CogVideo LoraSelect

This node selects a CogVideoX LoRA and bundles it for the model loader. It's the near-twin of CogVideo LoraSelect Comfy, and the only real difference is where it looks: this one reads from ComfyUI/models/CogVideo/loras, the wrapper's own dedicated LoRA folder, while the "Comfy" variant reads from the standard ComfyUI/models/loras. Same controls, same output, different directory. Pick whichever matches where your LoRA file actually lives.

Quick refresher on what a LoRA does here: it's a lightweight adapter that biases CogVideoX toward a specific character, style, or motion pattern without touching the base weights. This node doesn't apply it - it selects and packages it, and the package plugs into the loader's lora input, which is where the LoRA gets merged into the model for the run.

The inputs that matter

• lora - the dropdown of files in ComfyUI/models/CogVideo/loras. Empty means nothing's in that folder. If your LoRA is in the general loras directory instead, you want CogVideo LoraSelect Comfy.
• strength (default 1) - application strength. 1 is full; dial down to 0.5–0.8 if the LoRA is stomping on your prompt or hurting motion quality. Negative values push away from the concept.
• prev_lora (optional) - the stacking input. Chain one select node's lora output into the next node's prev_lora to run multiple LoRAs together.
• fuse_lora (optional, default off) - fuses the LoRA into the model weights instead of applying a live patch; a little faster at inference, less flexible.

The output, lora (a COGLORA), feeds the lora input on (Down)load CogVideo Model or CogVideoX Model Loader.

How to install it

Bundled with the wrapper:

• ComfyUI Manager - search ComfyUI-CogVideoXWrapper, install, restart.
• Manual - cd ComfyUI/custom_nodes && git clone https://github.com/kijai/ComfyUI-CogVideoXWrapper, then pip install -r ComfyUI-CogVideoXWrapper/requirements.txt, restart.

Common issues & troubleshooting

My LoRA isn't in the dropdown. This node only scans ComfyUI/models/CogVideo/loras. If you put the file in the normal ComfyUI/models/loras folder, switch to CogVideo LoraSelect Comfy - that's the whole reason both nodes exist. Two select nodes, two folders; the confusion between them is the single most common snag.

The LoRA has no effect. Check that the lora output goes into the loader's lora input (not the sampler), and that strength isn't 0. Also confirm the LoRA was actually trained for CogVideoX - a LoRA from another model family won't apply cleanly even if it appears in the list.

Stacked LoRAs overcook the result. When chaining via prev_lora, lower each LoRA's strength so their combined effect stays balanced rather than every one running flat out.