infotech

Alibaba Cloud Opens "Open-Source" Access For Video-generating AI Models

Alibaba Cloud has officially announced that it has opened up access to open-source artificial intelligence (AI) video generator models as part of its latest efforts to support the open-source community. 

In a press release by Alibaba Cloud on Wednesday, the cloud company opened up access to four models from the Wan2.1 series, the latest version of the Tongyi Wanxiang (Wan) video base model, with 14 billion(B) and 1.3 billion(B) parameters. 

The four models, namely T2V-14B, T2V-1.3B, I2V-14B-720P, and I2V-14B-480P, are designed to generate high-quality images and video content from text or image inputs. 

The models can be downloaded through Alibaba Cloud's AI model community, Model Scope, and the Hugging Face AI collaborative platform, and are accessible to academics, researchers, and commercial institutions globally. 

Launched earlier this year, the Wan2.1 series is the first video content generation model to support both Chinese and English text effects. 

The model excels in producing realistic visuals by accurately handling complex motion, enhancing pixel sharpness, following the principles of physics—thus improving the naturalness of motion, and optimizing the precision of instruction execution. 

The high accuracy of instruction execution earned the Wan2.1 the top spot on the VBench leaderboard, a comprehensive set of benchmarks for video content generation models. 

In addition, the Wan2.1 is the only open-source video generation model to rank in the top five of Hugging Face’s VBench leaderboard. 

According to VBench, the Wan2.1 series scored an overall score of 86.22 percent, with strengths in key dimensions such as motion dynamics, spatial relationships, color, and multi-object interactions. 

With this open access, more businesses can leverage AI to create high-quality, customized visual content at a lower cost. 

---

Tag: Alibaba Cloud Artificial Intelligence