Hacker News new | ask | show | jobs
Riffusion Release v0.3 – Stable Diffusion for audio (github.com)
53 points by haykmartiros 1277 days ago
2 comments

We were blown away by the community interest in our Riffusion experiment (see https://www.riffusion.com/about).

We're now catching up and did a full rewrite to go from a hack to a quality software project. It also includes an extensible CLI tool and an interactive streamlit app for common tasks, MPS backend support, stereo spectrogram encoding, a test suite run by CI, and more. These tools provide a solid foundation to build upon.

Is it a good idea to save spectrograms as JPEGs? Won't the lossy compression be an issue?
I guess you could listen to what different lossy image compression algorithms sound like
Love it!
Why haven't you guys used <title> <artist> <album> <lyrics> as input to generate the song? It would be much cooler.