Dans le cadre du colloque Les arts du sommeil à Paris les 12–14 octobre, je co-présente une conférence avec Albertine Thunier intitulée : Création automatique et restitution des rêves surréalistes : le sommeil (les rêveries) de Desnos (re)vu par TikTok et Dream AI.

//

As part of a symposium in Paris on “the arts of sleep”, I’m giving a presentation with Albertine Thunier that can be loosely translated as “Automatic creation and the restitution of surrealist dreams: Desnos’s sleep and dreams revisited by TikTok and Dream AI.”

ABSTRACT

In an effort to restore the work of the Surrealists into popular culture, and in particular that of Robert Desnos, we reimagine his hypnotic sleep experiments using the protocols of digital automation. We understand applications like Wombo.art and TikTok as sites of memetic culture, and as technologies of cultural glanage that recuperate, rework, reframe, remediate, and recirculate cultural works and logics of different periods. In this sense, we align ourselves with Nova and Kaplan (2016), who present internet memes as matrices for generating cultural innovations. This presentation will draw on our creative experiments, and place Desnos’ writing and drawings amidst the vernacular poetics of today’s digital culture.

In 2021, the Canadian software company Wombo AI launched Dream AI, an application that generates “AI-powered painting in seconds” (https://www.wombo.art/) based on two elements selected by participants: a keyword (what they call a prompt), and one of twenty visual styles (such as “fantasy” or “etching”). Most of these resulting images have a “dreamlike” quality, whether the styles are explicitly surrealistic (e.g., “psychedelic” or “Dalí”) or not (e.g., “dynamic” or “no style”). To understand these images as the product of knowledge or logic, the key is to know what dataset was used to train the artificial intelligence, and what algorithms were put into place as protocols of image generation (Audry 2022). As an example of generative art, then, the Dream AI neural network delivers images that look similar to other images it knows, working in one sense in a “cultural bubble” or a loop of continuous recursivity. The automatic image is both generic and impossible, an illogical result of programming and chance. If memes are sometimes understood as “tapestries of conversation” (Milner 2017), the “paintings” of Wombo.art could be thought of as surreal, dreamlike tapestries of hybridisation.

In order to explore the Dream AI program as part of a larger inquiry into sleep-influenced automated modes of creativity, we input prompts that draw on the language of sleep: we start with general related terms such as “sleep” or “dream” but then move into Desnos’ work such as titles and fragments from his automatic writing as found in the ALMé digital archives. Creating a response between text and image, we thus also create a feedback loop between human and machine automation, with the words of the hypnotic Desnos meeting the images of algorithmic Wombo.

We continue the loop and the remediations by moving our text-generated images onto TikTok, where they become uniquely interactive and dynamic, functioning as decorative backdrops for the mise-en-scène of readings and recitations of Desnos’ poems. Using other automated technologies such as text-to-voice and an AI text generator, as well as our own recorded attempts to read the text while in the threshold stage between wake and sleep, we create a spoken-image digital Desnos traversing time and space not through dream but through Dream and TikTok.