1 week ago
6
After announcing its new text-to-video AI generator, Sora, back in February, OpenAI has finally released it to the world.
Sora has experienced some significant changes since its limited debut in February. While a limited group of users had access to Sora in February, primarily for testing and safety purposes, the general public has had a lot of waiting to do — aside from a malicious leak in late November, anyways.
As OpenAI moves its new video generation model “out of research preview,” the company is also considering what Sora’s underlying technology might mean for its broader AI goals.
“Sora serves as a foundation for AI that understands and simulates reality — an important step towards developing models that can interact with the physical world,” OpenAI explains.
Concerning Sora’s underlying technology, how it was trained has remained quite mysterious. A month after unveiling Sora, the company’s CTO, Mira Murati, refused to explain Sora’s training set, saying only that the company used “publicly available data.”
Two months later, OpenAI stuck to its guns and again would not disclose precisely where the training data came from, although this time it was the company’s COO, Brad Lightcap, avoiding the questions. As an aside, Murati left OpenAI in September.
Back to Sora, OpenAI developed a new version of the video generator, Sora Turbo, that is “significantly faster” than the version seen in February and is available to ChatGPT Plus and Pro users. ChatGPT Plus starts at $20 per month, while Pro is $200.
The company also upped Sora’s capabilities. Sora can now create videos up to 1080p resolution and generate clips as long as 20 seconds. “You can bring your own assets to extend, remix, and blend, or generate entirely new content from text,” OpenAI explains.
Alongside the beefed-up performance, Sora has a new interface that mimics a storyboard, enabling users to input AI-generated content at specific frames. Much like new Firefly tools in Premiere that Adobe unveiled at MAX in October, OpenAI is implementing generative AI that can extend existing clips, filling in blank spaces or stretching content to fit specific needs.
Despite being a long time in the making, OpenAI admits that the version of Sora it is deploying to the public today “has many limitations.” Sora “often generates unrealistic physics and struggles with complex actions over long durations.”
“We’re introducing our video generation technology now to give society time to explore its possibilities and co-develop norms and safeguards that ensure it’s used responsibly as the field advances,” explains OpenAI.
All Sora-generated videos include C2PA metadata and the company has implemented safeguards, including visible watermarks by default — although these can be removed. Sora also restricts users from creating content with “particularly damaging forms of abuse,” including child sexual abuse materials and sexual deepfakes. Uploads involving people will be limited at launch as OpenAI works on deepfake mitigation tools.
Although Sora has just arrived for the general public, some big names in the tech space have been working with it ahead of today’s launch, including Marques Brownlee, better known as MKBHD.
Brownlee has been using the newest version of Sora for the past week and has found a lot of things it is good at, as well as many others that it struggles with.
“The results I’ve gotten from it are both horrifying and inspiring at the same time,” he says. Beyond this, Brownlee asks many interesting questions, including how Sora has been trained, how much energy it uses, and whether people are ready for something like this to be so accessible and widely available.
“We hope this early version of Sora will enable people everywhere to explore new forms of creativity, tell their stories, and push the boundaries of what’s possible with video storytelling. We’re excited to see what the world will create with Sora,” OpenAI concludes.
Pricing and Availability
Sora is available now from a new dedicated Sora website. Sora is available in ChatGPT Plus and Pro subscriptions, which are $20 and $200 monthly. The Plus version enables users to create up to 720p videos that are five seconds long, while the 1080p and 20-second clips are restricted to Pro users. Only Pro subscribers can download Sora-generated videos without the watermark, although anybody can crop it out, as it is in the bottom right corner of Sora’s generations. Many examples of what Sora can do are available on OpenAI’s website.
Image credits: OpenAI
English (US) ·