Generative AI · 2024
Unstable Diffusion: An AI Art Installation
Type a prompt into a vintage CRT monitor. Watch a giant screen turn your words into art. Live. In front of everyone.
→A physical prompting station feeding a live Stable Diffusion pipeline built in ComfyUI.
Highlights
- ✦Vintage CRT monitor as a "prompting machine"
- ✦Live image generation on a big screen
- ✦Custom tooling to save prompts, workflows & outputs
The idea
Show people how weird and wonderful generative AI can be.
Not in a browser tab. In a room, with a crowd.
What I built
A high-end CRT monitor became the prompting machine. Visitors typed whatever came to mind.
Behind it: a ComfyUI pipeline running Stable Diffusion, with custom tooling to queue prompts, save every prompt and workflow next to its image, and process media with FFmpeg.
The big screen showed the results. Live.
The lesson
An installation has to keep running when someone types something very unexpected. 😄
Turns out that’s the same skill set as an enterprise generative media pipeline: orchestration, queueing, prompt handling, safety, and an interface normal humans enjoy.
Just with more neon.
