Power Lora Loader

Power Lora Loader is the flagship of the EasySettingpipes pack: one node where you stack LoRAs the way rgthree's famous loader does - add entries, toggle them on and off, drag them into order, dial in a strength per entry - instead of chaining a row of stock LoraLoader nodes across your canvas. If that pitch sounds familiar, it's because this is a lightweight reimplementation of the pattern rgthree made famous, shipped here under a permissive MIT license. The honest framing up front: if you already run rgthree-comfy (most people do), you've got the original Power Lora Loader and you don't strictly need this one. This version is for people who want stacked-LoRA management without installing the whole rgthree pack, or who are already using this pack's SamplerSetup pipe and want the LoRA side in the same place.

How it works

The node has exactly two real sockets - model and clip, both optional - and outputs MODEL and CLIP. The LoRA list is not a set of inputs at all. It's widgets drawn by the pack's own frontend code, and each entry serializes into a small config (name, on/off, model strength, clip strength) that arrives on the backend as lora_1, lora_2, and so on. The loader filters for valid configs, skips anything toggled off, and applies each one with ComfyUI's own load_lora_for_models. Two details worth knowing: a zero-strength entry is skipped entirely, so no file is even read - a free performance win if you keep a LoRA parked at 0 while you compare - and each node instance caches the last file it loaded, so re-running doesn't re-read the same .safetensors off disk.

The strength controls have two modes. Single mode applies one number to both the model and the CLIP encoder. Click the mode toggle and you get two: orange for model, cyan for clip. That matters more than it sounds - clip-only LoRAs exist, and so do ones where you want a strong model weight but a light touch on the text encoder.

Right-clicking an entry opens an info dialog that reads trained words straight out of the safetensors header (ss_tag_frequency), and can also ask CivitAI's by-hash endpoint for the model card's trigger words. That's a port of rgthree's approach, and it's the same trick the modidex troubleshooting doc points to when a LoRA appears to "do nothing" because its trigger word is missing. It needs a working network connection to reach CivitAI, and hashing a big file takes a couple of seconds the first time.

Installing it

This is the one node in the pack that ships real UI, but it installs the same as everything else:

cd ComfyUI/custom_nodes
git clone https://github.com/hayde0096/Comfyui-EasySettingpipes

Then restart ComfyUI. Manager works too - search "Easy Setting Pipes" and hit install. There's no requirements.txt, no model downloads, nothing heavy: everything it imports (aiohttp, requests) ships with ComfyUI already. Beware one trap: the README's own manual-install snippet points at github.com/user-attachments/..., which is a copy-paste bug and won't clone. Use the repo URL above.

Where people get burned

• This pack has effectively zero community footprint. I searched; nobody on reddit is discussing it. Bugs and questions go to the GitHub repo, and a single-maintainer pack can answer slowly.
• Nodes 2.0. Dynamic-widget nodes like this one are exactly what broke under ComfyUI's Nodes 2.0 frontend for rgthree (its issue 632). If the rows render as stacked text or the properties won't open, switch Nodes 2.0 off before filing a bug.
• "My LoRA does nothing." Run the standard checklist before blaming the node: base-model compatibility (a Flux LoRA won't apply to SDXL), the trigger word, and the weight - 1.0 is often too strong, 0.5–0.8 is the common range.