_Autumn _and _Alcibiades on his knees before his mistress _are two of the 200 works in Rêveries, a NFT series made by Orkhan Mammadov with generative AI. The series exemplifies the cutting-edge explorations of AI’s artistic potential that have appeared on the blockchain in recent years.
The project focuses on convolutional neural networks (CNNs), a type of deep-learning model designed for image classification and object recognition. Mammadov was inspired to explore the generative possibilities of these neural networks in 2016 after reading a technical paper by a group of researchers entitled “Texture Synthesis Using Convolutional Neural Networks.” Another influence was Google’s DeepDream computer vision program, which used CNNs to create hallucinogenic versions of source images.
Each of the works in Rêveries, which Mammadov describes as “generative neural paintings,”_ _is based on a painting of a biblical or mythological scene. Alcibiades on his knees before his mistress, for instance, takes as its source an 18th-century canvas of the same title by the French painter Louis-Jean-François Lagrenée. The CNN transforms the input data into an abstracted texture. There is also an interactive element, powered by GLSL shaders, which allows users to manipulate the texture in real time.
The 200 outputs selected by the artist were minted as ERC-721 tokens and released on Artist Rooms, an initiative from the digital art platform Verse.