OpenAI Reveals Stunning Text-to-Video Model

Less than 16 months after the release of its ChatGPT model helped launch the artificial intelligence boom, OpenAI has revealed Sora, a text-to-video model that can create realistic video in seconds. "Sora can create videos of up to 60 seconds featuring highly detailed scenes, complex camera motion, and multiple characters with vibrant emotions," OpenAI said in a post on X. Analysts said Sora, named after the Japanese word for sky, appears to be far ahead of rival projects from Meta and Google, raising exciting possibilities—and troubling ones.

• A look at what it can do. The company shared stunningly realistic videos it said were created by Sora without modification, including a woman walking on a Tokyo street and mammoths walking in a snowy field. It also shared animations, including one it said was created with a prompt asking for a "gorgeously rendered papercraft world of a coral reef, rife with colorful fish and sea creatures."

• It's not currently available to the public. In a blog post, OpenAI said Sora has been made available to "red team" researchers who will "assess critical areas for harms or risks," along with selected artists and filmmakers, who will deliver feedback on how to make it "most helpful for creative professionals." OpenAI said that as with its DALL-E image generators, Sora will "reject prompts that are in violation of our usage policies, like those that request extreme violence, sexual content, hateful imagery, celebrity likeness, or the IP of others."
• Safety tools. OpenAI says it is working on tools that can detect when a video has been created by Sora, which will be deployed before it. A spokesperson tells NBC News that when experts have tested its ability to cause harm, a system card describing safety evaluations will be issued,
• How to try it. Cecily Mauran at Mashable notes that reactions on social media "can be divided into two camps: 'Is Sora going to make video production obsolete?' and 'How can I try it?'" For now, it seems the only way to get a Sora video made is to ask OpenAI Sam Altman, who has been sharing videos on X made with prompts suggested by users, including one of a "bicycle race on ocean with different animals as athletes riding the bicycles with drone camera view."

• This happened faster than expected. OpenAI warns that the model has limitations—it can "struggle with accurately simulating the physics of a complex scene," for one, but experts say it goes beyond any of its rivals in terms of the length and quality of videos, the Washington Post reports. University of Illinois information science professor Ted Underwood says he was surprised by the "jump in capacity." "I didn't expect this level of sustained, coherent video generation for another two to three years," he says.
• Some major concerns. Analysts warn Sora and similar tools could put a lot of digital artists out of work—and spread even more online disinformation. "I am absolutely terrified that this kind of thing will sway a narrowly contested election," Oren Etzioni, an AI expert at the University of Washington, tells the New York Times. In Hollywood, some predict that the technology could replace hundreds of workers on animated films—along with human extras.

(More OpenAI stories.)