Thursday, 7 January 2021


Via Waxy, we make the acquaintance of a namesake (a portmanteau of the Pixar character and Salvador Dalรญ) neural network that generates, using Open AI, images from captions. It’s still too brittle, its minders say, for free-text (see also) but one can play Mad-Libs with a certain string of prompts to get an idea of its virtuosity and capabilities. 

This first array of images is in response to the cutline a triangular, yellow manhole cover. The second, poetically, is a fox—made of voxels—sitting in a field. The network even demonstrates learning in geographical facts, fashion and dating styles and technology, though some seem better informed than others.