Videos are made up of thousands of still images per minute, but what if all you have is a single image and artificial intelligence? That is the question artist and Google developer Damien Henry asked in his latest project, where, outside of the first frame, the entire nearly hour-long video is constructed by an algorithm designed to predict the next frame. The result? Well, it is not going to make you ditch your camera anytime soon, but considering it was generated almost entirely by AI, the bad-Fantasia-remake-meets-view-from-a-road-trip-window is rather impressive.
Henry says the entire video was generated by an algorithm in one try, with no retries and no editing or post-processing. The only thing Henry did was to feed the system the very first frame. From that single image, the platform predicted the next frame and then the one after that over and over again around 100,000 times. It created a rather abstract-looking footage of clouds that is reminiscent of what you see if you strapped a GoPro to your window during an uneventful road trip, with the computer generating a drive-by view of scenes with trees, power lines, and buildings.
The video is far from perfect — some treetops appear to hang in mid-air, for example — but the project demonstrates what AI is capable of, even when it comes to creative projects. Skipping further into the video (you’re probably not going to watch in its entirety unless you need some help falling asleep) you can see that drive-by view change to a rural drive-by with what looks like cornfields to an urban view more populated with buildings.
The video itself wouldn’t be impressive under the typical standards, but the technological feat of asking a computer to generate an hour of footage from a single photo is either incredibly cool or shows just how long the tech has to go, depending on your view and tolerance of the footage’s oddities.
Henry has worked with Google on several projects, according to his website, including Google Cardboard as well as Arts and Culture VR projects.