Sora is a generative text-to-video AI model developed by OpenAI that can create realistic and imaginative scenes from written prompts, with some surreal video-game-like quality. It has been trained on an unspecified amount of video footage and uses "spacetime patches" to break down frames into smaller segments that are encoded in the spacetime patch. This allows it to generate videos that look great, with consistent details and accurate physics, but may struggle with complex scenes or precise descriptions of events. Sora can be used for a variety of tasks, including generating videos from text prompts, converting static images to videos, adding special effects, extending videos in time, and editing existing ones. However, there is also the potential for misuse, such as creating deepfakes, which could be a challenge for OpenAI's guardrails. The model is currently available only to "red teamers" and will be released to the general public once it has been trained on their testing results.