OpenAI has unveiled Sora, an AI tool capable of generating strikingly realistic videos of up to 60 seconds based on a simple text prompt. This marks a significant improvement in the quality of AI videos and “deepfakes,” which have previously been exploited to mislead voters.
The innovative tool, named “Sora,” will initially be accessible only to a select group of artists, filmmakers, and “red teamers.” The latter are researchers tasked with identifying potential malicious uses of an AI tool, as stated in OpenAI’s announcement on Thursday.
Sora is an extension of the technology that powers OpenAI’s image-generating tool, DALL-E. It interprets a user’s prompt, expands it into a more comprehensive set of instructions, and then employs an AI model trained on videos and images to generate the new video.
The past year has seen a rapid increase in the quality of AI-generated images, audio, and video. Companies like OpenAI, Google, Meta, and Stable Diffusion are in a race to develop more proficient tools and find avenues to commercialize them. However, advocates for democracy and AI researchers have raised concerns that these tools are already being used to deceive voters.
While this isn’t the first instance of such videos or audio being created, and other companies have developed their own text-to-video AI generators, the length and quality of the videos produced by Sora surpass what has been previously observed. Google is testing a tool called Lumiere, Meta has a model named Emu, and AI start-up Runway has been developing products to assist filmmakers in creating videos.
Ted Underwood, a professor of information science at the University of Illinois at Urbana-Champaign, expressed surprise at the level of sustained, coherent video generation. He cautioned that OpenAI likely selected videos that showcase the model at its best, but acknowledged that there seems to have been a significant leap in capacity from other text-to-video tools.
He further added that the need for a large team of artists over a lengthy period to create an animated feature could be eliminated. This prospect, he said, is exciting.
However, Gracey pointed out a significant issue: AI tools are trained on the work of real-life artists without providing them with compensation. She criticized this practice, stating that it’s unfair when AI takes advantage of people’s creativity, work, ideas, and execution without giving them the credit and financial compensation they deserve.