Just like ChatGPT,
OpenAI
’s
Sora
text-to-video generator took the world by storm. Its ability to create realistic videos with text prompts left people tempted to get their hands on the software but they may not have to wait for long. The company
CTO Mira Murati
has said that the company is making the
AI tool
generally available “this year” and that it “could be a few months.”
OpenAI working on sound with Sora
When OpenAI showed off Sora in February, none of its videos or even the ones shared on X (formerly Twitter) had any sound.
In an interview with The Wall Street Journal, Murati said that the company is working to bring sound to videos. However, the teams are “not working on [sound] in this moment” but eventually will.
Adding audio to Sora-created videos, which Murati agreed were imperfect, will potentially make them even more realistic. Additionally, Murati said that OpenAI is also working to provide an ability that will allow users to fix issues in the videos that Sora creates.
“We’re trying to figure out how to use this technology as a tool that people can edit and create with,” Murati said.
Murati on data used by OpenAI to train Sora
OpenAI has been mired into controversies as it is alleged to have used copyrighted data to train AI models. When asked about the data used to train Sora, Murati replied, “Publically available data and licensed data.”
When pressed on what data was used like YouTube or Facebook or Instagram, Murati did not specify and looked to dodge the question.
“I’m not going to go into the details of the data that was used, but it was publicly available or licensed data,” she added. She, however, confirmed that Sora uses content from Shutterstock, with which OpenAI has a partnership.
Cost to run Sora
Murati said that since Sora is still a research project, it is “much more expensive” to power unlike DALL-E which is optimised for public use. The company is trying to make the tool “available at similar costs” to DALL-E, which is OpenAI text-to-image model.
Murati also noted that OpenAI is re-considering the release of more AI tools as elections approach. She said that the company is putting guardrails which will limit the AI model to generate images or videos of public figures. She also said that the company was working on issues like misinformation and not releasing systems till problems are mitigated.
English (US) ·