OpenAI has plans to introduce its latest generative artificial intelligence (AI) model Sora later this year, as per a report. Sora, introduced last month, is an AI-powered text-to-video generator which the company claims can generate up to 60-second-long videos, much longer than any of its industry rivals. In a recent interview, the AI firm’s Chief Technical Officer also revealed that Sora can generate videos within a few minutes depending on the complexity of the prompt. The interview also covered topics such as the imperfection of the model, the data used to train it, and the possible price point of Sora.
In an interview with The Wall Street Journal, OpenAI CTO Mira Murati, on the question of when the AI video generator would be available, said, “I’m hoping definitely this year, but could be a few months.” On being asked whether the US presidential election, which is scheduled for November 2024, will play a part in the company’s considerations on when to release the AI model, Murati said that dealing with misinformation and harmful bias is important to the company and added, “We will not be releasing anything that we don’t feel confident on how it might affect global elections.”
Murati was also asked about the source of data on which Sora was trained. The question is also of interest to Italy’s data protection watchdog which is investigating possible data violations by OpenAI. The regulators gave the AI firm 20 days to provide information on the AI video generator.
In the interview, Murati also appeared not to reveal any details and just stated that the AI model was trained on data available in the public domain and which was licensed to the company. She also said that she was unsure whether data from YouTube or Facebook was used but confirmed to the publication that Sora used content from Shutterstock, with whom they have a partnership.
Sora is also a very expensive model to run compared to other OpenAI tools such as ChatGPT and DALL-E, explained Murati. “ChatGPT and DALL-E are optimised for the public to be using them whereas Sora is really a research output. It is much, much more expensive,” the CTO told the WSJ. While the company has not finalised a price point for the AI video model, Murati said the company is trying to make the model available, once it is released, at a similar cost as DALL-E. At present, the base pricing for DALL-E 3 model is $0.040 (roughly Rs. 3) per image in a 1,024 x 1,024 pixels resolution.
Further, the OpenAI CTO also highlighted that the company was working to bring audio capabilities to Sora, optimise the time of generation and its cost-effectiveness, and make further improvements to minimise imperfections.
