Alibaba Cloud Open Sources Video Foundation Models

Alibaba Cloud has gave his family of video foundation models Open-Sourcut models and researchers access to the functions to create video performances.

The four models in the Tongyi Wanxiang (WAN) 2.1 family, which comprises both 14 billion and 1.3 billion parameter versions, are designed in such a way that they create high-quality videos from text and image input. The models can be downloaded into Alibaba Clouds KI model community, model scope and in the hug of the face.

According to Alibaba Cloud, WAN 2.1 is the first model of videoogenization that supports text effects both in Chinese and in English. His skills in creating realistic visuals are due to his ability to manage complex movements, to improve the pixel quality, to keep the physical principles and to optimize the accuracy of the instructions.

These functions have written WAN 2.1 at the top of the VBECH ranking, a benchmark suite for video models. It is also the only open source model among the top 5 VBen ranking from Face.

The different models are aimed at different requirements and arithmetic resources. The 14B model is characterized in order to create high-quality visuals with complex movement dynamics, while the 1.3b model, which compensates for the quality of the generation with arithmetic efficiency, enables users with standard laptops to generate a 480p video with five seconds in about four minutes.

Training video foundation models requires a large amount of arithmetic resources and high-quality training data. Open Source can reduce the barrier for more companies to use AI so that you can create high -quality visual content that is tailored to your requirements in a cost -effective way.

In addition to WAN 2.1, Alibaba Cloud also gave its QWen Foundation models Open-Sourcut models on the Huggingface Open LLM components with performance with a leading global models. There are currently over 100,000 derivative models based on the QWen family on the hugs to one of the largest AI model families worldwide.

The company also provides the AI ​​model studio with which large companies can access its basic models and model training tools in order to accelerate the provision of large -scaling models in a controlled environment.

With Model Studio you can monitor and identify risky content and filter or block unwanted information based on responsible AI principles. You can also train Foundation models by creating, labeling and managing training data records, adapting model training with adjustable parameters and evaluating and providing foundation models without any problems.

At the beginning of this week, Alibaba Cloud announced that in the next three years she will invest 380 billion RMB (53 billion USD) in its cloud computing and artificial intelligence infrastructure (AI) and exceeded the total editions for cloud and AI in the past decade.

The Cloud Intelligence Unit from Alibaba reported sales growth of 11% compared to the year in the last quarter, except for consolidated subsidiaries. Sales with AI-related product income rose in the three-digit range in the sixth quarter in a row, thanks to the increasing demand for AI hosting and the associated offers.