Stable Audio Tools is Stability AI’s official training and inference code for generative audio models. It is the codebase behind Stable Audio Open, the open text-to-audio model for generating music and sound effects from text prompts.
The repository contains everything needed to run and train diffusion-based conditional audio generation models: inference scripts, a full training pipeline with dataset configuration, and a Gradio interface for trying models interactively. It targets modern PyTorch (2.5 or later) to take advantage of Flash Attention and Flex Attention, uses uv for fast, reproducible dependency management, and can also be installed from PyPI. The headline model, Stable Audio Open 1.0, is available on Hugging Face and generates variable-length stereo audio from text descriptions. The code is released under the MIT license and has over 3,800 stars on GitHub, with active development continuing through 2026; the model weights themselves are distributed under Stability AI’s license terms.
The Stable Audio Tools code is free and open source under the MIT license. Model weights such as Stable Audio Open are provided under Stability AI’s own license, which is free for research and many use cases.
Audio-AI researchers, music-tech developers, and sound designers who want to generate or study AI-composed music and sound effects with open models.