hunyuan gguf

Introduction

The Hunyuan-GGUF model is a text-to-video framework designed for generating high-quality video content from textual descriptions. This model leverages the GGUF quantized version of the Hunyuan-Video model and is provided by the user calcuis. It supports English language inputs and is part of the Comfy-Org's repackaged models.

Architecture

The model is based on Comfy-Org/HunyuanVideo_repackaged and utilizes a quantized version for efficient performance. It involves several components, including the GGUF quantized model and various supporting files such as text encoders and VAE models. The architecture is designed to facilitate text-to-video generation using advanced neural network techniques.

Training

Details on the specific training parameters or datasets used for Hunyuan-GGUF are not provided in the documentation. However, it builds upon Tencent's HunyuanVideo base model and incorporates quantization techniques to optimize video generation.

Guide: Running Locally

Setup

1. Download Required Files:
   • Download hunyuan-video-t2v-720p-q4_0.gguf (7.74GB) and place it in ./ComfyUI/models/unet.
   • Download clip_l.safetensors (246MB) and place it in ./ComfyUI/models/text_encoders.
   • Download llava_llama3_fp8_scaled.safetensors (9.09GB) and place it in ./ComfyUI/models/text_encoders.
   • Download hunyuan_video_vae_bf16.safetensors (493MB) and place it in ./ComfyUI/models/vae.

Running the Model

1. Execute the .bat File:

   • Run the batch (.bat) file located in the main directory to start the application without additional installation.

2. Load Workflow:

   • Drag the provided workflow JSON file into your web browser to load the model's processing workflow.

Recommended Cloud GPUs

For optimal performance, it is recommended to use cloud GPU services such as NVIDIA's Tesla V100 or A100, available through platforms like AWS, Google Cloud, or Microsoft Azure.

License

The Hunyuan-GGUF model is distributed under the tencent-hunyuan-community license. Further details can be found in the associated LICENSE file.