Kuaishou Technology has unveiled Kling AI, a tool for generating highly realistic videos from text inputs, competing with OpenAI’s Sora model.
Kuaishou Technology, a Chinese tech business renowned for its social platform and content community, has introduced a new tool named Kling AI. Its purpose is to create extremely lifelike films using written instructions; this could be a rival to OpenAI’s Sora model.
Based on textual input, the Sora AI model can create imaginative and realistic scenes. A deep learning architecture is the backbone of Sora’s video-generating capabilities. The model connects words and phrases to objects, activities, surroundings, and stylistic details as it learns.
In comparison to Kling AI’s two-minute realistic videos at 1080p and 30 frames per second, OpenAI’s Sora model can only generate videos up to one minute in length. The AI-generated films are almost indetectable from the originals since they support various aspect ratios and produce a high degree of detail using sophisticated 3D face and body reconstruction algorithms.
In order to generate complex facial expressions and limb motions from a single full-body image, Kling AI employs a 3D Variational Autoencoder (VAE) for body and face reconstruction. This technology’s 3D spatiotemporal joint attention mechanism enhances the model’s ability to handle complex situations and motions. That way, we know the created content won’t break any rules of physics.
Although there are regional limits, the application takes an open-access approach. The trial phase of the model is now underway.
Kuaishou Technology introduces a new text-to-image model: KwaiYii LLM and Colors
There are a number of Chinese tech businesses, including Kuaishou Technology, that are trying to displace Sora. Its most current AI advancement is text-to-video capacity; it followed in May with the launch of its text-to-image model Kolors and the introduction of its KwaiYii large language model (LLM).
The “AI Dancer” function in Kuaishou Technology and Kwaiying, two apps for making films, lets users add a still image of a character and then have them dance to predetermined beats and choreography.