Video generation AI has finally been released as open source. Video generation AI technology that was once the exclusive domain of just a few tech corporations is now available to everyone.
VentureBeat reported on the twenty seventh (local time) that researchers from China’s Jifu AI and Tsinghua University have released an open source text-to-video model called ‘CogVideoX’. Post your paper within the archiveIt was reported that.
The 5 billion-parameter CogVideoX-5B generates high-quality 720×480 video at 8 frames per second and as much as 6 seconds long, with text prompts.
The researchers claimed that in benchmarks, it outperformed competing products resembling ‘Pika’, ‘VideoCrafter-2.0’, and ‘OpenSora’ in several indicators.
The researchers implemented a ‘3D variational autoencoder (VAE)’ to efficiently compress videos and developed an ‘expert transformer’ to enable detailed interpretation of text prompts and accurate video generation.
Particularly, the code and model weights of the CogVideoX-5B On the cuddling face It has made video creation technology, once the exclusive domain of open, well-funded technology corporations, virtually free.
This is taken into account to be a big turning point within the AI ​​field, as technology that was previously inaccessible attributable to lack of resources is now available to everyone.
Because of this with the participation of the worldwide developer community, we will expect rapid development in the sphere of AI-generated videos.
Meanwhile, on the twenty seventh of last month, Jifu AI released the ‘Ying’ model, which might generate a 6-second video in 30 seconds.
Also, the language model ‘ChatGLM’ released last yr is evaluated as essentially the most performing in China, and it’s a representative AI startup that earned the nickname of ‘China’s OpenAI’.
Reporter Park Chan cpark@aitimes.com
