WanVideoLoraStack

Anyone who's done Wan video for more than a week ends up with a pile of LoRAs: a character LoRA here, a style LoRA there, maybe a lightx2v speed LoRA if they're impatient. Wiring five LoRA loaders into a sampler is miserable, and it only gets worse when you realize Wan 2.2 runs two models (high-noise and low-noise) that both need the same LoRAs applied. WanVideoLoraStack is the pack's answer: you build a list of LoRAs in one node, chain as many as you want, and hand the whole stack to the sampler over a single wire.

What it actually does

Fair warning up front: it does not load LoRAs. It builds a stack - a Python list of [name, model_strength, clip_strength] entries - that the pack's WanImageToVideoAdvancedSampler reads from its lora_stack input and applies for you, to both the high-noise and low-noise models plus the CLIP. If you wire this into a plain KSampler, nothing happens, because nothing here is doing the loading. It's a list builder in service of the pack's big sampler.

The inputs, all familiar if you've used ComfyUI's own LoraStack:

• lora_name - dropdown populated from your models/loras folder (plus a "None" option).
• strength_model - how hard it hits the diffusion model, -100 to 100, default 1. Negative strengths are allowed and genuinely useful for de-emphasizing something a LoRA added.
• strength_clip - same range for the CLIP/text side, default 1.
• previous_lora - optional. Chain stack into stack: feed one node's output into the next's previous_lora and the new entry gets appended, so you build up a stack across as many nodes as you like.

Output is a single lora wire. Two small behaviors worth knowing: if both strengths are 0, the node returns the previous stack untouched (handy for bypassing a LoRA without deleting the node), and the lora_name list comes straight from what's on disk - drop a file in models/loras and it shows up after a refresh.

The Wan-specific advice

Since you're stacking LoRAs for Wan, a couple of community lessons apply. Wan 2.1 LoRAs carry forward to 2.2, so don't assume you need 2.2-native files. Speed LoRAs (lightx2v, CausVid) are the standard acceleration, but they're a known quality trade: the community's landmark warning was that they "destroy everything that's good" in Wan 2.2 - composition, lighting, motion, skin texture. The practical compromise most people landed on is applying speed LoRAs to the low-noise pass only, at reduced strength like 0.4-0.6, and keeping the high-noise pass clean. This node hands equal strengths to both passes, so if you're chasing full quality, be deliberate about what you stack rather than throwing every LoRA you own at it at strength 1.

Install

From przewodo/ComfyUI-Przewodo-Utils - ComfyUI Manager search "ComfyUI-Przewodo-Utils", or:

cd ComfyUI/custom_nodes
git clone https://github.com/przewodo/ComfyUI-Przewodo-Utils.git

Restart, and it's under PrzewodoUtils/Wan. No model downloads for the node itself - just make sure your LoRA files live in ComfyUI/models/loras, since the dropdown reads that folder directly. And if you wire a stack in and see "no valid LoRA name" in the console, that's the sampler telling you an entry had lora_name: None - check which node in the chain actually had an empty dropdown.