In This Article:
BEIJING, June 10, 2024 /PRNewswire/ -- Kuaishou Technology (HKD Counter: 01024 / RMB Counter: 81024)(together with its subsidiaries and consolidated affiliated entities, hereinafter referred to as "Kuaishou"), a leading content community and social platform, recently launched its self-developed video generation model, the Kling Large Model (hereinafter referred to as "Kling" (可灵)). Kling is capable of generating complex spatiotemporal motions and simulating the characteristics of the physical world. Leveraging these capabilities, Kling transforms text prompts into high-quality AI videos that closely mimic the real world's complex motion patterns and physical characteristics. Kling also possesses powerful conceptual combination and imagination abilities. Kling can generate videos up to two minutes long with a frame rate of 30fps and video resolution up to 1080p while supporting a variety of aspect ratios.
Kling utilizes diffusion-based transformer architecture (DiT), enhanced with Kuaishou's upgrades to the model's latent space encoding/decoding and temporal modeling modules. In terms of latent space encoding/decoding, Kuaishou has self-developed a 3D VAE network, achieving synchronous spatiotemporal compression and obtaining a high reconstruction quality, striking an excellent balance between training performance and effectiveness. In terms of temporal modeling, Kuaishou designed a computationally efficient, full-attention mechanism as a spatiotemporal modeling module. This method integrates temporal and spatial information, enabling comprehensive analysis and processing of video data. It can accurately capture local spatial features within video frames and temporal dynamic features across frames for a more comprehensive understanding and reproduction of motion information in videos. As a result, Kling can accurately capture details from rapidly moving objects, drastic scene changes, complex human movements and more, empowering dynamic, highly realistic video content generation.
Kling is currently available for beta testing within "KuaiYing" (快影), Kuaishou's video editing application for users in China. Users may register and apply for a Kling trial through KuaiYing. For more information and a video demo, please visit Kling's official website at http://kling.kuaishou.com/
As a global leader in the short video industry, Kuaishou has developed a comprehensive AI strategy to usher in the large AI model era. Large AI models offer a rich array of application scenarios for Kuaishou, seamlessly integrating with Kuaishou's content and commercial ecosystems. Kuaishou has already released "KwaiYii" (快意), a general large language model with 175 billion parameters, and "KeTu" (可图), a large model product for text-to-image generation, both of which have attracted widespread attention. Kling's launch demonstrates Kuaishou's commitment to accelerating the research, development and application of large models, aiming to provide creators and users with more diverse AI-powered creation and interactive experiences.