abv1 · SA3 engine
Two demo experiences built on Stable Audio 3 — hear what your track sounds
like as a prompt, and hear the same prompt anchored to your audio vs. free.
The Playground tabs expose the raw knobs; headless /generate and
/generate_a2a endpoints sit behind the UI for batch callers.
Drop a track in, get a description a model can actually use.
We measure it (tempo, key, loudness, sections, what's playing) and write three prompts at different distances from the original. Copy any of them into Ableton MCP or your own tooling — or press a Generate button and hear a 10-second sketch.
Analysis runs off-GPU and takes ~20-40s (longer if the analysis Space is cold).
Faithful
closest to the original — structure and timings kept
Fresh arrangement
same feel, tempo and key — new arrangement
Exploratory
mood and tempo only — go somewhere else
Powered by Stability AI. Models are used under the
Stability AI Community License;
see NOTICE.md in this repo. All audio here is AI-generated — not real
releases by any artist.