Conceived as a reinterpretation of the first surrealist literary work, the project brings together the concepts of surrealism and artificial intelligence. To achieve this, it uses André Breton and Philippe Soupault's work ‘The Magnetic Fields’ (1920) as an object of study and content exploration, deconstructing and reorganising it through machine learning mechanisms. Namely, a Recurrent Neutral Network, or RNN, which generated the text presented in the book using Breton's text as input, and the VQGAN+CLIP mechanism, which generated the visual contributions to the work, culminating in a book composed mainly of text and generated images.
Magnetic Fields Reimagined explores two crucial and inverse ideas. Initially, it explores the relationship between man and machine through the automatic processes intrinsic to the surrealist movement, in this case, automatic writing. Conversely, the machine's approximation to man is translated by a text generated based on the processing of datasets and inputs, with the aim of approaching - or distancing itself from - surrealist writing.

This project and Breton's original work are similar in the way they present the final product, gradually becoming a blur. They differ in that in the original work, Breton accesses the subconscious through a method of automatic writing, which results in several loose phrases that are incoherent when read together. In contrast, the project makes use of the automatic writing processes already existing in the machine to obtain virtually identical results.

The project thus culminates in a rewriting of the surrealist work, resulting in a story told by contemporary automatic processes.

This work was developed as part of the KYRT project under the theme ‘Missing Stories,’ marking the completion of the Undergraduate in Design — Communication.