Skip to content
Search docs

Turn a musical idea into something you can hear.

Describe a style and write a few lyric lines. Make a first piece of music, then try transcription or change one input to find a version you want to keep.

An ideaStyle, lyrics, or Reference Audio
Listen and inspectHear the audio, read the score
Keep a versionSave explicitly when satisfied

This approximately 35-second piece was generated on a local GPU and saved: English lyrics, a warm voice, acoustic guitar, and piano. Listen first, then choose a starting point.

This is a compressed copy of the verified local piece, with its original recipe. Playback generates no new music. It does not demonstrate the untested experiments below.

Download the listening MP3 · View example provenance

Open the local workbench and start by creating a Project, then choose generation and listening or Reference Audio transcription. To control the next melody, follow Score editing and regeneration to inspect, audition, and explicitly select a Score. To try a new style from a reference, follow the Cover tutorial, inspect the score, and choose melody or full. Save the results you like, then stay in the same Project to compare two Versions and explore another direction. Music generation needs a ready real Runtime; default fake mode produces a test tone. Preparation and entrypoints →

This site provides tutorials, copies inputs, and plays existing examples. Creation, uploads, saving, and comparison take place in the local workbench.

For example: “Would piano suit the same lyrics better than guitar?” Keep the original inputs, change one dimension, then listen to the voice, accompaniment, and ending. Each generation creates a Candidate. It becomes a Version only when you explicitly save it.

Each tutorial helps you choose inputs, inspect results, recover from a failure, and try the next idea. Use help and reference when you need parameters or troubleshooting.

Source version: 1ce3268Page source: docs/learn/overview.en.md