In a web world dominated by videos, it is necessary to help create them, even if we only have a reference image.
That’s what Google has been working on, enabling the DeepMind neural network to create short videos from a single image frame.
This AI model, called “Transframer,” generates text based on partial prompts, and is now capable of generating 30-second videos from a single image.
The AI guesses the surroundings of the images using context images, so it can show what the outline of a piece of furniture looks like without actually seeing it. What it does is imagine a real object from any angle.
It’s easy to imagine how he does it. If I give a program thousands of images of a chair from all possible angles, and then send it a photo of a chair from the front, the program can imagine the rest thanks to the previous training.
The artificial depth perception and perspective are noticeable in the demonstration, which helps to imagine how video games can improve with something like this, not only the creation of videos for social networks.
Transframer is a general-purpose generative framework that can handle many image and video tasks in a probabilistic setting. New work shows it excels in video prediction and view synthesis, and can generate 30s videos from a single image: https://t.co/wX3nrrYEEa 1/ pic.twitter.com/gQk6f9nZyg
— DeepMind (@DeepMind) August 15, 2022
If we can create realistic images with DALL-E, and from it we can create videos, we are one step away from automatically creating videos from a single text message.
