Alibaba Group Holding will open source four Sora-style video-generation models, in the latest effort by the Chinese tech giant to promote its artificial intelligence (AI) technologies to developers.
Advertisement
Alibaba Cloud, the cloud computing and AI arm of Alibaba, said on Wednesday that it has made the four models of the Wan2.1 series – the latest version of its video foundational model Tongyi Wanxiang – freely available for academics, researchers and commercial institutions worldwide. Alibaba owns the South China Morning Post.
The four models are trained on 1.3 billion or 14 billion parameters, Alibaba Cloud said, adding that they are available for download on its own AI model community, called Model Scope, as well as the collaborative AI platform Hugging Face.
The open-source approach to offering software is gaining momentum, partly thanks to Hangzhou-based DeepSeek, the Chinese AI start-up that open-sources its popular large language models. The trend, in turn, has initiated a wave of AI applications among Chinese enterprises and consumers.
Open-sourcing provides public access to a software program’s source code, allowing third-party developers to modify or share its design, fix broken links or scale up its capabilities. Use of the approach has gained rapid adoption following the popularity of DeepSeek’s models, which rival the performance of OpenAI’s models at a fraction of the cost.
Advertisement
The Wan2.1 series, which was released in January, was the first video-generation model to support text effects in both Chinese and English. The model now ranks as the top video-generation model on VBench Leaderboard, a benchmark suite for video-generative models on Hugging Face. It is the only open-source model among the top five on the list.