Kling is a Text-to-video generation and image-to-video generation model developed by Kuaishou Technology, the Chinese short-video platform company behind the Kuaishou app, a rival to ByteDance's Douyin (the Chinese version of TikTok). Kuaishou released Kling in June 2024, positioning it as a domestic competitor to Sora, which OpenAI had previewed several months earlier but had not yet released publicly. Kling was made available first to a waitlist within Kuaishou's Kwaiying video-editing app in China before expanding to a broader international web application.
Capabilities
At launch, Kling could generate video clips up to two minutes long at 30 frames per second, a notably longer duration than most competing video-generation systems offered at the time, along with support for video generation at up to 1080p resolution. The model used a diffusion architecture combined with Transformer (architecture)-based components, following an architectural direction similar to that reported for Sora, modeling video as spacetime patches to handle varying length and resolution. Subsequent versions, including Kling 1.5, 1.6, and Kling 2.0 and 2.1 through 2025, improved motion coherence, added image-to-video and start-and-end-frame video generation (specifying both a first and last frame and letting the model generate the transition), and introduced tools for lip-syncing and character consistency across shots aimed at professional and semi-professional video creators.
Competitive landscape
Kling emerged as part of a broader wave of Chinese video-generation models released through 2024 and 2025, alongside ByteDance's Seedance and other entrants, that collectively pushed the pace of progress in text-to-video generation faster than many Western observers anticipated. On community-run comparison platforms and leaderboards such as Arena-style video rankings, Kling was repeatedly rated among the top-performing video models globally through 2025, competing closely with Veo from Google DeepMind, Sora from OpenAI, and Seedance, with different models trading the top spot as each released updates. Kling's rapid iteration cycle and willingness to expose newer capabilities, such as longer duration and finer motion control, to a broad international user base through its web app contributed to its adoption outside China, including by marketing and content-creation professionals.
Reception
Industry commentary framed Kling, alongside Seedance and other Chinese video models, as evidence that China's AI industry had become internationally competitive not just in language models but in resource-intensive generative media, an area assumed to favor labs with the largest compute budgets. Kling drew both interest and scrutiny typical of the broader text-to-video field, including questions about training data provenance and the use of generated video for deepfake-adjacent misuse, concerns Kuaishou addressed in part through visible watermarking of generated output and usage policies restricting certain content categories. As with other fast-moving generative video systems, Kling's precise training data and full technical specifications were not fully disclosed publicly, consistent with a broader industry pattern of limited transparency in video-generation model releases.