Riffusion is a free and open source real-time music and audio generation library with stable diffusion. Users only need to enter a music description, and AI can generate music of the corresponding style. This open source project was launched by Seth Forsgren and Hayk Martiros. The principle behind Riffusion is to fine-tune the spectrogram based on Stable Diffusion (text generation image model). The spectrogram is a visual representation of audio, showing the amplitude of different frequencies over time, and then high-fidelity Convert spectrogram image to audio.