What it does

This repo contains two Google Colab notebooks. One generates music from a text prompt and duration via the Mubert API; the other pairs that output with Deforum Stable Diffusion to make prompt-driven music videos. You run cells, wait for the cloud GPU, and download results.

The interesting bit

The generation pipeline is a human-AI collaboration sandwich: a transformer matches your prompt to Mubert’s internal tags in latent space, but the actual sounds—bass loops, leads, etc.—are recorded by human musicians, not synthesized. Mubert’s tech arranges and composites them on the fly. It’s less “AI composes music” and more “AI stage-directs a session player.”

Key highlights

• Two notebooks: bare-bones text-to-music, and Stable Diffusion + music video generation
• Each request produces a unique combination; Mubert claims repetition probability is “extremely small”
• Free tier exists but requires Twitter attribution (@mubertapp, #mubert)
• Music can be used for image/video sync; releasing on Spotify/Apple Music as your own work requires a commercial license
• Explicitly described as a “first step” toward more complex generation

Caveats

• The actual music generation happens entirely on Mubert’s servers; these notebooks are thin API clients
• License terms are custom and somewhat restrictive—read them before using output commercially
• No candidate images provided in the repo

Verdict

Grab this if you need quick, attribution-tagged background music for a video project and don’t mind running a Colab. Skip it if you’re looking for an open-source generative audio model you can hack on locally; the heavy lifting is proprietary and remote.