Sora is trained on publicly available data, confirms OpenAI's CTO

In an interview with The Wall Street Journal, OpenAI CTO, Mira Murati confirmed Sora is trained on publicly available and licensed data.

Sora is the new text-to-video model by OpenAI introduced in mid-February. The AI model immediately became the talk of the town as it can generate stunning videos using text prompts. That said, it is not yet available for public use.

OpenAI trained Sora on licensed and publicly available data

Recently, Mira Murtati talked about Sora in an interview with Joanna Stern, a columnist for The Wall Street Journal, and here are some of the highlights.

When asked Murati about the data OpenAI used to train Sora, she said:

We use publicly available data and licensed data.

However, when asked about the source of training data for Sora, Murati’s answers were not convincing enough. She mentioned:

I am actually not sure about that, and I will not go into the details about the data used.

It is worth noting that OpenAI never misses a chance to showcase demo videos generated using Sora AI. It won’t be wrong to say that Sora seems to have set high standards when we talk of video quality.

Talking of the same, Murati said that generating a 20-second long 720p video clip could take a few minutes. She further said it generally depends on the complexity of the text prompts a user inputs. CTO Murati also hinted at the Sora release date. According to Murati, Sora is all set to be available for the public later this year. However, there was no mention of an exact date.

Although OpenAI’s CTO seemed to dodge questions about the training data for Sora, she lauded the text-to-video AI model’s ability to create hyperrealistic, beautiful, and highly detailed videos.

Time will tell whether it really will be a game-changer. Are you excited about the release of Sora? Let us know what you feel about the future of this text-to-video model.