RM Power LoRA Loader

This is the marquee node of the RMAutomation pack, and it's an honest clone of a very good idea. The README credits rgthree's Power LoRA Loader - the single most-borrowed node in that ecosystem - and the concept is the same: instead of chaining ten separate LoraLoader nodes and apologizing for the spaghetti, you get one node with an Add Lora button, a row per LoRA, and a LORA_STACK output you hand to a single apply step. Where this one goes further than the original: per-LoRA random strength, a style-string import, and a "Civitai AIR" mode that downloads LoRAs by model ID on the spot.

How it works

Each row you add is a dict: the LoRA file, a strength, an on/off toggle, and a random-strength mode with a min/max range. When random is on, the node rolls a fresh weight from that range every execution and forces a re-run so you get a new draw each time. Everything compiles into a LORA_STACK - a list of {name, strength, ...} dicts - which is just data; nothing is applied to a model yet. That's what RM LoRA Apply is for.

Three ways to get LoRAs in:

1. Pick from the dropdown - the node lists everything in your loras folder.
2. Style Import - paste a <lora:filename:weight> string (the same format RM Styles Full emits from its "Image Loras" output) and the node parses it into stack entries automatically.
3. Civitai AIR - enter a Civitai model ID (or id@version), and the node fetches the model page, downloads the .safetensors into your loras folder, verifies the SHA256, and logs it in a download-history file so it never re-downloads. Gated or NSFW-flagged models need a Civitai API key, and a 401 is the giveaway.

Inputs and outputs that matter

The node is built around dynamic inputs, so the only fixed ones are:

• Style Import (STRING, optional) - the <lora:...> string mentioned above.
• Lora Stack (LORA_STACK, output) - feed this into RM LoRA Apply, or into RM LoRA Collector if you want to merge it with another stack.

How to install it

The whole pack installs the same way:

cd ComfyUI/custom_nodes
git clone https://github.com/Moser9815/ComfyUI-RMAutomation

Restart ComfyUI and it's under RMAutomation/LoRA. The pack ships no requirements.txt, so if the AIR downloader fails with an import error, install requests and tqdm into your ComfyUI Python environment manually:

pip install requests tqdm

Common issues

"LoRA not found" in the console means the filename doesn't match anything in your loras folder - check the exact path. AIR downloads that hit 401 need a Civitai API key (the node accepts one as an input) because Civitai now gates NSFW/gated models behind auth. And remember the stack only lists LoRAs - if nothing changes in your output, check that you actually connected Lora Stack into an RM LoRA Apply and that the apply node is on the model path. Also worth knowing, from the general LoRA playbook: weights are not additive fun - random strength is great for exploration, but a 0.5–1.0 random band on three LoRAs at once will give you results all over the map. Start with one randomized and the rest fixed.