OpenAI will launch its text-to-video generator Sora later this year, but the rollout will be slow amid concerns that the technology could threaten jobs and spread misinformation. 

In an interview with The Wall Street Journal, OpenAI CTO Mira Murati says it's expensive for Sora to generate realistic video clips, but the company is optimizing the technology to require less computing power for its upcoming release. 

Currently, Sora can only create clips up to 60 seconds long, without any voice or sound. But Murati says OpenAI plans on adding audio to the Sora technology "eventually."

Sora may leave some in Hollywood concerned about their future job prospects given that it can churn out realistic clips that are significantly better than other text-to-video generators. But Murati notes that Sora still makes mistakes, like misunderstanding the user’s prompt or struggling to faithfully render people’s hands or the movement of cars passing by. 

Recommended by Our Editors

Google's Agentic AI Tool Gemini Spark Is Now Available. Here's How to Try It

I Used ChatGPT to Build Complex Spreadsheets Fast. Here’s How You Can, Too

GitHub Copilot Costs Skyrocket As Users Are Pushed to Per-Token Billing

To address concerns about misuse, OpenAI plans on adding a watermark and metadata to all the video clips that Sora produces, making it easier for the public to identify AI-generated footage. In addition, Sora will be barred from generating images of public figures, tamping down its ability to deepfake politicians and celebrities. 

Sora may also face legal challenges—a problem OpenAI is already experiencing with ChatGPT. The Wall Street Journal asked Sora to generate a video of a mermaid reviewing a smartphone with her crab assistant, for example, and the result made the crab look strikingly similar to a character from SpongeBob SquarePants. 

According to Murati, OpenAI has been using publicly available videos and licensed data, including content from stock image provider Shutterstock, to train the program to render life-like video. But it's not clear if Sora has been trained on YouTube or Facebook videos, too. Surprisingly, Murati said she didn't know if videos from YouTube, Instagram and Facebook were included in the training data.