Meta’s open-source AI music model — text-to-audio for indie devs and researchers.
MusicGen is Meta’s open-source text-to-music model released in 2023 and continuously updated. Run it locally on a GPU, fine-tune on your own dataset, or use it via hosted APIs.
Who it’s for: Developers building music AI apps, indie game devs who want full ownership of generated audio, AI researchers, and technically-inclined creators who want a free, open-source option.
Meta released the weights, training code, and inference scripts under a permissive license. Run locally with a single GPU (8GB VRAM minimum).
Generate from a text prompt alone, OR condition on a reference melody for variations. The melody mode is unique — it preserves the tune while changing style.
300M, 1.5B, 3.3B, and the new Stereo-Large model. Trade quality for speed depending on your hardware.
Released under a research-friendly license that allows commercial use in derived products. Meta’s updated license explicitly permits monetization.
Run locally, or use Replicate, Hugging Face Inference API, or FAL.ai for pay-per-second hosted inference. No monthly subscription needed.
MusicGen is the open-source backbone of the AI music revolution. If you’re a developer or indie hacker building a music product, it’s the starting point. For non-technical creators, Suno and Mubert are easier paths.