HunyuanVideo (neural network)

History

2024: Product Launch

In early December 2024, the Chinese corporation Tencent Holdings announced the launch of the HunyuanVideo neural network. It is said to be the world's largest open AI model for video generation.

HunyuanVideo creates video based on text prompts. According to Tencent, the neural network is designed to generate high-quality "hyper-realistic" videos with the ability to change the camera angle. For example, you can ask an AI model to show a cat in a cyberpunk landscape, a surfer on a wave, or several balloons slowly climbing into the sky.

HunyuanVideo has more than 13 billion parameters. At the time of launch, the neural network perceives text tasks in English or Chinese. Various video style, quality, and size options are available. In particular, HunyuanVideo is capable of generating 720p content for up to five seconds. Users are offered advanced filmmaking settings.

The new neural network is available free of charge for both individual users and companies. Since the launch of Huanyuan's basic AI model in September 2023, Tencent has been gradually expanding the platform's capabilities. In particular, in March 2024, the company, in collaboration with the Hong Kong University of Science and Technology and Tsinghua University, introduced the Follow-Your-Click image-video conversion tool. And later it became possible to convert text into images. The HunyuanVideo model is based on the Diffusion Transformer architecture, which underlies Sora, an AI video generator introduced by OpenAI in early 2024. This system, it is claimed, can create realistic scenes according to text instructions.^[1]

Notes