When we draw, we cannot fully see what is happening on the paper—what someone else perceives. In this sense, we are deluded.

This work, created in August 2017, is based on a custom image-to-image translation model using a conditional GAN system (Isola et al., 2017), better known as pix2pix. The dataset consists of 21,000 pairs of images featuring people who sat for my robot, along with their corresponding drawings (both scanned and simulated). The model was modified for high-speed performance and integrated with a camera feed, allowing for real-time exploration.

This work was featured on fastcompany Design: https://www.fastcompany.com/90139023/watch-this-ai-paint-portraits-alongside-a-real-artist

|Patrick Tresset | 2024 | MP4 | 1280 x 640 | 39'' |