OpenAI has introduced Sora, its latest diffusion model, designed for text-to-video creation. This new AI model, from the team behind ChatGPT, enables the generation of videos in various resolutions and aspect ratios.
Furthermore, it allows editing existing videos, facilitating swift modifications to the scenery, lighting, and filming technique directly from a text command. Sora also can create videos from a still image or augment existing video content by adding in missing frames.
The sample videos created by Sora look very promising. OpenAI says that the AI model can generate minute-long videos in 1080p resolution. Here is a sample video from the company’s landing page for Sora.
OpenAI describes Sora as follows:
Sora can generate complex scenes with multiple characters, specific types of motion, and accurate details of the subject and background. The model understands not only what the user has asked for in the prompt, but also how those things exist in the physical world.
Sora works by breaking down an image or video and presenting it as smaller units called patches. ChatGPT also uses this transformer architecture. A video starts as static noise in Sora with the AI model working to remove the noise from the final product.
OpenAI has announced that it is applying the safety measures developed for DALL·E 3 to its new model, Sora. The model is undergoing rigorous testing by “red teamers,” specialists tasked with evaluating the model for any potential risks before its public release.
Furthermore, OpenAI is engaging in discussions with policymakers, artists, and educators to identify possible issues and applications for Sora. As of yet, no official release date has been announced. However, OpenAI says that Sora will launch as a product soon, meaning it is likely coming to ChatGPT as well.
