CogVideo LoraSelect Comfy

This node picks a CogVideoX LoRA and packages it for the model loader. The "Comfy" in the name is the important detail: it reads from your standard ComfyUI/models/loras folder - the same place all your other LoRAs live - rather than a CogVideoX-specific subfolder. So if you dropped a CogVideoX LoRA in with the rest of your LoRAs, this is the node that finds it.

A LoRA is a small add-on that shifts the model toward a particular subject, style, or motion without retraining the whole thing. For CogVideoX that might be a character LoRA, a motion LoRA, or a speed/distillation LoRA. This node doesn't apply the LoRA itself - it selects and bundles it, then you feed the bundle into the lora input on your model loader, which is where it actually gets baked in.

The inputs that matter

• lora - the dropdown of files found in ComfyUI/models/loras. If it's empty, there's nothing in that folder (or you need to refresh). This is where the "Comfy" variant differs from its sibling CogVideoLoraSelect, which reads from ComfyUI/models/CogVideo/loras instead. Same node otherwise; different folder.
• strength (default 1) - how strongly the LoRA is applied. 1 is full strength. Turn it down (0.5–0.8) if the LoRA is overpowering the prompt or degrading motion; the range even goes negative if you want to push away from a concept.
• prev_lora (optional) - this is how you stack LoRAs. Chain the lora output of one select node into the prev_lora input of the next, and both ride together. Repeat for as many as you want.
• fuse_lora (optional, default off) - fuses the LoRA weights into the model rather than applying them as a live patch. Can be faster at inference, less flexible to toggle.

The output, lora (a COGLORA), goes into the lora input of (Down)load CogVideo Model or CogVideoX Model Loader - not into the sampler directly.

How to install it

Comes 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

The dropdown doesn't show my LoRA. Two things to check. First, the file has to be in ComfyUI/models/loras for this node - if it's in ComfyUI/models/CogVideo/loras, use CogVideoLoraSelect instead. Second, a LoRA trained for a different model won't work; a CogVideoX LoRA is not interchangeable with a Wan or SDXL one, even if it shows up in the list.

The LoRA does nothing. Make sure the lora output is wired into the loader's lora input, not left dangling or plugged somewhere else. The select node only prepares the LoRA; the loader applies it. Also confirm strength isn't sitting at 0.

Stacked LoRAs fight each other. When you chain several via prev_lora, their effects combine and can overshoot. Drop each one's strength so the total stays balanced rather than running every LoRA at full 1.0.