DEMON: Diffusion Engine for Musical Orchestrated Noise
TL;DR AI
2 min readKey summary
Researchers introduced DEMON, a streaming diffusion system for real-time music generation with live, responsive control over denoising.
It uses per-slot schedules, shared mutable state, per-frame source blending, and windowed VAE decoding to keep latency low.
The system can separate control inputs by how quickly they affect the output, making model parameters feel like performance controls.
DEMON reaches high generation rates on a consumer GPU, showing diffusion-based audio synthesis can work as an interactive live instrument.
