[14:43 Tue,3.December 2024 by Rudi Schmidts] |
With Hunyuan Video, Tencent is presenting a new type of open source video model whose first demo clips demonstrate a quality in video generation that is at least comparable with many closed source models. The model also offers several modern additional functions under one hood:
;) Tencent&s HunyuanVideo AI model is set to beat current competitors According to Tencent, the Hunyuan video model was trained with over 13 billion parameters, making it the largest of all available open source models. However, the memory requirements for the model also exceed the VRAM requirements of standard PC GPUs. At least 45 GB V-RAM is required for clips in the smallest resolution (960 x 544 px). For better results, 60 - 80 GB are strongly recommended. This allows you to create 1280 x 720 px videos without cloud support. However, until you have recouped the approximately 20,000 euros for an NVIDIA A100 80GB, it might be better to rent such a GPU in the cloud. Corresponding offers can be found here from around 1 dollar / hour. This might be worthwhile, because according to Tencent, HunyuanVideo currently outperforms other current AI video models such as Runway Gen-3, Luma 1.6 and three of the most powerful Chinese video generation models. By releasing the code and weights of the base model and its applications, Tencent deliberately aims to bridge the gap between closed-source and open-source video base models. This initiative is intended to give the community the opportunity to experiment with their own ideas and thus promote a more dynamic and vibrant ecosystem for video generation. If you are interested in more background information on the model, you can find it here  the corresponding paper.
 more infos at bei aivideo.hunyuan.tencent.comdeutsche Version dieser Seite: Tencent Hunyuan - Offenes KI-Videomodell soll aktuelle Konkurrenz schlagen |
 |
|
|