After text-to-image, text-to-video, and AI-generated video in general, is definitely the next big thing. Shortly after introducing its own text-to-video AI, Google introduced another AI-powered video tool -and it’s pretty darn awesome!
Google’s new program InfiniteNature-Zero lets you “fly into” a landscape scene, similar to what we often see in drone landscape videos. But you don’t need a drone – all you need is a single landscape image and AI does all the magic, turning it into a video and giving you a three-dimensional feel.
The researchers presented InfiniteNature-Zero in a paper which explains a bit more about technology. As I mentioned, the technology takes a single landscape photo and turns it into a video that lets you fly like an eagle, to the seeea. But what’s super-special about it is that the algorithm was not trained on video. No my friends – it was trained on single images.
You may think that the researchers at least needed to train the algorithm on sets of photos taken from different angles. Nope – each scene is recreated from a single JPG photo, taken at just one angle. The algorithm is capable of recreating the scene beyond each frame, expanding the scene further, and creating a video.
“To achieve this, we propose a novel self-supervised view generation training paradigm, where we sample and rendering virtual camera trajectories, including cyclic ones, allowing our model to learn stable view generation from a collection of single views” the researchers write. “At test time, despite never seeing a video during training, our approach can take a single image and generate long camera trajectories comprised of hundreds of new views with realistic and diverse content.”
What you get is a flyover video that looks like a drone shot with some glitches. Considering that it’s still at the early stage, I think it’s pretty cool:
Sure, this technology still can’t replace drone videos (or maybe…?), and it will take some time to refine it. But considering how fast it develops and improves, I wouldn’t be surprised if, in a few years, we’ll get to turn our landscape photos into believable videos. And even though there are both good and bad sides to these technologies – I’m excited and curious to see where their development will take us.
[via PetaPixel]
FIND THIS INTERESTING? SHARE IT WITH YOUR FRIENDS!