OpenAI’s recent unveiling of its text-to-video AI, Sora, has generated both excitement and disappointment among AI enthusiasts and professionals. While the initial videos produced by Sora were impressive, further exploration into its capabilities has revealed flaws that seem to be common among video and image-generating AI programs.
One of the issues with Sora’s outputs is its flawed sense of physics and unrealistic depictions. For instance, in a video created from the prompt of a bicycle race on the ocean with different animals as athletes, some of the animals appear to be floating in the air instead of riding bicycles. Similarly, in a video depicting New York City submerged like Atlantis, the sea creatures are shown swimming above the waterline, defying any logical sense.
Moreover, Sora often produces clips with bizarre and distorted visuals, as well as body horror elements. For example, an animation of a cat waking up its owner shows the owner’s shoulder morphing into the comforter, and a hand suddenly appearing from the blankets in an unrealistic manner. Another clip featuring archeologists excavating a plastic chair in the desert starts off convincingly realistic, but then the chair begins to morph, melt, and float in mid-air.
Although Sora has not been made publicly accessible yet, it has been shared with a select group of professionals, including visual artists, designers, and filmmakers, who provide feedback to improve the model for creative professionals. However, based on its current state, it is unlikely that Sora is capable of producing fully-fledged films or animations that meet the standards of human animators.
Despite Sora’s flaws, other text-to-video AI services have garnered interest, and Hollywood studios, known for their reliance on computer-generated imagery, have shown interest in integrating AI technology. While Sora’s outputs may be questionable, there is still a likelihood that users, including individual creators or studios looking to reduce costs, will embrace AI technology once it becomes available. The impact of automation on creative professionals and the content economy remains to be seen.
Overall, Sora’s text-to-video AI has received mixed reviews, with both impressive and questionable outputs. Its flaws in physical understanding, unrealistic depictions, and distorted visuals raise concerns about its capabilities. However, the interest in AI-generated content and its potential impact on creative industries persists, despite the limitations of the current technology.
