Load Wan 2.2 Fun Lora

The thing that trips people up about applying LoRAs to Wan 2.2 is that there isn't one model to patch - there are two. Wan 2.2 is a mixture-of-experts setup: a high-noise expert that handles motion and scene composition, and a low-noise expert that refines detail, and they run as separate passes. That's why this node has two LoRA slots instead of one - lora_name and lora_high_name - where the single-slot loaders elsewhere in this pack only have one. If you've read the community wisdom on Wan 2.2 speed LoRAs (apply them to the low-noise pass only, keep the high-noise pass clean, because that's what preserves motion and composition quality), this node is where that advice actually gets implemented in the graph.

How it works

Feed it a funmodels from LoadWan2_2FunModel, pick a LoRA for the low-noise pass, optionally pick a second one for the high-noise pass, set a strength, and it returns the patched bundle. Leave lora_high_name unset if you only want to touch one expert - that matches the "speed LoRA on low-noise only" pattern the wider Wan community settled on.

The inputs that matter

• funmodels (required) - from LoadWan2_2FunModel.
• lora_name - the LoRA applied to the low-noise (detail) expert.
• lora_high_name - the LoRA applied to the high-noise (motion/composition) expert. Both dropdowns pull from your standard loras/ folder.
• strength_model - default 1, range -100 to 100; applies to both slots.
• lora_cache - default false. Turn it on if you're iterating on prompts without changing the LoRAs themselves, to skip re-merging weights every run.

Output: funmodels, patched, into Wan2_2FunInpaintSampler or Wan2_2FunV2VSampler.

Installing it

Part of the base VideoX-Fun pack - no separate install. Search ComfyUI Manager for VideoX-Fun, and if it's not listed (this pack isn't always reliably indexed there):

cd ComfyUI/custom_nodes
git clone https://github.com/aigc-apps/VideoX-Fun
pip install -r VideoX-Fun/requirements.txt

Restart ComfyUI afterward.

Common issues

You only wanted to apply one LoRA and got confused by the second slot. That's fine and normal - leave lora_high_name blank and it only touches the low-noise expert. This matches the well-documented community pattern for speed LoRAs specifically: putting a step-distillation LoRA on the high-noise pass tends to wreck composition and motion quality, so most people leave that slot alone and only patch low-noise.

Both LoRAs same file, different effect than expected. You can use the same LoRA in both slots if it was trained generically, but character or style LoRAs trained against one expert don't necessarily transfer cleanly to the other - the two experts see different noise regimes during training. If results look off, try clearing lora_high_name first before assuming the LoRA is broken.

Empty dropdowns. Standard fix - confirm the LoRA files are actually in ComfyUI's loras/ folder and restart ComfyUI; the list doesn't refresh live.

This node is for the "Fun" checkpoints specifically. If you're running plain (non-Fun) Wan 2.2 through LoadWan2_2Model, use LoadWan2_2Lora instead - same two-slot idea, different model family, and the two aren't interchangeable.