OpenAI's Sora Video Prototype Undergoes Major Improvements
OpenAI is making significant strides in enhancing its Sora video prototype, originally introduced in February. According to a report by The Information, the company aims to develop a new version of Sora that will be faster and capable of producing longer, higher-quality videos. The initial prototype faced criticism due to its slow production speed, taking over 10 minutes to create a single video, and struggling with consistency in patterns, objects, and characters. Filmmaker Patrick Cederberg reported that he had to generate hundreds of clips to obtain one that met his standards, highlighting the challenges with physics errors and overall quality.
To address these shortcomings, OpenAI is collecting millions of hours of high-resolution video data as part of its training process. This extensive dataset will help the new Sora model cover a wider range of styles and topics while reducing bias, ultimately improving the quality and consistency of the videos produced. In early September, OpenAI stated that Sora remains under investigation as discussions with policymakers continue. CEO Mira Murati indicated that the improved version of Sora could be available later this year, although production costs must be lowered, as it currently exceeds the expenses associated with existing AI systems.
