OpenAI introduces text-to-video model Sora

OpenAI CEO Sam Altman / Photo credit: Y Combinator

OpenAI has again flexed its generative AI prowess with Sora, a new model that can create videos from text.

Sora can generate videos up to a minute long while trying to match the user’s prompt. The model can create scenes that feature multiple characters and different types of motion.

While rivals like Google and Meta have already introduced similar tech, the content Sora produces appears to be more realistic. Sam Altman, OpenAI’s CEO, has shared output from the model based on prompts from users of X (formerly Twitter).

The company added that Sora has an understanding of language, visual concepts, and physics, enabling it to generate various subjects in different styles.

While Sora is not available to the public yet, OpenAI is already taking safety measures to prevent its misuse. The organization is working with red teamers – experts who test for security vulnerabilities – to check the model for possible abuse.

The firm is also building tools to help identify fake content generated through Sora, such as a detection classifier and a metadata tag. Moreover, it is implementing existing safety methods that it built for its products that use DALL-E 3, another AI model that can create images from text.

See also: What I learned about genAI from failing to scale my AI startup