Stability AI's Demon brings real-time music diffusion to local GPUs

Stability AI has released Demon, an open-source engine for real-time music generation built around diffusion models. Unlike batch-oriented systems that render a full track and hand it back when finished, Demon is designed to produce audio continuously, streaming sound as it computes rather than waiting for a complete render.

The headline figure is throughput: the project says it runs at 25Hz on a single local GPU. That update rate is what makes interactive use plausible, since it keeps the gap between input and output short enough to feel responsive rather than like a slow render job.

Why it matters

Most generative music tools today lean on remote servers and fixed-length outputs. A model that runs locally and in real time opens different doors:

• Latency: on-device inference avoids round trips to a cloud API.
• Control: continuous generation lends itself to live, interactive performance and experimentation.
• Access: an open-source release lets developers inspect, modify, and build on the system directly.

Demon arrives as an initial release, and key details — licensing terms, model size, and hardware requirements — remain light in the public record. Stability AI has been one of the more active companies pushing open audio and image generation, and a real-time music engine fits that pattern. For developers curious about interactive sound, the project page is the place to start.