Latent Jungle (2019).
A neuro-photographic collage playing with the typography of Amazonian jungle patterns, shapes and textures and their reinterpretation using generative neural networks and the material properties of Super8 analog film.
A latent image is an imprint of reality either on the photosensitive material, or on the trainable material of a neural network model. It is evoked, either by chemical processes in the darkroom (enlarging a negative), or in an abstract mathematic process of model inference (exploring a latent space). A newly created image sequence is, however, without time as we understand it in moving pictures. Instead, it has been replaced by similarity, by the compression of representation.
Recording on a Super 8 analog film was made manually to create a curated dataset, which was later scanned in high resolution. Progressively Growing GAN was trained to make use of the amount of details and post-processed using a super-resolution SFT GAN. Finally a collage has been assembled over a real photographic paper print made in a darkroom.
The subject of the recorded film, the study of natural processes and patterns, is inspired by the early experimental and scientific films made by Jean Painlevé and Jan Calábek (1955, “Pohyby rostlin” (Plant Movement)). This piece references aspects of the decomposition of motion such as that which can be seen in the photographs by Eadweard Muybridge. Extracting a typography of shapes to describe a particular system follows the work of Hilla and Bernd Bechers and the movement of New Topographics.
The work is presented as a selection of frames from the generative models and made as a collage with a sequence of a physical slides of film.