To use this site, please disable the ad blocking feature and reload the page.
This website uses cookies to collect information about your visit for purposes such as showing you personalized ads and content, and analyzing our website traffic. By clicking “Accept all,” you will allow the use of these cookies.
Users accessing this site from EEA countries and UK are unable to view this site without your consent. We apologize for any inconvenience caused.
REUTERS/Dado Ruvic/Illustration/File Photo OpenAI logo is seen in this illustration taken May 20, 2024.
The Washington Post / Will Oremus
16:10 JST, December 10, 2024
ChatGPT maker OpenAI on Monday launched Sora, a video generator that lets people create short, realistic-looking video clips just by typing out a description of a scene.
The new service is not the first widely available video generator, but OpenAI’s position as a market leader in artificial intelligence could help make synthetic footage increasingly common in commercials, art projects, movies and social posts – but also hoaxes and deepfakes.
Video generation has been billed by AI executives as the next leap for generative AI technology, which also powers chatbots, image generators and audio generators. Last week, Google opened its own AI video generator, Veo, to customers of its cloud services.
A feed on OpenAI’s website of clips recently created with Sora showed the tool capable of highly realistic footage and a range of visual styles but also distortions of everyday physics and human anatomy.
Clips of Christmas trees twinkling through the window of a house on a snowy street and an atmospheric black and white scene of monks poring over ancient tomes appeared close to photorealistic. In other clips, the limbs of acrobats and animals morphed into impossible shapes, showing the technology is capable of errors no human would make.
Interactions between human limbs and other objects or people – like playing soccer or a group hug – seemed particularly challenging for the technology.
The glitches suggest OpenAI has yet to resolve some of the problems highlighted in earlier tests of Sora, including by The Washington Post in February, that showed it struggling to depict processes such as a man lighting a cigarette.
“It is clear from glancing at Sora videos that the hard problems of video generation haven’t been solved,” said Arvind Narayanan, a professor of computer science at Princeton University. But he added that its release could have a commercial impact even if it doesn’t necessarily herald a major technological breakthrough.
“Despite these serious limitations, Sora will probably be very useful in some contexts because OpenAI seems to have put a lot of work into developing an actual product instead of simply releasing a model and letting users figure out what to do with it,” he said.
OpenAI’s entry into video generation has been highly anticipated, and on Monday afternoon, sign-ups were disabled on the company’s website. An error message blamed “heavy traffic,” and a Post journalist was not able to access the tool.
The company did not respond to a request for comment on whether it ran into unexpected problems with Sora’s public rollout.
Sora, initially announced in February, is available to only paying subscribers to OpenAI’s chatbot, ChatGPT.
The lowest cost plan, $20 monthly, allows creation of 50 videos with Sora per month at lengths up to five seconds. Paying more per month lets a user generate more videos, of greater length and higher resolution.
OpenAI said Sora would be available at launch to users in the United States and many other countries, but not in Britain or Europe. CEO Sam Altman said on X that the company wants to bring Sora there, but that “we also have to comply with regulation.”
On first look, Sora “feels competitive, although not far beyond the current standard for text-to-video world models,” said Gaurav Misra, CEO and co-founder of Captions, a start-up that makes AI tools for video creation and production.
He predicted that the stock-footage industry will be the first disrupted by Sora and other AI-based text-to-video programs.
Researchers who study AI and the impact of misinformation have warned that AI video tools can be used to create deepfakes, or realistic videos that deceive people by purporting to show real-world events that never happened. AI-generated or altered videos have already become a tool of harassment, especially against women, and experts fear they could also be used to manipulate elections.
In a document on Sora’s capabilities and vulnerabilities released Monday, OpenAI acknowledged that its service could “introduce novel risks, such as the potential for misuse of likeness or the generation of misleading or explicit video content.”
The company said it has tried to guard against those risks by filtering violent and sensitive content from the data used to create Sora and performing “red team” tests to identify loopholes.
As Sora was being readied for release, OpenAI temporarily suspended user access to the tool late last month. The move came after some artists whom the company invited to test the service launched a webpage that allowed the public to use Sora for free, alongside an open letter protesting that the company was using early testers as unpaid labor.
