公开信息显示，可灵大模型为快手AI团队自研，采用类Sora的技术路线并结合多项自研创新技术，具备诸多优势：1、能够生成大幅度的合理运动;2、能够模拟物理世界特性;3、具备强大的概念组合能力和想象力;4、生成的视频分辨率高达1080p，最长可达2分钟(帧率30fps)，且支持自由的宽高比。As of 9 am on June 13th, more than 64,000 people were queued up to apply for testing of the text-to-video large model "Keling" from Kuaishou (01024), which began its invitation-only beta test on June 6th. On average, tens of thousands of new applications are submitted each day.
According to the Wise News App, as of 9 am on June 13th, more than 64,000 people were queued up to apply for testing of the text-to-video large model "Keling" from Kuaishou (01024), which began its invitation-only beta test on June 6th. On average, tens of thousands of new applications are submitted each day.
Morgan Stanley recently published a research report stating that Keling's performance is comparable to that of Sora. Its successful release is expected to further increase Kuaishou's user base and activity, thereby having a long-term impact on Kuaishou's monetization potential in e-commerce and commercialization. According to Morgan Stanley, Keling's performance is superior to video generation models previously released by Douyin and Tencent. In terms of duration, Keling can generate videos up to 2 minutes long, while Douyin's Dreamina only supports videos up to a maximum of 3 seconds long and Tencent's Huan Yuan model can generate videos up to 16 seconds long.
Public information shows that Keling is a large model developed by the Kuaishou AI team, using a technology route similar to Sora and combining multiple self-developed innovative technologies. It has many advantages: 1. It can generate reasonable movements on a large scale; 2. It can simulate the physical properties of the real world; 3. It has strong ability to combine concepts and imagination; 4. The generated video has a resolution of up to 1080p, a maximum length of 2 minutes (30fps frame rate), and supports free aspect ratios.
"Since the beginning of 2023, Kuaishou has been steadily promoting the R&D and training of self-developed large models, accelerating the scenario-based landing of large models, including the research and investment in video generation technology. At the end of last year, Kuaishou also actively promoted the special research and development of video generation technology," said Cheng Yixiao, founder and CEO of Kuaishou Technology, at the 2023 Q4 performance conference call held earlier. "In the future, we will combine the generation models and producer tools to continuously help creators reduce the threshold for creation, improve the quality and efficiency of short video production."
"In the future, we will combine generation models with producer tools to continuously help creators reduce the barrier to creation and improve the quality and efficiency of short video production," said Cheng Yixiao.
According to Kuaishou, the currently open feature of Keling is the text-to-video function, which will soon be followed by the image-to-video function. Compared to the text-to-video function, the image-to-video function can reduce trial and error time, and at the same time, has stronger diversity of style and control over details.
