Quick Lora Stack

Most LoRA work doesn't need a clip weight. The full-fat LoRA loader gives you a model weight and a clip weight, and 95% of the time you set both to the same number anyway - which is exactly what Sage Utils' full Lora Stack node makes you do explicitly, knob by knob.

Quick Lora Stack is the "just let them match" version. One LoRA, an on/off switch, one weight. That's it. It feeds the exact same lora_stack type as the full node, so it drops into the same downstream tooling.

How it works

Three inputs:

• enabled - boolean, default on. Off means the entry is skipped and the incoming stack passes through untouched.
• lora_name - the LoRA file, from the standard loras dropdown.
• model_weight - one weight, default 1.0, range -100 to 100.

Under the hood the "quick" part is literal: add_lora_to_stack(lora_name, model_weight, model_weight, ...) - the same weight is written to both the model and clip slots of the stack entry. The optional lora_stack input lets you chain stacks, appending this entry onto whatever stack came before.

The output is lora_stack (LORA_STACK type). From there it goes into a loader that actually applies the LoRAs - Sage's Load Models + Loras or the Lora Stack Loader - or into Construct Metadata if you want the LoRA list recorded in the PNG's metadata.

Why the boolean matters

The on/off switch is the reason this node family exists at all. The author built these after a Reddit thread where someone asked for "LoRA loader nodes with boolean on/off switch" - they mentioned their own node set with boolean-toggled LoRA stacks in the thread. It's a real workflow need: flip a character LoRA on or off between runs without deleting wires or zeroing weights. Zero is a proxy; a switch is a switch.

Installation

Sage Utils, one install, as always. ComfyUI Manager, search "Sage Utils", or:

cd ComfyUI/custom_nodes
git clone https://github.com/arcum42/ComfyUI_SageUtils
cd ComfyUI_SageUtils
pip install -r requirements.txt

Restart ComfyUI. Only dependency is dynamicprompts; no models to download.

The honest take

The one real downside of the quick version: no separate clip weight. If you're doing SD 1.5-era tricks where clip and model weights genuinely differ (or a LoRA's card explicitly says "clip weight 0.5"), this node can't express that - use the full Lora Stack node instead. Otherwise, one weight is exactly the right number of weights, and the boolean keeps your workflow readable. Compatible with other packs' lora_stack types too, per the author's design - worth remembering when you're mixing node suites.