China Launches Self-Developed Global Video AI Model

On August 3, it was announced that Vidu, an AI video model, has officially launched globally at the 2024 Zhongguancun Forum. Vidu features two core functionalities: text-to-video and image-to-video, offering options for 4-second and 8-second clips with a maximum resolution of 1080P. Developed by Shengshu Technology in collaboration with Tsinghua University, Vidu is designed to deliver long-duration videos with high consistency and dynamism.

This global rollout of Vidu introduces additional capabilities, including anime-style creation and character consistency. A representative from Shengshu Technology highlighted that Vidu achieves the industry's fastest real-time inference speed, generating a 4-second clip in just 30 seconds. Users can access Vidu effortlessly by registering with their email, eliminating the need for an application process.

Earlier this year, the text-to-video model Sora garnered significant attention worldwide. The evaluation of video models primarily focuses on three key areas: semantic understanding accuracy, visual appeal, and dynamic consistency of the subject. Vidu adeptly balances these dimensions, accurately interpreting and generating text prompts, including letters and numbers, while also creating text effects. For various camera techniques such as first-person perspectives and time-lapse photography, Vidu offers precise expression, enabling users to enhance video control by refining their prompts.

Moreover, Vidu supports extensive and accurate motion generation, ensuring smooth and dynamic visuals. It excels in composition, storytelling, and lighting effects, approaching a cinematic quality. Additionally, Vidu can produce film-level visual effects, including smoke, glare effects, and computer-generated graphics.