OpenAI Can Now Create Videos From Prompts
OpenAI recently launched Sora, a new generative AI tool that can create realistic videos from text prompts. Users can create these videos from scratch or from an image and can extend existing content.
The videos created can contain multiple characters, movement, and complex backgrounds and scenes. Sora will use the text prompts to create representations in a physical world, understanding natural language and generating realistic videos that are up to a minute long while “maintaining visual quality and adherence to the user’s prompt.”
OpenAI cautions that the current version of Sora is still in its infancy and acknowledges that it may not understand all prompts accurately or circumstances of cause and effect. For example, if a person takes a bite of a cookie, the cookie may remain intact.
Before its wide release, Sora is being tested by a number of experts from outside the organization. OpenAI intends to publish their progress to keep the public up to date on the capabilities that are being created.
Some of the testers include artists, designers, and filmmakers who can provide feedback that will help OpenAI evolve the model to best help content creators. OpenAI is also relying on “read teamers,” experts in the domains of misinformation, hateful content, and bias to test the model and identify any potential risks.
OpenAI is implementing security features prior to any planned launch that will include adding metadata to videos to ensure that third parties can detect when a video is A-generated. There will also be restrictions on prompts that violate OpenAI policies, such as requesting content with “extreme violence, sexual content, hateful imagery, celebrity likeness, or the IP of others.”
OpenAI says that “despite extensive research and testing, [it] cannot predict all of the beneficial ways people will use our technology, nor all the ways people will abuse it. That’s why we believe that learning from real-world use is a critical component of creating and releasing increasingly safe AI systems over time.”
Sora is not the first generative AI model that is able to use text to generate video. However, ABI Research analyst Reece Hayden says it is the first to create them with the length and accuracy that OpenAI is claiming.
