Create images from text captions. This is the “super power” of the neural network DALL · E trained by OpenAI, which uses a set of data from pairs of texts and images, and which can test on a website.
On the OpenAI blog you can see examples of images generated by the system and test its capabilities in real time, since it is possible to exchange a few words of each sentence already available. You can, for example, choose to see a watch of various colors and shapes.
Click on the images to see the “super powers” of DALL · E
From cats to an avocado-shaped chair, the images show the system’s ability to “draw” through a caption.
The name of the system is inspired by the Disney film WALL-E and a suitcase used by artist Salvador Dali. DALL · E integrates a varied set of resources, including a functionality capable of creating animal and object shapes. For this, it combines unrelated concepts in several possible ways, taking advantage of the text and applying transformations to existing images.