Dynamic LoRA Stack
Stack unlimited Wan LoRAs without rebuilding the node — made for WanVideoWrapper
- lora_stack
If you run Wan video locally through kijai's WanVideoWrapper, you know its LoRA stacking can be a chore: the graph wants a WANVIDLORA stack, and most stock nodes give you a fixed number of slots. Add a third LoRA and you're rebuilding wiring. Dynamic LoRA Stack is MoonPack's answer - auto-expanding WANVIDLORA slots that grow as you plug more in, flattening nested lists and never choking on an empty stack.
It's a small node with a narrow job, and it's aimed squarely at one audience: people stacking multiple LoRAs onto a Wan 2.x video pipeline (the pack's README says "designed for kijai's WanVideoWrapper"). If you're not doing Wan video, this node does nothing for you. If you are, it removes a recurring annoyance.
How it works
The node starts with a single auto-expanding slot and the frontend adds more as you connect things - the same dynamic-input machinery the pack uses for its string concat node, but typed WANVIDLORA instead of STRING. Server-side, it gathers every connected slot in order, and here's the part that matters: it flattens. If a slot receives a list of LoRA entries (which happens when you chain another stack into it), the node extends the output rather than nesting lists. That composability means you can build a "base" stack node and merge it with a "style" stack node and get one flat stack out.
The other deliberate behavior: an empty stack returns [], not None. That sounds pedantic until you've debugged a workflow where a null blew up a downstream loader. Returning an empty list keeps the graph running and lets other nodes check length cleanly.
The output
One output, lora_stack (type WANVIDLORA) - a flat list of LoRA entries ready for WanVideoWrapper's loader. That's the whole deal: wire your LoRA loaders in, take lora_stack out, feed it downstream.
A word on Wan context
Wan 2.x (frozen at 2.2 since mid-2025, per the KB's model panel) is the open base most local video work consolidates on, and LoRA stacking is part of the workflow - speed LoRAs on the low-noise expert, style LoRAs on top. If you're tuning a Wan 2.2 pipeline with several LoRAs, the fixed-slot alternative will frustrate you within a session. This node exists to make that specific workflow painless.
Installing it
Standard MoonPack install:
cd ComfyUI/custom_nodes
git clone https://github.com/moonwhaler/comfyui-moonpack.git
or ComfyUI Manager → MoonPack → install, restart. Under MoonPack/lora. No models to download, no extra Python dependencies - though it only does its job if you already have WanVideoWrapper (kijai) installed for the WANVIDLORA type.
Verdict
Niche by design - Wan-only, LoRA-only. But within that niche it's genuinely the right tool: auto-expanding slots, flattening, and []-not-None are exactly the three behaviors that stop you from rebuilding a stack node every time you add a LoRA.
Inputs (0)
No inputs
Outputs (1)
| Name | Type | Description |
|---|---|---|
| lora_stack | WANVIDLORA | — |