Visual Art Python Machine Learning Video Generation Prompt Engineering

DreamGAN

GitHub

Authors: Conor Hayes

In late 2021, I became curious about the GAN-generated images I had seen bouncing around online, and wondered if the same technique could produce narrative videos instead.

After a night or two spent hacking around, I managed to adjust VQGAN-CLIP (Crowson et al) to produce narrative videos based on a series of prompts.

Inspired by the dream journal I was keeping at the time, I called it DreamGAN, and used it to make a number of renderings of my and others' dreams, as well as video accompaniments for my songs.

Videos

Visual accompaniment to my song Fall of Rome.

My first attempts were visualization of actual dreams I wrote down in a dream journal.

A brief explanatory video for a DreamGAN video I made during a livestream, as an accompaniment to my song Marry Another Man.

Back to Projects