(Image source from: twitter.com/OpenAI)
OpenAI finally announced Sora, its artificial intelligence (AI) video generation model, on Monday. The company previewed the Sora to select people back in February and has now launched a variant of the model called the Sora Turbo. Sora can produce videos up to 20 seconds long in 1080p resolution. The AI model is currently deployed on an independent platform available as a website. Notably, Sora is currently only available to paying ChatGPT subscribers with a certain tariff limit. This artificial intelligence company announced the launch of Sora in a blog post and described the capabilities of this model. The Sora was first announced earlier this year, but the model has been delayed multiple times. The company cited the tightening of the model's security and data protection parameters as the reason for the delay. But after a delay of almost nine months, OpenAI has launched Sora as a standalone platform, which can be accessed here. Currently only available to ChatGPT Plus and Pro subscribers. Those who do not have a subscription are currently unable to create a new account on the site. Plus users, on the other hand, are limited to 50 480p videos or 50 720p videos per month.
The recently launched ChatGPT Pro subscription for $200 (approximately Rs. 16,970) per month allows users to produce videos with "10x more ease of use, higher resolution and longer watch time." However, similar to Reduce Number of Videos, the company did not provide details on the amount required for higher resolution and longer duration. Sora can now produce videos in wide, vertical and square aspect ratios. Users can also upload videos and images to enhance, mix and match content in the generated video. AI models can also create videos from scratch using text messages. Additionally, the Story interface allows users to configure specific inputs for each frame. Speaking of technical details, OpenAI explained that Sora is a propagation model that allows AI to look forward multiple frames at once and maintain content integrity within 20 seconds. The artificial intelligence model uses the transformer architecture and uses DALL-E 3 recaptioning technology.
Now you can generate entirely new videos from text, bring images to life, or extend, remix, or blend videos you already have. We’ve developed new interfaces to allow easier prompting, creative controls, and community sharing: https://t.co/HMcAKwaCCw pic.twitter.com/gtMiFkT6GB
— OpenAI (@OpenAI) December 9, 2024
OpenAI also highlighted the details of the model data. The company said it obtained a wide range of publicly available data through its data partners and from people working with the model. Public data is collected from machine learning and web crawling datasets. The company partnered with Shutterstock's Pond5 and commissioned datasets to create its own data for AI modeling. Finally, data on Sora was collected from AI specialists, red teams and employees.
To reduce the risks associated with a realistic AI video production model, OpenAI also adds visible watermarks and metadata in accordance with Content and Authenticity Alliance (C2PA) standards. The company also said it extended the model to include security measures for people uploading media. The AI company also said that Sora is prohibited from making videos that contain harmful forms of violence, such as child sexual abuse and sexual deepfakes. Also, the number of downloads you can perform initially is limited.
