Kling AI is a generative artificial intelligence service created and hosted by Kuaishou, a Beijing-based technology company. The service generates videos from natural language descriptions, known as prompts, using a diffusion-based transformer architecture. Since its initial release in June 2024, Kling AI has undergone several major updates, expanding its capabilities and user base.
Kling AI operates as a cloud-hosted platform, allowing users to input text prompts and receive short video clips generated by the model. It is integrated into Kuaishou's video editing application, KuaiYing, for public testing, and has attracted attention for its ability to produce realistic motion and scene changes.
History
Kling AI was developed by Kuaishou, a Chinese technology company founded in 2011 that operates a popular short video platform. Kuaishou launched the first version of Kling AI in June 2024, making it available for public testing within its video editing app, KuaiYing. This initial release focused on demonstrating the model's core video generation capabilities.
In December 2024, Kuaishou released Kling 1.6, which featured improved video generation capabilities, including better handling of complex motion and higher visual fidelity. This was followed by Kling 2.0 in April 2025, which introduced significant enhancements to the underlying model. Kling 2.1 followed in May 2025, adding different quality modes for users to balance generation speed and output detail.
On June 5, 2025, celebrating its first anniversary, Kling launched a referral program, encouraging existing users to invite new users to the platform. This marked a shift toward broader user acquisition and community building.
Technology
Kling AI uses a diffusion-based transformer architecture (DiT), which has been enhanced by Kuaishou with a self-developed 3D variational autoencoder (VAE) network. This 3D VAE network allows for synchronous spatiotemporal compression, which helps to improve the quality of the generated videos while maintaining training efficiency. The compression reduces the computational load during both training and inference, enabling the model to handle longer or higher-resolution clips.
The model also features a computationally efficient, full-attention mechanism that serves as a spatiotemporal modeling module. This allows Kling AI to accurately capture complex motion and details in videos, including fast-moving objects and drastic scene changes. The architecture builds on advances in Transformer (architecture) models and Deep learning techniques, particularly those used in large language models and neural networks.
Kling AI's development reflects broader trends in Artificial intelligence research, where diffusion models and transformers are increasingly combined for multimodal generation tasks. The use of a 3D VAE distinguishes it from many other video generation systems that rely on 2D or frame-by-frame processing.
Reception
Kling AI's content moderation policies have faced criticism. The model operates under strict censorship rules that align with Chinese government regulations, preventing the generation of content related to sensitive topics such as politics, protests, or criticism of the government. This has limited its appeal among users seeking unrestricted creative expression, though it aligns with local legal requirements.
Furthermore, the popularity of Kling AI has attracted malicious actors. In May 2025, a malware campaign was discovered that used fake Kling AI websites and advertisements to distribute infostealer malware to unsuspecting users. These fake sites mimicked the official platform to trick users into downloading malicious software, highlighting security risks associated with the service's rapid growth.
Despite these issues, Kling AI has been noted for its technical achievements in video generation, particularly its ability to handle dynamic scenes and maintain temporal consistency. Its release has contributed to the competitive landscape of generative AI tools, alongside offerings from other companies in the field.
Impact and Future
Kling AI's iterative release schedule, with updates in December 2024, April 2025, and May 2025, demonstrates a rapid development cycle typical of the generative AI industry. The introduction of quality modes in Kling 2.1 suggests a focus on user customization and accessibility, potentially broadening its adoption beyond professional creators.
The service's integration with Kuaishou's existing ecosystem, including the KuaiYing app, provides a distribution advantage in the Chinese market. As of mid-2025, Kling AI continues to evolve, with ongoing improvements likely to address both technical performance and user concerns around moderation and security.