OpenAI introduced Sora 2, its new AI model capable of generating lifelike clips of real people speaking in multiple languages. The release includes a TikTok-style iOS app called Sora, which lets users create and remix “cameos” using their own likenesses, which The Verge's Hayden Field describes as “essentially an app full of deepfakes, on purpose.”
Currently the Sora app is available to users on iOS in the U.S. and Canada on an invite-only basis, with more countries and invitations coming soon.
The team behind Sora 2 said that they had been working on the model for over 20 months and that the new version:
- is way more accurate when it comes to following user prompts.
- introduces the ability to synchronize audio and video.
- is “moving us closer to useful world simulators.”
- brings much smarter physics, introducing the ability to accurately do backflips on top of a paddleboard with all of the fluid dynamics and buoyancy accurately modeled.
The Sora app also introduces the ability for users to remix your videos into “Cameos” using your AI-generated likenesses. After recording a short head-movement video in the iOS app, users can authorize others to generate clips featuring them.
“Hey everyone in my life, make videos of me saying ridiculous thing and publish them on the Internet!”
OpenAI employees said they now use Sora 2 internally for communication in lieu of text messages, emojis, and voice notes.
Some people criticized the new Sora as a cash grab by OpenAI that's going to be responsible for creating an endless amount of AI slop on the web, to which Sam Altman replied:
“We do mostly need the capital for build[ing] AI that can do science, and for sure we are focused on AGI with almost all of our research effort. It is also nice to show people cool new tech/products along the way, make them smile, and hopefully make some money given all that compute need.”
The Wall Street Journal reported that prior to Sora's launch last week, OpenAI had been telling Hollywood studios and agencies that they needed to explicitly opt out if they didn't want their IP to be included in Sora-generated videos. However now the company has changed its tune and said that it is planning to give copyright holders “more granular control over generation of characters, similar to the opt-in model for likeness but with additional controls.”
Holy shit, OpenAI — pay people for their copyrighted IP! First our books and website content; now our copyrighted characters and personal likeness?
I guess why wouldn't they ask for forgiveness instead of permission. They've effectively gotten away with every other copyright violation thus far.