Siren Experiment #1

Klingemann describes the work as an experimental neural shader incorporating a pretrained SIREN model, and credits code by Blackle Mori. SIREN, short for sinusoidal representation network, was introduced by Vincent Sitzmann and collaborators in 2020. It represents signals such as images through a neural network that uses sine functions; an image can be encoded in the network’s learned parameters.

Here those parameters are embedded in the shader, which calculates the color of each pixel. Time enters the network’s calculations and the coordinates sampled by the renderer, continually changing the result. The work is therefore executable software rather than a prerecorded animation: the NFT distributes a small running system, with the trained model itself functioning as artistic material.