RM LoRA Collector

The RM LoRA nodes split the job in three: a loader builds a stack, an apply node uses it, and RMLoraCollector is the plumbing in between that lets you combine stacks. Its whole job is merging up to five LORA_STACK inputs and up to three lora_string inputs into one stack. Chainable, boring, and quietly essential when your workflow has LoRAs coming from two different directions.

How it works

It concatenates every LORA_STACK input it receives (stacks are just lists, so merging is list extension), then parses any lora_string inputs with the same <lora:name:weight> syntax used elsewhere in the pack and appends those entries too. Invalid filenames in a string get skipped with a console warning. The output is a single LORA_STACK you can feed straight into RM LoRA Apply - or into another Collector, which is how you build arbitrary merge trees.

Why would you bother when RM LoRA Apply already accepts multiple stacks? Two reasons. First, the Collector gives you a single point where any stack source can join: a styles-pipe stack from the Power LoRA Loader (Pipe), a manual stack from the plain loader, and a raw string from a text node. Second, it decouples the merge from the apply - you can route the combined stack anywhere, preview it, or send the same merged set to two different applies.

Inputs and outputs that matter

• lora_stack_1 through lora_stack_5 (LORA_STACK, optional) - connect up to five stacks.
• lora_string_1 through lora_string_3 (STRING, optional) - <lora:name:weight> strings.
• Output: lora_stack (LORA_STACK) - one merged stack.

How to install it

Ships in the RMAutomation pack:

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

Restart ComfyUI; it's under RMAutomation/LoRA. No dependencies.

Common issues

Order is the one thing to keep an eye on: inputs are concatenated in fixed order (stack 1→5, then strings), so if you care which LoRA applies first - and you should, order affects the result - connect your most structural LoRA stack into lora_stack_1. The collector is a merge, not a dedupe: if the same LoRA shows up in two stacks it gets applied twice at both weights, which is usually an accident. And as with everything in the stack system, this node only builds data - no LoRA actually touches your model until the stack hits an RM LoRA Apply.