Another day, another AI model. Alibaba Cloud, a subsidiary of Chinese conglomerate Alibaba Group and one of the world’s largest cloud computing companies, has unveiled the I2VGen-XL AI tool. This is an advanced text-to-video system intended to compete with high-end models such as those released by Pika Labs or Stability AI.
The company announced that it announced the model’s weights today, following the publication of a research paper on the model last month.
I2VGen-XL is designed using a cascade diffusion model, a sophisticated AI technique that ensures that the generated videos are not only visually impressive, but also contextually consistent and semantically accurate, the paper explains. This works as a two-step process. The base phase focuses on maintaining consistency with input text and images, while the enhancement phase improves the detail and resolution of the video, achieving up to 1280×720 pixels.
This technique may sound similar to the technique used to create images with SDXL. Unlike SD 1.5 and SD 2.1, which relied on a single model, Stability AI developed two different models: Base and Refiner, which must be combined to produce the highest quality images possible.
Alibaba Cloud says the model’s training leveraged an extensive dataset of approximately 35 million text-video pairs and a whopping 6 billion text-image pairs. These massive data sets ensure model versatility and accuracy across a variety of scenarios and topics.
A new model in the AI arms race
The launch comes amid heightened tensions and competition in the global technology environment, particularly between the United States and China. Alibaba’s move is timely and strategically important for China amid trade restrictions and a push for technological independence.
Alibaba’s latest innovation is not an isolated development but part of a longer story about technological competition. As the United States restricted chip exports and China responded, the competition for AI hegemony accelerated. This environment has fostered the development of indigenous technologies, with both countries competing for leadership positions in AI, semiconductor technology, and 5G innovation.
Compared to other notable advances in the field, such as Pika Labs’ model and Stable Video Diffusion, I2VGen-XL stands out through its unique approach and high semantic accuracy. A demo with several examples of using HiGen (a diffusion model) with I2VGen-XL shows significant improvements in temporal and frame consistency compared to using HiGen alone.
Alibaba’s I2VGen-XL model represents an important milestone in the AI landscape because it provides Chinese users with an alternative to models that have been banned or may be restricted by the U.S. or Chinese governments in the future.
Alibaba’s Emerging Technology Play
Alibaba is more than just e-commerce. It has been a prominent player in emerging technologies for some time, continually driving new developments in the areas of AI, metaverse, software, and even digital currencies.
In AI-based animation, in addition to sI2VGen-XL, Alibaba’s ‘Animate Anyone’ model stands out. This tool uses a new framework called ReferenceNet to convert static images into dynamic animations. By incorporating a sophisticated diffusion model, temporally stable and visually consistent videos can be obtained.
Alibaba Cloud also partnered with Avalanche to launch the Cloudverse platform. This technology provides a seamless path for businesses to create and maintain their digital world. The strategic partnership with Avalanche and Metaverse Universal Assets DAO’s participation in middleware solutions highlights Alibaba’s collaborative approach and commitment to leveraging Web3 technology.
Moreover, Jack Ma’s insights into digital currencies demonstrate Alibaba’s deep interest in the future of global finance. Ma’s advocacy for the transformative role of digital currencies in building a new financial system is consistent with the global trend toward digitalization of the financial sector. Alibaba’s CEO has described himself as a cryptocurrency skeptic, but this stance is far from being a cryptocurrency hater, as Alibaba launched its Blockchain as a Service business in the middle of 2018’s infamous cryptocurrency winter.
Edited by Ryan Ozawa.
