Running Hunyuan Video Model Locally

Learn how to set up and run the powerful Hunyuan video model on your local machine for video generation and processing using ComfyUI.

Prerequisites

• GPU: VRAM of 12GB or 10GB. For convenience, here is a list of recent GPU with at least 10GB VRAM: Nvidia Desktop GPU: RTX 3060, 3080, 3080TI, 3090, 3090Ti, most of the 4000 series RTX GPU. Nvidia Mobile GPU: 3080, 3080Ti (both have 16GB VRAM). AMD GPU: 6700XT and up (6700–6950XT). 7600–7900XT

• RAM: 30GB of memory is need to load the model files.

• Disk(SSD): need at least 100GB for model storage.

Installation Steps...

Ensure you have the latest ComfyUI (either fresh install or upgrade). I made a video guide on install it.

Download the hunyuan_video_t2v_720p_bf16.safetensors file and put it in your ComfyUI/models/diffusion_models folder. Download the clip_l.safetensors and llava_llama3_fp8_scaled.safetensors files from here and put them in your ComfyUI/models/text_encoders directory. Download the hunyuan_video_vae_bf16.safetensors file and put it in your ComfyUI/models/vae folder.

Then download and load the workflow to ComfyUI .

My run results:

I tested on my 3080Ti GPU (12GB VRAM) on a 32GB RAM machine (Ubuntu OS). Below is the performance (848x480 73 frames, 20 steps):

6 Minutes is needed for 20 steps.