifies itself with two things: a NONE passthrough and a string output that reports what it applied. If you've ever compared "the same image with and without this LoRA" by duplicating a whole branch of the graph, you'll get the appeal immediately - this node makes the with/without comparison a dropdown flip instead of a graph edit.

Mechanically it's a standard single-LoRA loader: give it a model and clip, pick lora_name from your models/loras folder, and set the two strengths. Under the hood it applies the LoRA through comfy.sd.load_lora, falling back to the built-in LoraLoader class and then a low-level apply_lora path if the API shifts under it - the author clearly wants this to keep working across ComfyUI versions. When lora_name is NONE (or null/empty), nothing is applied and your input model and CLIP pass through byte-for-byte unchanged.

The inputs that matter

• model / clip - the base values the LoRA patches
• lora_name - dropdown of models/loras files; defaults to NONE
• strength_model / strength_clip - LoRA strength, range -20 to 20, step 0.05, default 1.0

The range is wider than you'll ever want (the KB's advice about 0.5–0.8 being the common sweet spot still holds for SDXL-family LoRAs), but it means negative-strength experiments are on the table without changing nodes.

Why the third output is the point

• model - MODEL after the LoRA is applied
• clip - CLIP after the LoRA is applied
• applied lora - a STRING with the applied LoRA token, or "" for NONE

That string comes out formatted like <lora:filename:strength_model:strength_clip> - the exact token syntax you'd type into a prompt in other UIs. Wire it into a text-concatenation node before CLIP encoding and your recorded prompt automatically reflects which LoRA actually ran. That's genuinely handy for logging, A/B runs, and keeping workflow shares honest.

Installation

Standard pack install:

cd path/to/ComfyUI/custom_nodes
git clone https://github.com/hybskgks28275/ComfyUI-hybs-nodes.git
pip install -r ComfyUI-hybs-nodes/requirements.txt

Or "ComfyUI-hybs-nodes" via ComfyUI Manager and restart. No extra models, no heavy deps - the pack's toml requirement belongs to a different node.

The honest verdict: for one LoRA in a stable graph, the built-in loader is fine and you already have it. Reach for this one when the reporting matters - comparing applied vs. not, generating the prompt token automatically, or keeping a workflow where the "no LoRA" baseline has to coexist with LoRA versions in the same canvas.