In 2019, I became aware of Microsoft's Attentional Generative Adversarial Network (AttnGAN) – an early generative image model that produced wild, surreal images that interested me. I built a quick experiment in 2020 to ask: what does it mean for a machine to create art, and what new forms or value might it bring?(Video)

In 2025, generative AI is everywhere. But the core question of the experiment still matters. I've updated the experiment to use OpenAI's image API. This version is limited to low-res images with a prompt to replicate AttnGAN using DALL-E 2. It's slower, less surreal, and has lost the original charm, but can machine-made art still surprise us?