✨ OpenAI Unveils its New Video-Generation Model Sora
posted 16 Feb 2024
OpenAI has introduced a new neural network named Sora, capable of generating high-quality videos up to one minute long based on textual prompts. Initially, this technology will undergo testing by a select group of users, including researchers and artists, before being rolled out to the general public.
According to the company, the standout feature of Sora is its exceptional ability to grasp the interactions between objects in the real world and to conduct comprehensive analyses of input prompts. It is capable of generating not only entirely new video content but also animating static images (art, photographs) and interpolating missing frames.
Despite showcasing some impressive capabilities, such as creating cartoon animations, OpenAI acknowledges Sora's limitations. The neural network faces challenges with complex dynamic scenes and accurately capturing cause-and-effect relationships, which can lead to mixed outcomes, like the scenarios with wolf pups or a Chinese parade.
OpenAI is also actively developing strategies to prevent undesirable text prompts and to ensure videos are free from violence, hate speech, and other disallowed content. To combat misinformation, developers plan to release a separate mechanism to identify Sora-generated videos.
More breaking news
