Forget OpenAI's Sora: Kling AI is producing mesmerizing videos.
Keeping up with developments in the field of artificial intelligence, let alone following them, is quite challenging. While OpenAI’s Sora is quite impressive, now the Chinese company Kling AI is raising the bar in video production.
The growth in the field of artificial intelligence continues at an incredible pace. Just a few months after the introduction of OpenAI’s text-to-video AI Sora in February, a serious competitor has unexpectedly emerged from China. The Chinese video platform company Kuaishou has unveiled its generative video model Kling, which appears impressive with its advanced capabilities.
Kling has overshadowed Sora
While OpenAI showcased Sora earlier this year, it has not yet been released for general use. On the other hand, Kuaishou has already launched Kling. Thus, Kling goes down in history as one of the first text-to-video generative AI models available for public testing for free.
The short video platform, which boasts over 600 million active users, announced its new tool earlier this month. Kling can produce videos up to two minutes long with a frame rate of 30 FPS and a resolution of up to 1080p. As known, Sora was capable of producing 1-minute clips. Looking at the examples of Kling published by Kuaishou and users, we see that the generative video model adheres to prompts and performs quite well in terms of physics. Kling can also transform a photo input into a video.
At its core, Kling is based on the same type of diffusion transformer model as Sora and supports various aspect ratios and different shot types. It is worth mentioning that the model used in the AI is proprietary to the company. According to the company’s website, Kling has advanced 3D face and body configuration capabilities to improve limb movement, contributing to its physical accuracy.
Upon closer examination of Kling’s examples, it is evident that it can produce impressive photorealistic scenes, although similar blurriness seen in other AI-generated videos is also present. In some cases, such as the parrot video, even trained eyes might struggle to identify that the video was AI-generated, showcasing the model’s power.
Another significant strength of Kling is its handling of liquids. Sora, Runway, and others often struggle with liquids, but Kling can consistently pour milk into a cup of coffee. Kling and similar AIs operate similarly: they combine diffusion models traditionally used in video generation AI with a transformer architecture that helps them understand larger video data sets and produce more efficient results.
Kling’s greatest strength lies in its mastery of these data sets. The company operates one of China’s largest short video platforms, with users uploading millions of videos. Despite this, Kling currently produces only a few seconds of output in its public version. It’s worth noting that Sora also creates short videos in many examples. This approach helps reduce the issue of hallucinations. The longer the video, the more possibilities the model has to predict, increasing the likelihood of errors.
Increasing Competition
Currently, text-to-video AIs available on the market can typically produce videos that are only a few seconds long. However, even these few seconds (5-20 seconds) of videos are expected to bring significant changes to the industry. On today’s social media platforms, people usually watch a few seconds of vertical format videos and then continue scrolling. Short but impactful videos produced by AIs align with this consumption pattern. Additionally, they can be used in advertising.
For now, these tools will remain productivity tools supporting content creators. In the future, we will see these tools being used in music videos and even in the cinema industry.
There is nothing new to say about Sora, but its main competitor, Runway, has made significant improvements in generative video AI with Gen-3 Alpha. It now offers finer controls, greater consistency, and higher quality.
A startup named Luma AI, which we haven’t heard of before, has also released a similar model for general use, and it looks impressive as well. It is said that ByteDance, the biggest competitor of Kuaishou, will soon launch its own generative video tool. ByteDance, as is well known, is the owner of TikTok. Therefore, competition in the field of generative AI for video production is rapidly increasing.
