Sora Team Announces Ongoing Testing Phase - Not Yet Open for Public Use

Recently, the YouTube channel WVFRM Podcast featured an interview with the core team behind Sora, an innovative video generation model developed by OpenAI.

In the discussion, project leaders Bill Peebles, Tim Brooks, and Aditya Ramesh shared insights on Sora's current status and future direction. They emphasized that Sora is still in the user feedback phase and has not yet reached the point of being a complete product. As a result, it will not be available to the public in the near term, and there is no set timeline for when it will evolve into a fully realized offering.

When explaining how Sora operates, the team noted that it analyzes vast amounts of video data to learn how to generate realistic videos. The technology merges the capabilities of large language models, such as GPT, with diffusion models like DALL-E. While Sora's framework is more akin to GPT, its training techniques draw closer to those used in DALL-E.

Regarding Sora's training data sources, the team could only reveal that it utilizes a combination of publicly available datasets and those authorized by OpenAI. Although there have been reports that Sora can generate relatively high-quality videos from text prompts at a rapid pace, industry experts still point out the limitations of current AI tools, which cannot fully replace the work of human creators.

Film concept artist Reid Southen remarked, "Many believe Hollywood is coming to an end, but I hold a different view. The production processes in Hollywood are complex, and these generated videos still face numerous challenges, particularly concerning temporal consistency and other details."