The neural network has learned to encrypt text messages in pictures

Categories: Technology |

The capabilities of artificial intelligence are growing at a tremendous pace. Could anyone have imagined at the dawn of the development of neural networks that they would be able to create photorealistic images and full-fledged videos? Not long ago, another talent of this technology was revealed - the Glif neural network learned to write text into pictures.

The neural network has learned to encrypt text messages in pictures

The new feature of the Glif neural network has caused a whole wave of experiments on the network. Users generate pictures with cats, raccoons, plants and even household appliances, encrypting messages in them. Artificial intelligence, following the text description given to it, creates images in which text is inscribed in different ways.

The neural network has learned to encrypt text messages in pictures

Letters are formed from the objects themselves, from shadows, folds, bends. The text cannot always be read immediately - some images look like riddle pictures and you need to look at them carefully before the essence becomes clear. For example, in this cute photo with red cats the phrase “We have paws” is encrypted. To see it, you need to move away from the monitor or squint.

The neural network has learned to encrypt text messages in pictures

Anyone can experiment with a neural network. To do this, you need to follow the link, select the Controlnet Any Word function and fill out three fields. The first contains a text description of the image itself in English. In the second, you need to enter a word or phrase in any language that you want to encrypt. In the third field, indicate the size of the letters from 100 to 170. Experienced users of the service advise choosing the “golden mean” - 130.

The neural network has learned to encrypt text messages in pictures

The process is fun and very creative. Users post hundreds of images online, including real masterpieces. For example, in boats against the background of a sunset, you can hide the “Dream” inscription.

The neural network has learned to encrypt text messages in pictures

Experiments with Glif are not always successful. In some cases, the neural network “hides” the text too well, making it unreadable, or cuts off words.

Unfortunately, neural networks still make mistakes, but their mistakes sometimes cause joy with their unpredictability and absurdity.

     

source