Stable Audio Open

Stable Audio Open generates audio from text for artists and researchers experimenting with sound effects, field recordings and music samples. It performs better on effects and environmental sounds than music, and it can't generate realistic vocals.

• Text-guided audio: Generate stereo clips of up to 47 seconds from English descriptions, with a choice of clip length.
• Generation controls: Use negative prompts to discourage unwanted qualities and fixed seeds for repeatable generation.
• Local inference: Run the model on your own hardware through stable-audio-tools or Hugging Face Diffusers, using a CPU or CUDA GPU.

The model was trained on English descriptions, so prompts in other languages perform less well. Results also vary across musical styles and cultures. Its intended use is research and experimentation; downstream applications require further risk evaluation and mitigation.

Access to the model files requires a Hugging Face account, acceptance of the terms and agreement to share contact information. The model uses the Stability AI Community License, with commercial use subject to Stability AI's licensing terms. The separate stable-audio-tools library uses MIT.

Training recordings come from Freesound and the Free Music Archive under CC0, CC BY and CC Sampling+ licenses, with attribution available for the recordings.