Merge LoRA Stacks

A lora_stack in ComfyUI is just a list of LoRA entries - (lora_name, strength_model, strength_clip) tuples that some node will eventually apply to the model. Stacker nodes bundle several LoRAs into one of those lists, and that's great until you have two of them. A character pack in one stacker, a style pack in another, and suddenly you're stuck feeding two lists somewhere that wants one.

Merge LoRA Stacks is the glue. It takes however many lora_stack inputs you care to connect and concatenates them into a single LORA_STACK output. That's the whole job, and it's exactly the sort of small plumbing node that keeps a big workflow from becoming a tangle.

How it works

The left-side inputs are dynamic (the V3 node API's "Autogrow" - click the + and another slot appears, up to a hard limit of 100). Under the hood it sorts the slots by index (lora_stack_0, then lora_stack_1, …) and concatenates the entries in that order, so the order you wire them is the order they end up applied.

Two behaviors are worth knowing because they're quietly useful:

• Strength-zero entries are dropped. If a stacker has a LoRA toggled off (strength_model == 0), that entry is filtered out of the merged result instead of being carried through. That's almost always what you want - it means you can keep LoRAs loaded in a stacker and switch them off without them polluting the merge.
• Empty or None inputs are skipped rather than erroring, so an unplugged slot doesn't break the whole merge.

The single lora_stack output uses the same type identifier as the classic V1 LORA_STACK nodes, so it wires into the usual suspects - the easy-use and lora-optimizer families, or anything else that consumes a stack. If you're not already thinking in stacks, the thing to know is that a stack is just a convenience wrapper over "apply these LoRAs at these strengths," and this node lets you compose those bundles instead of hand-editing one giant list.

Why you'd reach for it

The honest use case is organization. Instead of one monster stacker with twelve LoRAs and a wall of per-entry toggles - the Power Lora Loader approach, which is great until it isn't - you get to keep logical groups as separate stackers (character, style, clothing) and merge them at the point where they actually apply. It also makes A/B testing tidier: swap one stacker's group in or out and the merge just rebuilds.

Install

Standard for this pack - ComfyUI Manager (search "ComfyUI-zyd232-Nodes") or:

cd ComfyUI/custom_nodes
git clone https://github.com/zyd232/ComfyUI-zyd232-Nodes.git

Restart, find Merge LoRA Stacks under zyd232 Nodes. No dependencies beyond what ComfyUI already has. Same caveat as the rest of the pack: it's written against the newer V3 backend node API, so keep ComfyUI updated or the node won't appear.

Bottom line

A tiny, well-behaved utility that does one thing: combine LoRA stacks in order, drop the disabled entries, pass the result along. If you've ever wished you could organize your LoRAs into groups instead of one endless list, this is the missing join node.