Load Wan Fun Lora

If you've looked at LoadWan2_2Lora or LoadWan2_2FunLora and wondered why they need two LoRA slots, the answer is Wan 2.2's split high-noise/low-noise expert architecture. LoadWanFunLora doesn't have that problem - it's for the Wan 2.1-Fun family loaded through LoadWanFunModel, and Wan 2.1 is a single dense model, not a mixture of experts. So this node has exactly one LoRA slot, and that's not a missing feature, it's just what the underlying architecture calls for.

How it works

Standard LoRA-merge pattern: take a funmodels bundle, apply one LoRA at a given strength, return the patched bundle for the sampler downstream.

The inputs that matter

• funmodels (required) - from LoadWanFunModel.
• lora_name - pulled from your loras/ folder.
• strength_model - default 1, range -100 to 100. Most people stay well inside 0.5-1.2; the far ends of that range are either invisible or badly distorted.
• lora_cache - default false. Enable it if you're re-running the same LoRA/strength combination repeatedly and want to skip the re-merge cost each time.

Output: funmodels, patched, into WanI2VSampler or WanT2VSampler.

Installing it

No separate install - it's part of the VideoX-Fun pack. Search ComfyUI Manager for VideoX-Fun, and if it's not indexed there:

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

Restart ComfyUI to pick it up.

Common issues

Applying a Wan 2.2 LoRA here and getting weak or broken results. Wan LoRAs trained on 2.1 are broadly reported as forward-compatible with 2.2, but the reverse isn't guaranteed - a LoRA trained specifically against Wan 2.2's split-expert setup may not transfer cleanly onto the single Wan 2.1 model this node targets. If a LoRA does nothing, check which generation it was actually trained for before assuming the node is broken.

Dropdown's empty. LoRA files need to be in ComfyUI's standard loras/ folder, and the list only populates on startup - restart after adding new files.

Strength feels off. Video LoRAs in general tend to degrade motion smoothness before they finish "kicking in" at high strength, more so than image LoRAs do. If 1.0 looks overcooked, dial back to 0.6-0.8 rather than assuming the training itself was bad.

Should you be on Wan 2.1-Fun at all in 2026? Worth being honest about: Wan 2.2 is the stronger base across motion quality, emotional expressiveness, and prompt adherence, and it's where the bulk of new LoRA training is happening. Stick with 2.1-Fun if you've got an existing pipeline built on it or specifically need the lighter 1.3B tier - otherwise LoadWan2_2FunLora paired with LoadWan2_2FunModel is the more actively-supported path.