Chinese tech company Kuaishou, also known internationally as Kwai, has expanded the reach of its generative artificial intelligence platform, Kling AI, to a global audience. The beta version of Kling AI is now accessible worldwide, requiring only an email address for registration.
Upon signing up, users receive 66 free credits daily for image and video generation. Kling AI supports both text-to-video and image-plus-text-to-video generation, positioning it as a competitor to OpenAI’s Sora, which is still limited to a select group of users.
The efficiency of Kling AI’s generation capabilities is notable. Each 5-second video clip takes approximately 5 minutes to generate, equating to 1 minute per second of video, or half a second per frame. Image generation is almost instantaneous, taking a matter of seconds per image.
The platform currently offers three video resolution options: a 960×960 square format, or 720p in landscape or portrait mode. It also features camera control, allowing users to select between static, horizontal, vertical, pan, tilt, or rolling movements.
Users can fine-tune their creations by adjusting the balance between creativity and relevance. Favoring relevance ensures the output closely adheres to the prompt, while emphasizing creativity yields more imaginative results. Kling AI also accepts negative prompts to enhance accuracy.
The platform supports multilingual prompts, delivering comparable results to identical prompts written in both English and Mandarin.
Video generated by Kling AI in response to a prompt requesting a “cute kitten eating lunch like a human.
Video generated by Kling AI in response to a prompt written in Chinese, also requesting a cute kitten eating lunch like a human. The result did not significantly differ from the version generated in response to the English prompt.
Based on a brief round of testing by KrASIA, Kling AI demonstrated the ability to generate realistic video content with most prompt components present. However, the majority of videos produced in the sample lacked coherence, highlighting the ongoing challenge of achieving smooth temporal dynamics in video generation—a challenge that affects Kling AI as well.
For instance, in response to a prompt requesting a “realistic puppy driving a car,” Kling AI generated a video snippet showing a dog at the wheel from the driver’s perspective. However, the dog was positioned on the wrong side of the wheel, and there were inconsistencies in the rearview mirror within the video.
Video generated by Kling AI in response to a prompt requesting for a “realistic puppy driving a car.”
Currently, video generation is limited to 5-second clips, but Kuaishou plans to introduce long-form generations in a future update. Kuaishou has segmented Kling AI into two platforms, with a new web portal for international users and a separate domain for the original Chinese version. This strategy mirrors those of other major Chinese tech companies, such as ByteDance with TikTok (known as Douyin in China) and Tencent with WeChat (domestically known as Weixin).