Sora is the new artificial intelligence tool developed by OpenAI that allows for the astonishing generation of videos from text. Its objective is to teach AI to understand and simulate the physical world in motion, with the purpose of assisting people in solving problems that require real interaction with the environment.
This innovative tool, called Sora, is capable of generating videos up to one minute in duration while maintaining exceptional visual quality and accurately following user instructions. Initially, access is being granted to risk assessment teams (red teamers) to analyze critical areas for potential damages or risks. Additionally, visual artists, designers, and filmmakers have been invited to provide feedback on how to improve the model and make it more useful for creative professionals.
OpenAI has decided to share its early research progress to collaborate with external individuals and obtain broader feedback. Sora has the ability to generate complex scenes with multiple characters, specific movements, and precise details of the subject and background. The model not only understands what the user requests in the text but also how those elements exist in the physical world. Moreover, thanks to its deep understanding of language, Sora accurately interprets instructions and generates compelling characters that express vibrant emotions. It can also create multiple shots within a single generated video, maintaining coherence in characters and visual style.
It is important to note that the current model has some limitations. It may struggle to accurately simulate the physics of complex scenes and may not always comprehend specific causal relationships. However, OpenAI is taking important safety measures before implementing Sora in its products. Rigorous testing will be conducted in collaboration with experts in areas such as misinformation, offensive content, and biases to identify potential issues.
Furthermore, tools are being developed to detect deceptive content, and classifiers will be implemented to identify videos generated by Sora. OpenAI is also committed to involving policymakers, educators, and artists worldwide to address their concerns and find positive use cases for this new technology.
Sora is based on a transformer architecture similar to that of GPT models, ensuring scalable and superior performance. It represents videos and images as collections of smaller data units called patches, which enables training the model on a wide range of visual data with different durations, resolutions, and aspect ratios.
This is a groundbreaking tool in video generation from text, with the ability to understand and simulate the real world, representing a significant milestone in the quest for Artificial General Intelligence (AGI). OpenAI continues to work on improving and securing Sora and is committed to collaborating with the global community to maximize the benefits and minimize potential abuses of this new technology. Have you already seen some of its results?