Every song has a shape. When you upload audio, the engine extracts an amplitude envelope and a frequency spectrum, producing a numeric fingerprint unique to your track - energy, dynamic range, bass-to-treble ratio, and peak positions.
That fingerprint drives the 3D geometry: bass shapes broad distortions, midrange adds folds, treble carves fine detail. A per-song hash seed sets unique phase offsets and twist rates, so every song produces a fundamentally different silhouette.
The result is captured at max resolution and sent to the render server for DPI upscaling and text compositing.