Chinese tech giant Alibaba has unveiled a new AI video tool
11 months ago Benito Santiago
Another day, another artificial intelligence model. Alibaba Cloud—a subsidiary of Chinese conglomerate Alibaba Group and one of the world's largest cloud computing companies—unveiled its I2VGen-XL AI tool. It is an advanced text-to-video system intended to compete with top-of-the-line models released by Pika Labs or Stability AI.
The company has revealed the model's weight today after publishing the model's research paper last month.
I2VGen-XL was designed using fragmented distribution models, the paper explains, and advanced AI techniques ensure that the generated videos are not only visually stunning, but also contextually consistent and semantically accurate. It works in a two-stage process: the base stage focuses on maintaining coherence with the input text and images, while the refinement stage enhances the details and resolution of the video up to 1280×720 pixels.
This method may look similar to those used to create images with SDXL. Unlike SD 1.5 and SD 2.1, which rely on a single model, Stability AI has developed two separate models, a base and a filter, which combine to generate the highest quality images possible.
Alibaba Cloud says the model's training used a vast data set containing about 35 million text-to-video pairs and a staggering 6 billion text-to-image pairs. Such a wide range of data ensures the model's versatility and accuracy in a wide range of situations and cases.
Table of Contents
ToggleA new model in the AI arms race
The release comes as the global technology landscape is witnessing heightened tension and competition, particularly between the US and China. Amid trade restrictions and the push for technological self-sufficiency, Alibaba's move has both timely and strategic significance for the country.
Alibaba's latest innovation is not an isolated development, but part of a long narrative of technological competition. With the US imposing restrictions on chip exports and China imposing countermeasures, the race for AI dominance has accelerated. This environment has spurred advances in indigenous technologies, with both countries vying for a leading position in AI, semiconductor technology and 5G innovation.
Compared to other advances in the field, such as Pika Labs's model and Stable Video Diffusion, I2VGen-XL is distinguished by its unique approach and high localization accuracy. A demonstration with several examples of using HiGen (diffusion model) with I2VGen-XL shows a significant improvement in temporal and frame consistency compared to using HiGen.
Alibaba's I2VGen-XL model represents a significant milestone in the AI landscape because it offers Chinese users an alternative to models that are banned or may be restricted in the future by the US or Chinese government.
Alibaba's new technology games
Alibaba goes beyond e-commerce. It has been a significant player in emerging technologies for a while, consistently pushing new developments in AI, the metaverse, software and even digital currencies.
In addition to AI-driven animation, sI2VGen-XL, Alibaba's “Animate Anyone” model stands out. This tool converts static images into dynamic animations, using a novel framework called ReferenceNet. Integrating sophisticated distribution models achieves temporally stable and visually consistent videos.
Alibaba Cloud has partnered with Avalanche to launch the Cloudverse platform. This technology gives businesses a seamless way to create and maintain a digital universe. The strategic alliance with Avalanche and Metaverse Universal Assets, DAO's involvement in middleware solutions, underscores Alibaba's collaborative approach and commitment to leveraging Web3 technologies.
Moreover, Jack Ma's insight into digital currencies shows Alibaba's keen interest in the future of global finance. Mau's support for the transformative role of digital currencies in establishing a new financial system coincides with a growing global trend in finance. Alibaba's CEO has described himself as a crypto skeptic, but such a stance is far from crypto-hating, in the middle of the crypto summer of 2018, Alibaba launched Blockchain as a Service.
Edited by Ryan Ozawa.