Google is building its cornerstone artificial intelligence services into generating images from text. As the name suggests, Image Video creates high-quality videos from just a few words.
A week after the launch of Meta’s Make-a-Video, Google’s turn is to launch an artificial intelligence-based video generator. By entering a few words, the service will create a short video, respecting style restrictions if necessary (for example, Van Gogh’s paintings, etc.). Image Video can create 3D objects while preserving their structure or create text according to many animation styles.
Not open to the public
The purpose of the video collection models created by Google is ” Significantly reduces the difficulties of producing high-quality content “. High quality, AI creates videos lasting 5.3 seconds at 1,280 x 768, at a rate of 24 frames / second. After entering text, Imagen Video creates the first basic sequence of 16 images at 24 x 48, animated at 3 FPS. 6 additional broadcasts A “cascade” structure of 2 leads to the finished video.
Google provides many examples ranging from simple (” Ice cream drips onto the cone “) is more complicated (” Fly through an intense battle between pirate ships on a stormy sea “). The proposed videos still have graphic artifacts, but the fluidity and detail are very impressive and all this bodes well for the future.
However, Imagen Video will not be offered as “self-service,” at least not for a while. Google wants to fix problems with explicit, hateful or dangerous content first. The service models are based on the LAION-400M database, which contains 14 million text-video pairs and 60 million text-image pairs. If they are trained to filter this type of content, the solution is still not available to everyone.