AI Artists Logo.png

Robbie Barrat

Robbie Barrat is a young AI prodigy whose projects push the boundaries of neural networks and the traditional art world.

Robbie Barrat - Nudes.jpg

Selected AI artworks:

Infinite Skulls

The L'Avant Galerie Vossen emailed Robbie after seeing his AI-generated nude portraits and asked if he’d be willing to fly out to Paris to work with the famous French painter Ronan Barrot. Robbie agreed and flew out last July to meet with Ronan, and the two have been working together ever since.

Robbie Barrat - Infinite Skulls.jpg

Their exhibition “BARRAT/BARROT: Infinite Skulls – An unprecedented encounter between a painter and an artist researcher in artificial intelligence” featured an “infinite” number of skulls. 450 paintings of skulls which Ronan Barrot painted over the last few years were digitally scanned so that Robbie Barrat could train a neural network to create new images of skulls from these works. The resulting collaboration – or confrontation – between its digital and analog creators marked the birth of an entirely new realm of creative partnership.

AI Generated Nude Portraits

Using an implementation of Progressive Growing of GANs and a corpus of thousands of nude portraits scraped primarily from WikiArt, a neural network was trained to create nude portraits. The resulting portraits are not realistic. The machine failed to learn all of the proper attributes found in nude portraits and instead has fallen into a local minima where it generates surreal blobs of flesh. Is this how machines see people?

AI Generated Landscapes

Using an implementation of Progressive Growing of GANs and a corpus of tens of thousands of landscape oil paintings scraped from WikiArt, a neural network was trained to create new landscape paintings. The resulting landscape paintings are fairly realistic, while also being strangely surreal. One can tell that whatever painted these has little idea of how the physical world works; often painting trees with floating branches, and multiple trees having joined trunks. The longer the network is trained, the more it tends to producing more murky and dark paintings. These paintings can be seen towards the bottom of this page. Towards the end of training, it started to paint more and more paintings with unrecognizable objects and forms included. The probable answer for this is some form of mode collapse. Nevertheless, the results are interesting.

Neural Network as a Video Game Shader

Using CycleGAN, a corpus of minecraft gameplay frames was collected through Barrat’s own gameplay, and a corpus of landscape oil paintings he scraped from WikiArt - he was trained the network to turn minecraft into a "playable oil painting".

Neural Network Balenciaga

Using a corpus of Balenciaga runway shows, catalogues, and campaigns, a Pix2PixHD network was trained to reconstruct Balenciaga outfits from Densepose silhouettes. The results are outfits which are novel but at the same time heavily inspired by Balenciaga's past few years under Demna Gvasalia. The network lacks any contextual awareness of the non-visual functions of clothing (e.g. why people carry bags, whether or not bags are separate from pants, why people prefer symmetrical outfits) - and in turn produces more strange outfits that completely disregard these functions. Note that this is an ongoing project.

AI Generated Sculptures

3D DCGAN inspired GAN trained on 32x32x32 voxelizations of Thingi10k - a corpus of 10,000 3D printable objects and sculptures - as a result, the generated sculptures are almost always 3D-printable, but usually do not have any real 'meaning', and are just abstract sorts of shapes. Barrat originally began this project as an attempt at generative architecture; but lack of an appropriate dataset held him back from that. He plans to try to get something usable from the Google sketchup 3d workshop.

Rapping Neural Network

The first AI+Art project Barrat completed, during his senior year of high school. He wrote and taught a Recurrent Neural Network to rap like Kanye West. This project was written about in a profile by Quartz. NVIDIA noticed this project and invited Barrat to work with them on AI directly out of high school. The idea to build an AI that can create its own bars came after Barrat and his classmates were in a heated and challenging argument as to whether AI could perform "artistic" or "human" better tasks than Humans themselves. After a week of working profusely on the project, he completed RAPBOT after only a week. This is the first notable project that he's worked on using AI in this manner. He trained the AI on using 6,000 Kanye West lines and can generate speech that can rap the words assembled by the AI, with natural pauses and a cadence similar to Kanye.

Learn more about Robbie Barrat: