Run open-weight AI video models on RunPod
Rent a GPU on RunPod, set up ComfyUI, and generate video with open-weight models you run yourself.
Open-weight video models can run on a GPU you rent on RunPod, so you choose the model, the settings and what happens to the output. It takes more setup than a hosted tool, and this guide covers it.
What you need
- A RunPod account with some credit
- A Hugging Face account, for downloading model weights
- About 30 minutes for the first setup
Set up the pod
Create a network volume
Model weights are large. A network volume keeps them between sessions, so you download them once instead of every time you start a pod.
Pick a GPU
Video models need a lot of VRAM. Check the model's page for its minimum, and pick a GPU that meets it with some headroom.
Start from a ComfyUI template
RunPod has community templates with ComfyUI preinstalled. Attach your network volume when you deploy it.
Open ComfyUI
Once the pod is running, connect to the ComfyUI port from the pod's Connect menu.
Download a model
Open-weight video models such as Wan and LTX-Video are published on Hugging Face. From the pod's terminal, download the weights into ComfyUI's model folders on your network volume, then load the matching example workflow in ComfyUI.
pip install -U "huggingface_hub[cli]"
huggingface-cli login
huggingface-cli download <model-repo> --local-dir /workspace/ComfyUI/models/<folder>
Stop or terminate the pod when you're done. You're billed for every minute a GPU pod runs, even when it's idle.
Generate your first clip
Start with a short, low-resolution clip to check the workflow runs, then raise the resolution and length. Each model's page lists the settings it was trained for, and staying close to them gives the best results.