HunyuanVideo: Tencent releases the weights of a 13-billion-parameter video model
In one sentence On December 3, 2024 Tencent releases weights and inference code for HunyuanVideo, a text-to-video model with over 13B parameters that generates 129-frame 720p clips. In Tencent's evaluation with professional raters it ranks ahead of Runway Gen-3 alpha and Luma 1.6. The license explicitly excludes the European Union.
At the end of 2024 open video models were still a step behind commercial services. Tencent publishes HunyuanVideo with weights and inference code: over 13 billion parameters, the largest open video model at the time according to the paper.
The model generates 720p (1280x720) videos of 129 frames. That takes datacenter GPUs: about 60 GB of peak memory at 720p, 45 GB at 540p; Tencent recommends an 80 GB GPU.
In Tencent's evaluation, over 60 professional raters compared results on 1,533 prompts: HunyuanVideo comes first, ahead of Runway Gen-3 alpha and Luma 1.6.
An important detail for anyone working in Italy: the Tencent Hunyuan Community License states that it does not apply in the European Union, the United Kingdom and South Korea.
Companies
Tencent
Tools
HunyuanVideo
Tags
Sources