List LoRA Loader

Stacking two or three LoRAs is easy with a chain of the stock LoRA loader node. Stacking eight, remembering what each one's trigger word is, and toggling half of them off for a quick A/B test is where the stock node falls apart. List LoRA Loader is the CrasH Utils answer: one node, a folder browser, and a vertical list where every LoRA shows its name and trigger keywords right next to a toggle.

What it is and where it fits

This isn't a new idea - the ComfyUI ecosystem already has a well-loved precedent in rgthree's Power Lora Loader, which does per-entry toggles and trigger-word lookups in a single stacked node, and the CrasH Utils README credits that pack directly as the inspiration. What List LoRA Loader adds on top is a folder browser with search, an NSFW-folder filter, and automatic trigger-word extraction from companion text files, in a compact list layout rather than a thumbnail grid (that's its sibling, Grid LoRA Loader). If your LoRA library is large and well-named, the list view is the one you'll actually want - you can scan names faster than you can scan a wall of thumbnails.

How it works

Browse your LoRA folders inside the node itself, search, and double-click to add a LoRA to your active stack. Each active entry gets its own enable/disable toggle and a strength control, so you can drop a LoRA out of the mix without deleting it from the graph. There's a dedicated SFW toggle that hides any folder named NSFW (case-insensitive, at any depth) from browsing and search - flip it off if you need to see those folders again.

Trigger words come from a companion .txt file next to each .safetensors, same base filename, one word per line or comma-separated. A Trigger Tags panel lets you pick which of those get sent to the triggers output - so you can load a LoRA without automatically dumping every one of its trigger words into your prompt.

Inputs and outputs that matter

Both real inputs are optional: model and clip, which is where you wire in your checkpoint's model and CLIP so the loaded LoRAs actually get applied. In normal use you'll always connect both - leaving either one out just means that half of the LoRA patch (model or text-encoder side) never happens.

Outputs: MODEL and CLIP - the patched versions, ready for your sampler and text encoder - plus triggers, a comma-separated STRING of whichever trigger tags you toggled on. Wire triggers into a text-concat node feeding your positive prompt if you want it automatic instead of copy-pasting trigger words by hand.

Installing it

ComfyUI Manager: search "CrasH Utils". Manual:

cd ComfyUI/custom_nodes/
git clone https://github.com/chrish-slingshot/CrasHUtils.git

Restart ComfyUI. No extra dependencies.

Common issues

LoRAs vanished from the browser. Check the SFW toggle first - if any of your folders have "NSFW" in the name, it hides them at any depth, silently.

Trigger words not showing up. The .txt file needs the exact same base name as the .safetensors file and needs to sit right next to it. A mismatch (even just a typo) means the loader won't find it.

A loaded LoRA seems to do nothing. This is the general LoRA checklist, not specific to this node: confirm the LoRA actually matches your checkpoint's architecture (an SDXL LoRA does nothing on Flux, and Pony/Illustrious/SDXL LoRAs are often not cross-compatible despite sharing code), that its trigger word is actually in your prompt, and that its weight isn't stuck too high or too low for that particular LoRA - the old "0.5–0.8" rule of thumb is SDXL-era advice and doesn't generalize to every architecture.

Thumbnails not appearing (Grid view specifically). Only relevant if you're using the Grid variant - List LoRA Loader doesn't show thumbnails by design, so if you want the visual browser, that's the sibling node, not this one.