Power LoRA Configurator

Power LoRA Configurator is the "describe one LoRA" node in the CornmeisterNL PowerPack. It doesn't apply anything to your model - it builds a reusable config card for a single LoRA, then hands that card to the pack's Power LoRA Selector, which does the actual applying. If you've ever wired up five LoRAs and lost track of which wire goes where, this is the fix: each configurator is a labeled card, and the label takes care of itself.

The neat bit is the self-naming. The node has a trigger field - the word you'd inject into your prompt to activate the LoRA's effect (rare-token trigger words are the standard trick for character and style LoRAs, since the model keys on them). As you type, a tiny frontend script renames the node's title to that trigger word. So you end up with a canvas full of nodes literally titled "zephyr_style", "sarah_v3", "vintage_paper" instead of five identical "Power LoRA Configurator" boxes. It's a small quality-of-life thing, but once you have a dozen LoRAs in a character-testing workflow, it's the difference between reading your graph and squinting at it.

What it does mechanically

It's a thin data builder. Pick a LoRA from the dropdown (populated from models/loras, with a (none) option to skip), set your strengths, and it emits a LORA_CFG object - a small dict carrying the filename, trigger, and both strengths. That object is the pack's private data type, so the Configurator is only useful wired into the Selector (or into Power Text Concat's trigger input, where the trigger word gets picked up as the source label).

Inputs that matter

• lora - dropdown of everything in models/loras. (none) means "no LoRA, pass through".
• trigger - the prompt word associated with this LoRA. Doubles as the node's title.
• strength_model and strength_clip - both float, default 1.0, range −5 to +5 in 0.05 steps. Model strength is the big knob; CLIP strength affects the text-encoder side and is usually left at 1.0. The old "0.5–0.8 is the sweet spot" advice is SDXL-era; on newer architectures people routinely run 1.0 and above, so trust the LoRA's own page over a global rule. Negative values exist for "unlearning" effects - you'll rarely touch them.

Output is one wire: cfg (LORA_CFG), which goes into the Selector's cfg_1 (and cfg_2, cfg_3, …) inputs.

Install

Part of the CornmeisterNL PowerPack - install once for the whole set:

cd ComfyUI/custom_nodes
git clone https://github.com/Cornmeister/ComfyUI-CornmeisterNL-PowerPack.git

Restart ComfyUI (console should print ⚡ [CornmeisterNL Powerpack] Backend loaded), or just use ComfyUI Manager and search "CornmeisterNL PowerPack". Only requests and Pillow under the hood; nothing heavy.

Where people get burned

• The node alone does nothing. If you wire a Configurator straight into a KSampler, nothing gets applied. It outputs a config, not a model - the Selector is the missing link, and that's by design.
• Confusing title changes. The title becomes the trigger, so if you leave trigger empty the node just says "Power LoRA Configurator". Give every LoRA a trigger even if you don't use the word in prompts - it's your label.
• Expecting it to merge multiple LoRAs. It configures one. Stacking happens on the Selector side, which picks exactly one active config at a time. If you want three LoRAs simultaneously, this isn't the pack for that job.

The Configurator is boring in the best way: it captures the facts about a LoRA in one place, labels itself, and gets out of the way. The clever part is what the Selector does with those cards.