RM Power LoRA Loader

Take the RM Power LoRA Loader, and instead of pasting LoRA strings into a Style Import box, hand it an RM_STYLES_PIPE and let it pull the LoRAs out of the pipe for you. That's the whole difference, and it's a workflow-packing move: one thick wire carries your prompt, negatives, motion prompt and LoRA lists, and this node slices out exactly the LoRA list you want.

How it works

The styles pipe is a tuple with a fixed layout - positive, negative, motion, image LoRAs, motion LoRAs high, motion LoRAs low, prompt number, prompt name. This node reads that pipe, picks one of the three LoRA fields via the lora_source dropdown, and feeds it into the exact same stack-building engine as the plain Power LoRA Loader. So you still get all the dynamic UI - Add Lora rows, per-LoRA toggles, random strength ranges, Civitai AIR downloads - on top of the pipe-fed base list.

Where this shines is video workflows. "Motion Loras High" and "Motion Loras Low" aren't just names - the author's styles file distinguishes stronger and weaker motion LoRAs, and with this node you can swap which tier is active by changing one enum instead of rewiring the graph. Each selected list is still parsed as <lora:name:weight> strings, so any invalid filename is skipped with a console warning rather than killing the run.

Inputs and outputs that matter

• Styles Pipe (RM_STYLES_PIPE, optional) - produced by the pack's RM Styles Pipe node, or by anything else that emits the same pipe type.
• lora_source (enum, optional) - Image Loras, Motion Loras High, or Motion Loras Low. This is the field you'll actually change between runs.
• Lora Stack (LORA_STACK, output) - feed into RM LoRA Apply.

There's no Style Import string input on this variant; the pipe is the source. If you want both a pipe and a manual string, that's what RM LoRA Collector is for - merge this node's stack with a string from somewhere else.

How to install it

Part of the RMAutomation pack - one install, all nodes:

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

Restart and it appears under RMAutomation/LoRA. No extra Python deps beyond what the base loader needs (requests/tqdm if you use AIR downloads).

Common issues

The pipe is order-sensitive - if you connect a pipe from a different node pack, confirm it packs values in the same order (positive, negative, motion, three LoRA lists, number, name). Wrong order means the node silently reads the wrong field as LoRAs. Also remember the pipe only contains whatever the style node emitted; if lora_source is set to Motion Loras High but your styles file has an empty motionLorasHigh array, the stack comes back empty and the model runs with no LoRAs at all. A quick sanity check: peek at the console log, which lists exactly what got loaded.