Infinite Images ∞
Holly Herndon and Mat Dryhurst.
Infinite Images ∞ 1, 2022
“We can’t emphasize enough how surreal the experience was. We were pulling 14 hour sessions daily trying to get as much as possible out of these tools just to understand them, stopping to eat and sleep. It felt like we were holding a secret, and that also came with a sense of responsibility to think about what it all means in the bigger picture."
Holly Herndon & Mat Dryhurst
“In 2021 we were invited by OpenAI to work with the DALL-E 1 research model as part of a commission. “Infinite Images” are the product of these experiments, large images meticulously constructed piece by piece to explore the models capacity for coherence.
This context hints at why themes of reflection in water and glass, horizon lines and mountain ranges appear so frequently, as the works were constructed by playing an exquisite corpse game with the model. A single initial generation could now be infinitely expanded upon in any direction. To our knowledge these are the only artworks produced with the original DALL-E model. They represent a transitional phase from the more abstract generations of the models that preceded it, and the high fidelity models we are used to today.”
- Holly Herndon and Mat Dryhurst
Arriving onto the AI scene by way of electronic music and experimentation with voice and sound, Holly Herndon and Mat Dryhurst have since broadened their practice to include image, sculpture and wider community projects that tackle the current ethical issues with the emerging technologies of AI. With their work spanning all corners of the ecosystem – from the artistic to the societal, it is not a surprise that as OpenAI began to develop the new DALL-E tools, the first of the contemporary text-to-image models, it was Herndon and Dryhurst who were invited to test them.
Text-to-image models such as DALL-E, Stable Diffusion and Midjourney are now the default choice for producing high quality, realistic images based on a prompt of a few words. However, this concept of communicating with neural networks through human language to produce images was still reasonably new in 2021. First conceived by Elman Mansimov in his 2015 alignDRAW paper, this idea took the research community a number of years to develop into a model that could achieve the desired performance and realism. Therefore, the arrival of DALL-E on the AI art scene caused much excitement, particularly as the process of generating images was now made instant, instead of requiring days or weeks to train a GAN model. This opened up a far shorter feedback loop between the human and the AI system, enabling artists such as Herndon and Dryhurst to improvise and jam with the machine.
As part of their experimentation with the tool, Herndon and Dryhurst produced a series of works called Infinite Images, which reflect the duo’s philosophy of the media world of the future – anyone will be able to create, adapt, and iterate on the works of others. This series are the first images made with DALLE-1, representing a special moment when it was a mere handful of people testing the too, including the two artists and a few engineers from OpenAI. Now, the tool has millions of fans.
The series begins with a large-scale canvas, a patchwork quilt of images that are presented as if they are screens in an exhibition, extending as a queue towards infinity on the right of the original idyllic mountain landscape. Upon a closer look, we realize that not everything is as it seems – out of the two young boys in the foreground one is actually composed of two round, skin-colored shapes - this depiction of the human form signals that the system is not perfect, yet. The artists’ use of DALL-E 1’s outpainting capabilities allows each image to be extended, showcasing coherence of style and subject matter and applying imagination in a much broader sense based on the relationships between images themselves as opposed to associations between prompt and image.
Despite the impressive realism offered by these tools, Herndon and Dryhurst push this emerging technology to the extent of its new limits, probing the spectrum of its generative abilities: moving beyond photorealism to make use both of the creativity of the technology and its potential to draw on a multitude of artistic styles. The Infinite Images references a variety of movements and techniques - from the surrealism of imaginary creatures that could have walked out of a Max Ernst canvas to the strong colors of Fauvism or the grainy texture of gouache paint. After all, the notion of infinite images should encompass the breadth of human artistic traditions - both past and future, the latter more increasingly now in the form of co-creation with the machine.
Artworks