OpenAI announced its new diffusion model Sora that harnesses the power of text in video creation. ChatGPT Creator’s new AI model is capable of creating videos of different resolutions and aspect ratios and can also edit existing videos for quick changes of scenery, lighting and shooting style from text prompts. Sora can create videos based on a still image or expand existing videos by filling in missing frames.
OpenAI shared that Sora is currently capable of producing up to one minute of Full HD video content, and the examples we’ve seen are promising. For more generated video samples you can visit Sora’s landing page.
It works using a transformer architecture similar to ChatGPT where videos and images are represented as small units of data called patches. The videos generated by Sora start out as static sound, and the model gradually removes the noise to create the final product.
OpenAI shared that it is using its existing security protocol used in DALL·E 3. Sora is currently being tested by “Red Teamers” – experts who will run tests and assess the model for potential risks before it is officially launched.
OpenAI will conduct discussions with policymakers, artists and academics to look at potential concerns and use cases for Psora. No official launch date has been given yet.
