[Sora功能介绍] Detailed explanation of Sora's video generation features: video parameters, text-to-video, image-to-video, and the characteristics of video-to-video.

Sora's main functions fall into five aspects. First are the basic video generation methods, including text-to-video, image-to-video, and video-to-video. In addition, there is its outstanding duration—at least 1 minute per generation—as well as its outstanding 1080P resolution. Generating a 1-minute video takes about 30 minutes. According to BBC's interview with OpenAI's female president, Sora is expected to launch around November this year.


[baike]Video Generation Methods[/baike]
(1) Text-to-video: you input a piece of text, and a video can be generated according to the meaning of your text.
(2) Image-to-video: Sora can generate videos from provided images and prompts as input. On the left is the image, and on the right is the video; it makes the image move, directly replacing did heygen digital humans




(3) Video-to-video: a simple video is optimized into a higher-quality video, such as changing a barren environment into a lush jungle



[baike]Video Duration[/baike]Each video generated is at least 1 minute long. According to the March BBC interview with OpenAI's female president's news report, generating a 1-minute video takes about 30 minutes and requires enormous computing power.


[baike]Video Resolution[/baike]Sora can sample widescreen 1920x1080p videos, vertical 1080x1920 videos, and everything in between. This allows Sora to directly create content for different devices in their native aspect ratios. It also enables us to quickly prototype content at a lower size before generating at full resolution - all using the same model.





[baike]Multi-shot Perspectives[/baike]We found that video models exhibit many interesting emergent capabilities when trained at scale. These capabilities enable Sora to simulate certain aspects of people, animals, and environments in the real world. The emergence of these properties has no explicit inductive bias toward 3D, objects, etc.—they are purely a scale phenomenon.
3D consistency. Sora can generate videos with dynamic camera motion. As the camera moves and rotates, people and scene elements move consistently in three-dimensional space.



[baike]Video Extension[/baike]Sora can also extend videos forward or backward in time. Below are four videos, all of which extend backward from a generated video clip. Therefore, all four videos have different beginnings, but all four videos end the same way.