Realtime Audio is here: Introducing DEMON

Ryan Fosdick demos DEMON, our real-time music engine we just open-sourced, and answers your questions live. WHERE TO WATCH: On Wednesday, May 27, we open-sourced DEMON, a real-time music engine built on top of ACE-Step 1.5. This event is a live walkthrough of what it does and how it works. Most AI music tools break your creative flow with even a few seconds of latency between prompt and output. DEMON responds in about 100 milliseconds - fast enough to actually play like a musical instrument. That changes what you can do with it. For DJs and producers, the engine turns a source loop into raw material. Feed it audio you already have, then sweep through styles, intensities, and structural transformations live. You can keep a verse close to the original while the drop fully transforms, inside a single generation. Music style LoRAs swap in real time. MIDI controllers map to any knob. Ryan will demo this end-to-end. For developers and researchers, DEMON is an open engine built for continuous frame-level control of music diffusion. A ring-buffer pipeline gives you six per-frame control curves at 25 Hz, with parameter changes propagating across in-flight generations in a single tick. TensorRT acceleration runs the whole thing at 12.3 generations per second of 60-second audio on a single RTX 5090. The code is on GitHub (we'd appreciate your stars there if you are finding it useful!), and the project page has all the details. What we'll cover: - Live walkthrough of DEMON, from source upload to live playing - The remixing story: per-frame source blending, LoRA hot-swap, MIDI - Under the hood: streaming ring buffer, control curves, shared mutable state, TensorRT - What's next, including the VST in development - Open Q&A with Ryan Speaker: Ryan Fosdick - Lead author of DEMON, Daydream When: Tuesday, June 2, 6:00 PM CET / 12:00 PM ET / 9:00 AM PT Where: Streaming live on YouTube: Plus our Discord: https://discord.gg/g7F2HCa9VB Who this is for: Musicians and producers curious about real-time AI as a remix surface. AI and audio ML researchers working on streaming diffusion, control, or music models. Engineers building on ACE-Step, StreamDiffusion, or adjacent open-source music infrastructure.