1 week ago
6
OpenAI set the industry pace for artificial intelligence (AI) developments when it launched ChatGPT and DALL-E 2, creating a chatbot and image generator craze. Now, OpenAI is expanding its offerings by delving into the AI video generation space with the much-anticipated launch of Sora.
Also: What is ChatGPT Pro? Here's what $200 per month gets you
On Monday, during its "12 days of OpenAI" event -- a holiday-themed campaign in which the company live streams a new demo or product launch every day for 12 days -- OpenAI unveiled Sora Turbo, a model that can generate videos from text or users' assets.
This version builds on the Sora video generation model launched in February to a select group of red teamers and testers, with OpenAI touting that it is "significantly" faster. The experience is available as a standalone product at Sora.com to ChatGPT Plus and Pro users at no additional cost.
The new Sora interface is similar to the standalone DALL-E 2 experience when it first launched. It includes a Featured and Recent feed that allows users to explore other creations from the community and see how they made them for inspiration.
Also: OpenAI's AI transcription tool hallucinates excessively - here's a better alternative
Beyond that, users can generate videos using text or their own assets with up to 1080p resolution. Videos can be up to 20 seconds long and can be widescreen, vertical, or square aspect ratios. When using their own images and videos, users can use Sora Turbo to extend, remix, replace, remove, or blend existing content.
The interface also features a storyboard tool that allows users to generate different input frame-by-frame to create a new video sequence.
OpenAI warned that Sora Turbo has "many limitations." For example, it "generates unrealistic physics and struggles with complex actions over long durations."
Also: These AI avatars now come with human-like expressions
To prevent Sora from generating harmful content, OpenAI has blocked it from creating damaging forms of abuse, including child sexual abuse materials (CSAM) and sexual deepfakes (the company did not specify restrictions on any other types of deepfakes). At launch, uploading assets with people in it will be limited, with the feature rolling out to more users over time as deepfake mitigations are refined.
All the videos generated by the model will also include C2PA metadata, which helps ensure the content's metadata discloses that it was created by Sora. Content will also have visible watermarks, although OpenAI labels them as "imperfect." Lastly, OpenAI built an internal search tool to help verify if content came from Sora.
For more details on the model performance, the company released a Sora System Card that includes the model's data, risk analysis, external red teaming findings, learnings from Early Artist Access, and more.
Both ChatGPT Plus and Pro subscribers can access Sora with varying limitations. With the ChatGPT Plus subscription, which costs $20 per month, users can generate up to 50 videos per month at 480p resolution or fewer videos at 720p.
Also: How Apple, Google, and Microsoft can save us from AI deepfakes
With the recently unveiled Pro Plan, which costs $200 per month, users get "10x more usage, higher resolutions, and longer durations," OpenAI said. Other paid subscribers, such as ChatGPT Enterprise, Team, and Edu users, do not have Sora access included in their plans.
OpenAI shares that it is working on a tailored pricing model for different types of users, which will be available next year. The company also plans to expand access, which is available everywhere ChatGPT, with the exception of the UK, Switzerland, and the European Economic Area, in the coming months.
OpenAI also shared a video on how to get started with the new model.
English (US) ·