Stability AI
Stability AI's open-weight music generator — full tracks up to 6 minutes
Stable Audio 3.0 (May 20, 2026) is a family of four text-to-audio diffusion models trained entirely on licensed data. The Medium model generates music compositions up to 6 minutes 20 seconds from a text prompt. Three models (Small SFX, Small, Medium) are open-weight on Hugging Face; the Large model is API-only for music platforms. **Step-by-step:** 1) `pip install diffusers transformers torch`. 2) Load: `stabilityai/stable-audio-3-medium` from HuggingFace (or `-optimized` for lower VRAM). 3) Pass text prompt + desired duration. 4) Output: stereo WAV file. **Model sizes:** Small SFX (~2 GB, SFX only) · Small (~3 GB, ~90s) · Medium (~6.5 GB, 6:20 min) · Large (API only). **Best for:** background music for video, game audio, podcast intros, rapid music prototyping.
Three open-weight models generate tracks up to 6:20 min — all trained on licensed data.
Small/Medium open-weight (free) · Large via Stability AI API