Google's AI can reproduce images from text

The research team at Google has found a way for their AI to reproduce images from text. The more detailed the text, the more accurately reproduced the image.

Do you still remember DeepMind, the company that created artificial intelligence called AlphaGo and beat the Chinese goalkeeper number one Ke Jie? Immediately after that historic victory, AlphaGo decided to " retire " because no one else was strong enough to rival it. After a period of "silent" , DeepMind has come back with a new breakthrough in the field of artificial intelligence: their AI is capable of reproducing true images based on data in the form of Text that people enter.

Images re-created by Google AI.

According to Futurism, this is one of the latest breakthroughs in leveraging the capabilities of AI into image processing. In February, Google Brain researchers discovered a way to "improve" image quality like the ones you would find in science fiction films such as Blade Runner or Crime Investigative Films. Like the CSI series (Crime Scene Investigation). Using PixelCNN, the computer will be able to turn low resolution photos into high resolution images with impressive accuracy.

Now, that technology will be used to turn text into images . The researchers found that the more detailed the input data, the more accurate the resulting image would be. For example, if the input is "a yellow bird with a black head, eyes and an orange beak" will bring the image exactly the same. The algorithm that DeepMind uses can be selective from a huge collection of images and distinguish between concepts like birds and humans, and from there produce results according to what it "learns". .