Google continues to show us the capabilities of its artificial intelligence. This time, it has announced the creation of a new tool for YouTube Stories. This for now can only be used by content creators and consists of being able to change the background of published stories to make it seem like they in other places.
For example, we will probably see our favorite youtuber make a story as if it were at night without using editing softwares. Normally it would require not so simple techniques with ‘rotoscopia’ and ‘chroma’ (the typical green screen that we see in the movies).
How does it work?
Google, which describes this technology as “video segmentation,” explained that one of the goals was to make the animations look smooth and natural with at least 30 frames per second. This in order to stand out against other platforms with similar tools.
In this way the first thing they did was to train their artificial intelligence by teaching them thousands of photos of human faces in different poses. In this way, the system could differentiate the face it was recording against the background it had to remove.
Also, as the objective was to work with moving images, they started working with videos in which there were obstacles in motion such as cars or other people. It worked from the first frame so that new objects that appear in front of the camera will not generate disturbances in the simulation. In addition, smoothing was added so that the changes of movement of the person being recorded did not look too rough.
Google reported that the result was remarkably good. The tests showed a fluency of up to 100 frames per second (Fps) with an iPhone 7, and up to 40 Fps with a Pixel 2.
For now this option can only be used by a limited number of content creators and it is not yet known when it will be available to the rest of the public.