Riffusion
Generates music by diffusing spectrogram images — the original open-source project, now unmaintained.
| Language | Python |
|---|---|
| Category | ML & generative |
| License | MIT (code) · CreativeML OpenRAIL-M (weights) |
| Platforms | Linux Windows macOS Web |
| Install | Clone github.com/riffusion/riffusion-hobby, then pip install -r requirements.txt |
| First released | 2022 |
| Maintained | Not actively maintained (last activity 2024) |
| Links | Source |
Strengths
- Novel spectrogram-image approach to audio synthesis
- Original code and model are open (MIT)
- The web app adds vocals and full songs
Limitations
- The open model is experimental and lower-fidelity
- No longer maintained (last commit 2024)
- The hosted web app was rebranded to Producer.ai, then acquired by Google as Flow Music, and no longer uses this spectrogram method
Best for
Experimenting with spectrogram-based music generation.
Minimal example
# The open model runs Stable Diffusion on spectrogram images; see the repo:
# https://github.com/riffusion/riffusion-hobby