OpenAI has launched a new AI model that can generate realistic videos from text prompts. This generative AI model is called Sora and can generate up to 1 minute long 1080p videos with multiple characters, complex scenes, different motions, and proper details.
Sora is a diffusion model that uses a transformer architecture to unlock superior scaling performance Sora has been trained with publicly available videos and videos licensed from copyright holders.. OpenAI claims that Sora is built upon past research conducted for DALL-E and GPT Models. Sora is also capable of generating a video from the input image. It can also extend existing videos by adding more frames to it.
“Our largest model, Sora, is capable of generating a minute of high-fidelity video. Our results suggest that scaling video generation models is a promising path towards building general purpose simulators of the physical world,” says OpenAI.
On the official website, OpenAI has shared several videos generated by Sora along with the text prompt the company used to generate the video. I observed all the videos and found them impressive. Although the AI model is not perfect and I noticed a few mistakes, overall it is good.
It is worth noting that we already have some good text-to-video models from Meta, Google, and Runway. But OpenAI’s Sora stands out with its ability to generate videos as long as 60 seconds.
OpenAI is also working to fix misinformation, hateful content, and bias. The company is working to build a tool that can detect videos generated by Sora.
