Mustango
Controllable text-to-music diffusion with chords, beats, tempo, and key control.
About
Mustango is a text-to-music system based on diffusion models that introduces explicit musical control. At its core is MuNet, a music-domain-knowledge-informed UNet module that steers generation toward music-specific conditions such as chords, beats, tempo, and key predicted from the text prompt.
To overcome the scarcity of captioned music data, the authors created the MusicBench dataset with about 53K music-caption pairs enriched with music-theory descriptions. In evaluations, Mustango achieved state-of-the-art quality and greatly outperformed models like MusicGen and AudioLDM2 on controllability of chords, beat, key, and tempo.
What it can do
- Text-to-music generation with diffusion
- Control over chords, beats, tempo, and key
- MusicBench dataset of 53K captioned clips
- State-of-the-art controllability benchmarks
Setup
Clone the repo or load the model from Hugging Face with diffusers.
Prerequisites
Python, PyTorch, and a GPU for local generation.
Use cases
Works with
Trust score — 7/10
Scored 7/10 (metrics tier, metrics-v1). Evidence: Verified via vendor site/docs/GitHub and reputable coverage, wave 103; Trending score 66/100 from researcher assessment.
Sources
Every fact above was checked against these sources. Where a claim comes from the provider and couldn't be independently verified, the text says so.