AI Cosmic Dust Animation

This dataset is a stop-motion animation of 1,169 AI-generated images of cosmic dust, made using Midjourney. Each image was produced by uploading either a NASA cosmic dust image or a previous AI generation of one, and prompting the model to reimagine the same object from behind, without naming or describing it. The method functioned simultaneously as an investigation into Midjourney's training data. NASA's cosmic dust images represent a tiny fraction of the imagery circulating online, and the model's generations reflected that scarcity, frequently resolving into eyeballs or human figures; the kinds of imagery that presumably dominate its training data. Although these generations were systematically discarded in favour of those more closely resembling geological matter, the images selected for further AI cosmic dust propagation would sometimes reveal, on reinvestigation, faint ghosts of those human forms which would then be reinforced in subsequent regenerations. 

The animation shares the same 1:1 aspect ratio and stop-motion structure as the companion dataset NASA Cosmic Dust Animation [10.5281/zenodo.18915910], of which it forms a chronological continuation. Where that dataset tracks the evolution of scientific imaging technology across thirty years of NASA documentation, this dataset speculatively extends that trajectory into the era of AI image generation. The dust grain remains the constant; what changes is the apparatus imagining it.

The dataset was produced as part of the practice-led research project We Are Supernova (MSCA Fellowship, Goldsmiths University of London) and functions as source material for the two-channel video installation SPECKS.