StyleGAN2

This is where it all started. I accidentally stumbled upon AI DMT entity generation via training StyleGAN2 on a dataset of all of the visionary art on the internet that I could possibly find. The results were shocking - the model began generating imagery that looked remarkably similar to DMT hallucinations.

While the resolution was limited compared to modern models, StyleGAN2 produced some of the most authentically psychedelic imagery I had seen from any AI system at the time. The entities that emerged from the latent space had an organic, flowing quality that later diffusion models sometimes struggle to replicate.

This experiment laid the groundwork for everything that followed, from StyleGAN3 to the Midjourney prompts I eventually developed. The video below showcases the interpolation capabilities of the model, smoothly transitioning through its learned latent space: