Hunyuan Video in ComfyUI

NOTE: This is WIP, while people work tirelessly in this community to bring things to everyone as fast as humanly possible, please understand that "Native ComfyUI support" comes later and takes time.

We will look at setting up Kijai's Wrapper for Hunyuan Video from Tencent.

Hoda Hunyuan Video Pack ComfyUI Workflows:
https://civitai.com/models/1013919/hoda-hunyuan-video-pack
https://github.com/MushroomFleet/DJZ-Workflows/tree/main/Hoda-Hunyuan-Video

this video was filmed without Sage Attention installed on day one

You will need:

Kijai's Wrapper (custom nodes) (update this often!)
https://github.com/kijai/ComfyUI-HunyuanVideoWrapper

place inside (or git clone): ComfyUI/custom_nodes/

If you use ComfyUI portable, you might need my "install-portable.bat" for windows as (its so new) this is not in the manager yet. You can run this from the inside the custom node folder.

Transformer Weights:
https://huggingface.co/Kijai/HunyuanVideo_comfy/blob/main/hunyuan_video_720_cfgdistill_fp8_e4m3fn.safetensors
https://huggingface.co/Kijai/HunyuanVideo_comfy/blob/main/hunyuan_video_720_cfgdistill_bf16.safetensors (25.6GB)
or
https://huggingface.co/Kijai/HunyuanVideo_comfy/blob/main/hunyuan_video_720_cfgdistill_fp8_e4m3fn.safetensors (13.2GB)

place inside: ComfyUI/models/diffusion_models/

VAE:
https://huggingface.co/Kijai/HunyuanVideo_comfy/blob/main/hunyuan_video_vae_bf16.safetensors (493mb)
or

https://huggingface.co/Kijai/HunyuanVideo_comfy/blob/main/hunyuan_video_vae_fp32.safetensors (986mb)

place inside: ComfyUI/models/vae/

LLM Text Encoder (has autodownload)
https://huggingface.co/Kijai/llava-llama-3-8b-text-encoder-tokenizer (~10GB)

place inside: ComfyUI/models/LLM/llava-llama-3-8b-text-encoder-tokenizer/

CLIP text encoder

https://huggingface.co/openai/clip-vit-large-patch14

place inside: ComfyUI/models/clip/clip-vit-large-patch14/

Troubleshooting Incomplete Downloads

If you have problems with interrupted downloads when grabbing the last two Hugging face repos, or if your ComfyUI auto-download is failing to retrieve the complete files, consider using my HF Downloader webUI, and copy them into the correct locations. I have to do this a lot, so you might find that easier.

I'm sure that ComfyOrg is working hard to bring us support for this new Video Generation model, so watch this space for more information !