Power Lora Loader V2 (QQ)

Power Lora Loader V2 is the node you end up reaching for once you've wrestled with Wan 2.2's two-pass architecture and wondered why your LoRA stack needs to be applied twice. Wan 2.2 isn't one model you patch - it's a high-noise expert that handles motion and composition and a low-noise expert that handles detail, and the community routinely trains separate High and Low LoRA pairs for them. This loader is built around that split instead of pretending it doesn't exist.

The name is a mouthful and the "(QQ)" suffix just marks the pack, but the feature set is genuinely useful: a dynamic rgthree-style UI where you stack a bunch of LoRAs, auto-detection of Low-variant pairs from their High counterparts (the README's smart pattern matching handles -/_/space separators and case variants), separate strength sliders for the high and low side, and per-LoRA block selection.

The inputs that matter

The info_schema is refreshingly small for something this powerful:

• model (required) - your main Wan model, the one that comes out of the model loader.
• model_low (optional) - the low-noise expert model if your workflow hands it to you separately. This is where the two-pass trick lives: community best practice is to keep the high-noise pass clean and apply speed LoRAs like lightx2v to the low-noise pass at 0.4–0.6 strength (the "speed LoRAs kill everything good" PSA is one of the most-upvoted Wan threads for a reason).
• clip (optional) - wire it through if your graph passes a CLIP.

It outputs three things: MODEL and MODEL_LOW (both patched with whatever stack you built in the UI) and CLIP. You wire MODEL straight into your sampler or WanVideoWrapper pipeline and it just works.

Why this one instead of the older loader?

The pack also ships WanVideoPowerLoraLoader, which is the "v1" format. The practical difference: v1 outputs WANVIDLORA objects you're expected to chain into Kijai's WanVideoSetLoRAs node inside his ComfyUI-WanVideoWrapper graph. V2 collapses that - it returns ready-to-use patched MODELS. If you're building a workflow from scratch in the wrapper, v2 is fewer nodes and less ceremony. If you're editing an existing wrapper workflow that already has WanVideoSetLoRAs, keep it and use the older loader.

Installing it

This pack is the community "QQ" extension layer that sits on top of Kijai's ComfyUI-WanVideoWrapper - the README says it needs the KJ wrapper, and for actual Wan inference you'll want that installed plus the Wan 2.1/2.2 checkpoints, VAE, and text encoders. Install the pack itself via ComfyUI Manager (search WanVideoWrapper_QQ or SA-Nodes-QQ), or:

cd ComfyUI/custom_nodes
git clone https://github.com/siraxe/ComfyUI-WanVideoWrapper_QQ.git

Then restart ComfyUI.

Where people get burned

• The rename. The repo was renamed to ComfyUI-SA-Nodes-QQ in v1.3.4. Old workflow JSONs that reference the old wanwrapper_qq node pack can trigger an unintended install. If you see a stray wanwrapper_qq folder in custom_nodes, delete it and install the current pack.
• LoRAs not appearing. Drop files into ComfyUI/models/loras and refresh; the dynamic UI populates from there.
• Thinking speed LoRAs are free. They're not - applying lightx2v/CausVid to the high-noise pass is where Wan quality goes to die. Use the dual-strength sliders to keep the high side at 1.0 and the low side tame.

If you're on Wan 2.1 or run a single-model graph, model_low just stays empty and this behaves like a very good multi-LoRA loader. The magic is optional, which is the right design.