riffusion/riffusion-hobby

Generates audio from text prompts by converting between spectrogram images and audio waveforms, using a diffusion pipeline with prompt interpolation and image conditioning. Provides multiple interfaces including a CLI, Streamlit playground, and Flask API server for model inference. Supports CPU, CUDA, and Apple Silicon backends, with real-time generation requiring ~50 inference steps under five seconds on high-end GPUs like RTX 3090 or A10G.

3,883 stars. No commits in the last 6 months.