The way we consume video has changed a lot over the course of the last decade. We now watch videos on our mobile devices from anywhere and because of this, video content comes in a wide variety of formats. Google recognizes this shift and so last week their AI team announced ‘AutoFlip’ an open-source framework for “intelligent video reframing.”
Traditional video for television and desktop viewing are filmed in a landscape format, normally in 16:9, 4:3 or 2:1. If you are like me and are curious why this is, here is a quick lesson. First, our surroundings are landscape – the room around you right now is organized horizontally, thanks to gravity. So if films are created to represent our daily lives, then it makes sense for the images to be captured in a landscape format. Second, human vision is about 200 degrees horizontally and 135 degrees vertically so original 35mm film was created to mimic this field of view. This is why I always recommend that you turn your phone horizontally when you record a video or take a picture.
So why do we need a AutoFlip tool then? Well, our modern smartphones are vertically oriented. Quick videos like Instagram Stories and YouTube Stories are formated vertically so that users don’t have to rotate their devices to view the content. Creating this vertical format from a video that was shot horizontally can be done manually using a static crop in post-production but this process will often result in bad framing. To overcome this issue, creators can adjust the shot frame-by-frame but this process is time-consuming and also error-prone.
Using the AutoFlip tool, a creator can now film a project horizontally and then use AutoFlip to dynamically resize the same video for other platforms. Google accomplishes this using face and image detection models to “find interesting, salient content in the frame.” Google says this machine learning process will work with a variety of content, like animals, text/logos in the frame, or a ball in sports. In the end, ‘AutoFlip’ will give the creator the most flexibility in post-production.
The full write-up for this project is worth a read if you are interested in video science like me. For anyone who is just curious to try out the AutoFlip tool, this isn’t something you can use online or go download at this point. The open-source code might get included in a Google service in the future but for now, you will have to compile the project from Github or wait for developers to include the code in their app. Similar technology can be found in some Adobe video editing software.