The Chinese Internet giant, Tencenthas unveiled the new version of its model artificial intelligence (AI) for video generation open source, DynamiCrafter. An announcement that comes little Google’s for Lumiere, which performs the same function. And it underlines the fact that the technological competition between the United States and China is very lively, also with regards to AI.
Tencent announces AI to create videos
Tencent announced the news on GitHub on Monday. This move confirms Chinese tech companies’ growing focus in the generative AI space, focusing on creating videos from images and text — a true technological breakthrough.
DynamiCrafter, like other generative video tools, makes use of the method of diffusiontransforming captions and static images into videos of lasting a few seconds. Diffusion models allow you to transform simple data into more complex and realistic data, similar to the physical motion of particles moving from an area of high concentration to an area of low concentration. ChatGPT and Google Bard also work this way, although the algorithms that enable this effect are well-kept secrets.
Video from images or a few words
The new version of DynamiCrafter generates videos with a pixel resolution of 640×1024, representing a significant upgrade over the previous version which produced 320×512 video. An academic article published by the team highlights the diversity of this technology compared to other similar ones. The key innovation is to incorporate the image into the generative process as a guide. This contrasts with traditional techniques that focus primarily on animating natural scenes with stochastic dynamics or domain-specific movements.
In a comparative demo between DynamiCrafter, Stable Video Diffusion (launched in November) e Pika Labs (the video above), the Tencent model appears slightly more animated. But as TechCrunch points out, one can only assume that Tencent has only chosen examples that favor its model.
Growing competition in AI
AI has demonstrated the ability to generate texts and images quite successfully. ChatGPT is becoming more and more normal in the world of work, as many people use AI to create images. Now, several companies are investing in generative videos. And it seems like that OpenAI and Google will not be alone: China is also working on this front.
To prove this, startups and technology companies in China, including ByteDance, Baidu and Alibabawho have released their own video diffusion models. MagicVideo the ByteDance e UniVG of Baidu have published demos on GitHub, while Alibaba has open sourced its VGen model. In short, the competition on AI between the United States and China seems more alive than ever. Like a video generated by artificial intelligence.
