- Presented by
HERE Opening event
Saturday 25th November 2023
4 - 5pm: Jan Robert Leegte with Pita Arreola (Digital Art Curator, V&A)-
5 - 6pm: Haroon Mirza with Kostas Kostas Stasinopoulos (Curator, Serpentine)
6 - 10pm: DJ set by AHMD (Artists Eddie Peake, Prem Sahib and George Henry Longly)
Free & Open to Public
25 - 26 Nov 2023
9 Cork St
London W1S 3LL
Elman Masimov’s alignDRAW, presented by Fellowship, is widely credited as the first text-to-image model, developed in 2015. With text prompts revolutionizing the guidance of AI’s creative process, these works represent the beginning of a new paradigm in image making.
In 2015 Elman Mansimov developed a new idea to create images via a computer model called alignDRAW. By this time, neural networks could already label images, i.e. convert them to descriptions of their content. That same year, a group of researchers introduced a model called DRAW to generate coherent images using a series of numbers.
Mansimov combined the two ideas and showed that text descriptions can be converted by a deep network into images. He and his collaborators published a paper “Generating Images from Captions with Attention” which used as illustrations images produced by their new alignDRAW model. These 168 images were generated using prompts describing non-existent scenes such as “a herd of elephants flying in the blue skies.” At that time, Mansimov also saved another 2,057 images created by the same alignDRAW model.
The alignDRAW project marks the beginning of a new paradigm in image creation. Photography was already two hundred years old at the time, while computer graphics were fifty-five years old. The first 3D wireframe computer graphics appeared in 1960. Ivan Sutherland, an MIT student, created Sketchpad, the first interactive CAD system, in 1961-1962. Sketchpad gives visual commands to the computer via the interactive screen, and the computer responds by drawing lines.
However, generative media has brought yet another surprising method of creating visuals with computers. You can now describe a desired image using natural language rather than creating it by hand, using mechanical or software devices, or capturing it with a lens. You are able to generate an infinite number of visual universes simply by naming them, which is extremely amazing. While we can debate whether AI is “creative” or not, in my perspective, it is already technically more proficient than most art students and adult artists.
When we look at the initial low-resolution and hazy images from nearly two centuries ago, we see the whole future potential of photography, which eventually became the dominating imaging and communication technology of our time. And when I look at comparable low-resolution alignDRAW pictures, I see a similar promise for a new major visual method that could very soon become as essential as lens photography was in the last two hundred years. This is why these images are valuable, magical, and worth collecting.
Text by Lev Manovich