In the ever-evolving landscape of AI, OpenAI released yet another groundbreaking demonstration of their newest AI model, Sora. Sora is a Text-to-Video model, taking a text input as a prompt and generating a corresponding video. Despite being relatively new, Sora is already producing some pretty impressive results, many of which could easily be mistaken for real video footage if not scrutinized.
With an architecture similar to ChatGPT’s, Sora uses DALL-E’s diffusion method to generate videos. Essentially, Sora begins with random static and continuously fills it in to get closer and closer to the text prompt. However, it doesn’t go frame by frame. Instead, it generates the whole video at once, creating earlier parts of the video knowing what it will look like later in the video. This allows it to keep details consistent throughout the entire video, one of the main reasons why the results are so impressive. At the end of the process, Sora has created a long, detailed, dynamic clip that adheres to the prompt.
These capabilities make it perfect for certain tasks:
- Adding CGI to existing videos
- Creating a video out of a static image
- Extending existing videos
- Filling in gaps between existing videos
- Removing/replacing objects from existing videos
Sora would make an excellent tool for movies and other visual productions. However, it’s much more than a creative tool to be used by graphic designers and filmmakers. To create realistic simulations, it has to understand how objects interact with each other in the real world. Essentially, Sora has “accidentally” learned a physics engine in the process of learning how to create videos.
All in all, Sora is a revolutionary AI system. It’s a strong video production tool and potentially a soft physics simulator. More importantly, however, Sora is a symbol of AI evolution. By building on ChatGPT and DALL-E to perform a unique task, Sora demonstrates how the growth of AI is exponential. The more advanced AI technology is, the easier it will be to build on it and improve it even further. In the future, Sora might provide the foundation for an even more groundbreaking AI.
See more like this:
