Manswi Panchbhai
Sora 2 is OpenAI’s newest text-to-video generation model, which is designed to offer greater realism, improved controllability, and better adherence to the laws of physics.
Sora 2 avoids bizarre errors e.g., a basketball bouncing correctly instead of teleporting.
It is capable of adhering to intricate instructions over several scenes and can manage cinematic, anime, and realistic styles.
OpenAI is also launching an iOS app for Sora, with a social twist, users can create, remix videos, and even add themselves or friends into scenes using the “cameo” feature after a one-time video and audio recording.
Sora is capable of producing videos with a resolution of up to 1080p and a maximum duration of 20 seconds. Videos can be made in widescreen, vertical, or square formats.
Sora 2 includes features for background audio, conversations, and sound effects. Additionally, it allows users to integrate themselves or actual objects into the created scenes through cameo.
The application allows users to manage recommendations, utilizing a natural-language algorithm that favors content from individuals they follow.
In the initial access period, the model processed over 500,000 user requests from more than 60 different nations. The expense of training Sora 2's video generation model was around US $200,000 for computing and infrastructure.