Chinese App Kling Emerges as Rival to OpenAI's Video Generator Sora


A panda sitting by a calm lake is seen playing an acoustic guitar. The panda's surroundings include green grass and a rocky shore, with the lake reflecting the sky above. The scene is serene and whimsical, blending nature with a playful, musical element.
A screengrab of an AI video showing a panda playing guitar. Made by Kling.

Chinese app maker Kuaishou has released an AI video generator called Kling capable of producing impressive results that rival OpenAI’s Sora.

Although Sora has not been available to the public yet, Kling is being operated via a waitlist with remarkable AI-generated videos being created from the service. Although thus far it is exclusively available in China.

Users can generate 1080p videos at 30 frames per second up to two minutes long. The model adopts a “3D spatio-temporal joint attention mechanism, which can better model complex spatio-temporal movements, generate video content with larger movements, and at the same time conform to the laws of motion.”

Tom’s Guide notes that the diffusion transformer model is capable of generating 3D faces and bodies that help with facial expressions and movements.

The company behind Kling, Kuaishou, is already a major app player. Better known as Kwai overseas, it is the main competitor to TikTok inside of China boasting nearly 400 million daily active users, making it the second-largest short-video app in the country.

AI Video is Coming

The race for AI video is on with OpenAI heavily teasing Sora; collaboating with artists and music videos being made with the state-of-the-art technology.

Despite all the teasing, there is no release date for Sora yet and this is widely thought to be because of safety issues — especially given the proximity to the 2024 presidential elections.

Then there are copyright issues. Of course, we don’t know what training data was used to build Kling but then we don’t know what was used to build Sora either.

OpenAI has repeatedly refused to discuss Sora’s training data set in any detail or shed light on where OpenAI might have taken it from. The repeated line from OpenAI is that it is “publicly available” data which was also used for the AI image generator DALL-E.





Source link

Post a Comment

Previous Post Next Post