wtf is music-maker?

MusicMaker AI is an open source platform that generates long, non repeating music tracks, up to 60 minutes, from an AI model in a short amount of time. It uses a music generation model called MusicGen to create a short core track, roughly three minutes long, and then extends it into a much longer mix. Instead of asking the AI to generate 60 minutes directly, which would be slow and prone to sounding repetitive, the tool builds the long mix from that shorter core piece using math based crossfading and looping. It also supports feeding previously generated audio back into the model so it can keep extending a track with new material rather than just repeating the same loop. A crossfade stitching step smooths over the points where separate segments are joined so listeners do not hear clicks or pops at the seams. Finished audio is saved in a 16-bit format that plays reliably in any web browser. The project has a React and Vite frontend for the user interface and a Python backend built on FastAPI that runs the actual music generation using PyTorch and a GPU. Because generation is GPU intensive and can take a little time, the backend queues each request and tracks it asynchronously so the connection does not time out while a long track is being built. To run it locally, you set up a Python virtual environment for the backend, install its dependencies, and start it with uvicorn, then separately install and run the frontend with npm. For deployment, the frontend can go on Vercel, while the backend needs an actual GPU, so the README recommends a cloud GPU provider called RunPod and includes a Dockerfile to make that setup straightforward. It is built by a single developer as an audio engineering project combining modern AI models with modern web tools.