In April, the amazing result achieved by DALLE-2, an artificial intelligence (AI) capable of creating images based on descriptions of objects, subjects and scenarios, was unveiled.
DALL.E mini makes it possible through a web page to enter a description and obtain in the form of an illustration the “artistic interpretation” made by an artificial intelligence
Initially this AI was not open to everyone and was only accessible to researchers or developers interested in investigating or exploring the possibilities of AI. Fortunately, it is now sufficient to access the DALL.E mini website to have the possibility of using it.
All you have to do is propose a description (in English) as detailed as possible of what you want to see reflected in a synthetically generated image, and obtain up to nine results with different artistic interpretations.
The web page takes a few seconds to provide the results, and may even display some message informing about the saturation of the service and asking to retry later. The generation of the image itself also takes a few seconds (sometimes almost a minute or even more).
Some of the examples, as a sample of DALL.E mini’s work, are the following:
-A rabbit wise man reading a book at the foot of a tree
-A surfer space surfer in front of an exploding star
-A river in the middle of the sunny countryside with some trees in the background and a wolf
–A baby sleeping on a light blue sofa
-Street of cobblestones between menacing old buildings
As can be seen, among the various “artistic interpretations” of DALL.E mini’s AI there are more fortunate and successful examples than others, noting a greater success in the representation of some natural scenarios and some rather disturbing samples in the interpretation of a seemingly innocent baby sleeping on a sofa, with almost terrifying results.
Although the algorithm of DALL.E mini is less powerful than that of DALLE-2 its performance is more than correct, admitting from mentions of famous people who would appear in the final image to “style hints”. This would be the result of describing “Donald Trump dancing as if he had been drawn by The Simpsons.”
The basis of the operation of this AI is large-scale capture of images from the Internet to, based on their classification and labeling, reinterpret the result that should appear in an image combining the elements mentioned in the proposed description.
