Stable Audio

Freemium

Stable Audio is an AI toolkit for making instrumental music and sound effects. You type what you want, or feed it audio to change. It gives you files you can drop into a video, a game, or a DAW.

It runs as a web app, a plugin for Ableton and Logic, or code you host yourself. The open models let you run it on your own machine if you want control and no per-use fees.

It sits in music generation. It is built for beds, loops, and textures—not songs with lyrics.

Open it if you need instrumental audio you can license cleanly. Skip it if you need vocals or polished pop tracks.

You get the most from it when this is your job.

• Video and game producers needing instrumental beds: You get six minutes of usable production music. It sits behind dialogue without fighting for attention.
• Sound designers wanting loops and effects: You generate foley and ambient beds on demand. The Small SFX model runs on a laptop CPU with no GPU.
• Electronic and ambient music creators: You get stereo width and believable layering. The output sounds deliberate, not random.
• DAW-based workflows needing plugin integration: You stay inside your host. No round-trip to a browser.
• Developers building generative audio into their own apps: You get API access and self-hosted models. You control the stack.
• Audio producers needing commercially safe generated music: Stability licenses the training data. You get clearer rights than most AI tools.
• ComfyUI users wanting audio in node workflows: You wire generative audio into existing pipelines with official templates.
• Content creators needing instrumental music at scale: Open weights mean no per-generation fees. You pay for compute, not credits.

These are the parts you keep coming back to.

• Six-minute generation ceiling: You make full instrumentals, not just eight-second stings. Other tools cap out sooner.
• Audio-to-audio and inpainting: You upload what you have and edit it. You are not stuck with the first output.
• DAW plugin with VST3/AU: You treat it like any other instrument. Your workflow does not change.
• Open weights for three models: You run Small and Medium locally. You fine-tune with your own samples on Apple Silicon.
• Fast iteration on the right hardware: You generate in seconds on an M4 MacBook Pro or H200. CPU-only Small models are quick enough for SFX.

The path depends on where you want to work.

1. Pick your path: web app, DAW plugin, or self-hosted code.
2. Describe what you want: tempo, instruments, mood, or BPM.
3. Generate and listen. Use inpainting to fix moments that miss.
4. Export stems or full tracks. Take them to your editor or DAW.

Skip it when these limits apply.

• No vocals or lyrics: You get instruments only. If you need a singer, look elsewhere.
• Learning curve for local setup: You need GPU knowledge to self-host Medium. Flash Attention 2 compilation is not friendly.
• Weak seed variation: You get similar melodies from the same prompt. You work harder for surprise.
• Rhythmic genres flatten out: You get average drums, not standout grooves. Electronic and ambient suit it better.