stable-audio-tools
Stability-AI/stable-audio-tools · Python · MIT
Training and inference toolkit for Stable Audio open models
stable-audio-tools packages the open Stable Audio stack for text-to-audio and music-oriented generation. Expect higher VRAM than ACE-Step on consumer cards.
Score
Setup at a glance
pip install stable-audio-tools Deep dive
stable-audio-tools is Stability AI's training and inference toolkit for the Stable Audio family of generative audio models — the open counterpart to their hosted music and sound generation.
It exposes the full loop: load a pretrained model from Hugging Face with --pretrained-name (prioritizing model.safetensors over model.ckpt), run inference, and fine-tune or train your own on custom audio.
The repo ships with model configs, training scripts and inference examples, and pairs with the published stable-audio-open-1.0 weights so you can start from an open base instead of scratch.