Load CogVideoX-Fun Lora

Once you've got a LoadCogVideoXFunModel in your graph, this is how you layer a LoRA on top of it. It's the standard "patch a model's weights with a low-rank adapter" move, just scoped to this pack's FunModels type instead of ComfyUI's native MODEL type - which is exactly why you can't use ComfyUI's built-in LoRA loader here. VideoX-Fun's models don't look like normal ComfyUI models internally, so the pack ships its own LoRA nodes to match.

How it works

You feed it a cogvideoxfun_model, point it at a LoRA file, set a strength, and it returns the same model type with the LoRA's weight delta merged in at that strength. Chain several of these in sequence if you're stacking multiple LoRAs - each one just takes the previous node's output as its own model input.

The inputs that matter

• cogvideoxfun_model (required) - the model coming out of LoadCogVideoXFunModel.
• lora_name - a dropdown populated from whatever's in your LoRA folder. If this list is empty, ComfyUI isn't finding any LoRA files, not a VideoX-Fun-specific problem.
• strength_model - default 1, range -100 to 100. In practice you'll live in the 0.5-1.2 range; anything far outside that is either barely noticeable or badly overcooked. Negative values invert the LoRA's effect, which is occasionally useful but rarely what you want by accident.
• lora_cache - default false. When enabled, the node keeps the merged weights cached rather than re-merging every run - useful if you're iterating on prompts/settings without changing the LoRA or its strength, since it skips redundant work on repeated executions.

Output: cogvideoxfun_model, same type, now LoRA-patched - feed it into a CogVideoXFun*Sampler node.

Installing it

This node ships as part of VideoX-Fun, so there's nothing to install separately from the base pack. Search ComfyUI Manager for VideoX-Fun; if it's not showing up there, install it directly:

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

Restart ComfyUI and it'll show up alongside the loaders and samplers under the CogVideoXFUNWrapper category.

Common issues

Your LoRA isn't in the dropdown. VideoX-Fun's nodes read from ComfyUI's standard loras/ folder - make sure the file actually landed there and not in a custom_nodes subdirectory, and restart ComfyUI after adding new files (the dropdown doesn't hot-reload).

Wrong architecture LoRA. A LoRA trained for Wan or for a different CogVideoX generation won't apply meaningfully here even if ComfyUI lets you select the file - the tensor shapes it's patching are architecture-specific. If a LoRA does nothing or errors on load, check it was actually trained against the CogVideoX-Fun family you're running.

Effect too strong or too weak. Don't fight it by cranking strength to the extremes - CogVideoX-Fun LoRAs, like most video LoRAs, tend to degrade motion coherence at high strength before they finish "arriving." If 1.0 is too much, try 0.6-0.8 before you assume the LoRA itself is bad.

You're not sure whether this is worth the effort. Honest take: CogVideoX-Fun has a much smaller LoRA ecosystem than Wan does, since most of the community's LoRA-training energy has gone into Wan since it took over. If you're picking a model fresh and LoRA support matters to you, that's one more point in favor of the Wan-Fun loaders instead.