Mousai
Efficient cascaded diffusion for minutes of 48kHz stereo music on one consumer GPU.
About
Mousai showed in early 2023 that diffusion could do long-form music without a data-center GPU. It is a cascading two-stage latent diffusion model: a diffusion autoencoder first compresses music into a compact latent, then a text-conditional latent diffusion model generates in that space. The result is multiple minutes of high-quality stereo music at 48kHz from a text description, with inference fast enough to run in real time on a single consumer GPU.
The authors released trained models and a set of open-source libraries explicitly to grow an open-source culture around music diffusion, and the paper backs the design with property analyses against existing music models. At 71-plus citations it became a reference point for efficient music generation, though its evaluations are provider-run and the project has seen no successor.
Specs
Context window
Pricing
Free quota
API access
Weights
Released
Benchmarks
Property analyses vs existing music generation models (provider-reported); 71+ citations as influence signal.
Provider-reported unless a source says otherwise. Benchmarks are marketing-adjacent by nature — treat them as directional.
Use cases
Trust score — 7/10
Scored 7/10 (verified tier, metrics-v1). Evidence: published paper with open models and libraries, 71+ citations, real-time consumer-GPU inference claim. Strengths: efficient two-stage design, 48kHz stereo long-form output, genuinely open release. Docked for: (1) evaluations are provider-run; (2) 2023-era, no active maintenance signals. License: open-source release; check repo license before commercial use.
How the trust score is built
Full methodology →- Adoption · stars/downloads, log-scaled vs peers
- 7.0/10
- Maintenance · recency, releases, issue load
- 5.0/10
- Openness · license + docs
- 9.0/10
- Manual review · sanity read + red-flag check
- 7.0/10
- · Paper arXiv 2301.11757, Jan 2023
- · Cascading two-stage latent diffusion
- · Minutes of 48kHz stereo music; real-time on a consumer GPU (provider-reported)
- · Trained models and open-source libraries released
- · No logo staged (research project, no brand mark)