ComfyUI Hunyuan Latent Nodes

Custom nodes for ComfyUI to generate empty latent space compatible with Hunyuan models for both image and video generation.

For video

For Images

Features

• Fixed aspect ratio selection from common formats (1:1, 16:9, 9:16, etc.)
• Support for both image and video latent generation
• All dimensions are automatically adjusted to be multiples of 16
• Maintains compatibility with Hunyuan model tensor format

Available Nodes

EmptyHunyuanLatentForImage

Creates an empty latent space for image generation with the following parameters:

• resolution: Dropdown selection of common aspect ratios and resolutions
• batch_size: Number of images to generate in parallel (default: 1)

EmptyHunyuanLatentForVideo

Creates an empty latent space for video generation with the following parameters:

• resolution: Same resolution options as image node
• frames: Number of frames to generate (default: 16, range: 1-256)
• batch_size: Number of videos to generate in parallel (default: 1)

Supported Resolutions

The nodes support various standard resolutions with their aspect ratios:

• Portrait formats:
  • 576x1024 (9:16)
  • 704x1408 (1:2)
  • 768x1344 (4:7)
  • And more...
• Square format:
  • 1024x1024 (1:1)
• Landscape formats:
  • 1024x576 (16:9)
  • 1344x704 (16:9)
  • 1408x704 (2:1)
  • And more...

Installation

1. Create a custom_nodes directory in ComfyUI if it doesn't exist

cd ComfyUI/custom_nodes

2. Clone this repository:

git clone https://github.com/ShmuelRonen/ComfyUI-Hunyuan-Latent.git

3. Restart ComfyUI

The nodes will appear in the node menu under:

• latent for EmptyHunyuanLatentForImage
• latent/video for EmptyHunyuanLatentForVideo

Usage

1. Right-click in the ComfyUI workspace
2. Navigate to:
   • latent → EmptyHunyuanLatentForImage for image generation
   • latent/video → EmptyHunyuanLatentForVideo for video generation
3. Select desired resolution from the dropdown
4. Connect to other nodes in your workflow

Node Output Format

Both nodes output a latent dictionary with the key "samples" containing a torch tensor with the following dimensions:

• Image: [batch_size, 4, 1, height//8, width//8]
• Video: [batch_size, 4, frames, height//8, width//8]

Requirements

• ComfyUI
• PyTorch
• Hunyuan model dependencies

Latest Hunyuan models from HuggingFace

https://huggingface.co/Comfy-Org/HunyuanVideo_repackaged/tree/main/split_files/text_encoders

I recomended this hunyuan_video_720_cfgdistill_fp8_e4m3fn model for fast operation:

https://drive.google.com/file/d/1BvGHjR4Mb60ZPx9tqzA1AabAwZc47ctx/view?usp=sharing

License

MIT

Contributing

Contributions are welcome! Please feel free to submit a Pull Request.