OpenAI launches new video generation model Sora with cutting-edge video preview

Leading AI company OpenAI on Friday launched Sora, an AI video generation model that can create hyper-realistic videos. OpenAI has released a preview of the video it created. This AI model comes in different resolutions and different layouts, and the resolution can also be adjusted.

By training this AI, it can take raw video as input and output latent representations that are both temporally and spatially compressed. Sora is trained and later creates videos within this compressed latent space. It also trains an associated decoder model that maps latent values ​​back to pixel space.

By Sora's ability to create videos. By specifying a period of time Video resolution and various ratios as desired. Sora's sampling flexibility It can sample 1920x1080p widescreen video, 1080x1920 vertical video and everything in between, helping Sora create content for a variety of devices. directly in the original aspect ratio. It also allows us to quickly prototype content at a lower scale before building it at full resolution. And all of this is in the same model.

Elements within the video can be updated to the same aspect ratio. Improved composition and improved frame rate management. Applying the subtitle transition techniques introduced in DALL·E 3 to the video, we first train a highly descriptive subtitle model. This model is then used to generate text captions for all videos in the training set. And can create animations from DALL·E 3, able to extend the duration of the video from the original video. Edit photos and videos from messages And believe that the video between the two videos doesn't flow smoothly either. In addition, Sora can create 3D models to accompany videos. and can display more details along with simulating the real world as well