Lora Loader Stack (rgthree)

Let's get the awkward part out of the way first: rgthree himself has put a line through this node's name in the README. It's officially deprecated, and the note next to it says to use the Power Lora Loader instead. So if you're landing here from a search, the honest answer is that this is the older, simpler tool and there's a nicer one in the same pack. But it still works fine, and there are reasons you might genuinely want the simple one, so let's actually talk about it.

What it does

It loads up to four LoRAs in a single node instead of chaining four separate LoraLoader nodes across your canvas. That's the whole pitch. A LoRA is a small add-on file that nudges a checkpoint toward a style, character, or concept, and it's normal to stack two or three at once. Doing that with stock nodes means a little conga line of loaders, each one taking a MODEL and CLIP in and handing them to the next. This collapses that line into one box.

Under the hood it's doing exactly what the stock loader does, four times in sequence: each LoRA it applies patches the model weights and (optionally) the CLIP weights, and the result feeds into the next slot. Nothing clever, just tidier.

The inputs and outputs that matter

You wire a model and a clip into the top (usually straight from your Load Checkpoint), and then for each slot you set:

• lora_01 through lora_04 - a dropdown of the LoRA files in your models/loras folder. Leave any set to None to skip that slot.
• strength_01 through strength_04 - how hard each LoRA pulls, defaulting to 1.0. The range runs from -10 to 10, but you'll basically live between 0.4 and 1.2. Note this one strength value drives both the model and CLIP weight for that LoRA - there's no separate CLIP strength here, which is one of the things the newer Power Lora Loader adds.

Out the other side you get MODEL and CLIP, patched with every enabled LoRA. Those go on to your KSampler and your text encode nodes exactly like the checkpoint's own model and clip would.

How to install it

The node ships in rgthree-comfy, a quality-of-life pack that's in more or less every serious ComfyUI install. Easiest route is ComfyUI Manager: open it, search for rgthree's ComfyUI Nodes, install, restart. Or clone it by hand:

cd ComfyUI/custom_nodes
git clone https://github.com/rgthree/rgthree-comfy.git

Then restart ComfyUI. There are no models to download and no heavy Python dependencies - the whole pack is UI and graph plumbing, so install is genuinely just the clone and a restart. It's MIT-licensed and maintained by rgthree (Regis Gaughan, a Google engineer who builds this on the side), so it's a solo project, but a widely-trusted one.

Common issues

The big one is philosophical: you're on the deprecated node. If you find yourself wishing for more than four slots, wanting to toggle a LoRA on and off without re-selecting None, or wanting separate model/CLIP strengths, that's the exact list of things the Power Lora Loader was built to fix - it stacks unlimited LoRAs with per-entry toggles and, via its right-click info dialog, will even pull a LoRA's trigger words off CivitAI and keep a local copy. Worth the switch.

The other classic isn't the node's fault at all: a LoRA that seems to do nothing. Nine times out of ten that's a missing trigger word in your prompt, not a loading problem - plenty of LoRAs are inert until you type their keyword. Load it, generate, and if the effect's absent, go find the trigger word (the Power Lora Loader's info dialog is the community-favorite way to track those down).

Finally, if the node renders as a stack of broken-looking fields, check whether you've got ComfyUI's Nodes 2.0 frontend enabled - the Vue rewrite broke several rgthree nodes and the standing workaround is to leave Nodes 2.0 off. This node is simple enough to be less affected than the fancy loaders, but it's the first thing to rule out when any rgthree node looks wrong.