cantor generate

Usage

cantor generate "a slow blues, brushed drums" -o blues.wav

The caption is the song you are asking for. Everything else is optional.

Option	What it sets
--model model:tag	Which installed model; the first one installed otherwise
-o, --output PATH	Also copy the finished song here, as WAV
--lyrics TEXT	The words to sing
--duration SECONDS	How long, up to ten minutes
--seed N	The same seed and caption make the same song again
--steps N, --cfg N	The model’s own sampling controls; leave them for the model’s defaults
--detach	Queue it and return at once

What it does

It queues the song on the running node and follows it as it moves through the model’s stages, then prints where the finished song is. With --detach it prints the job’s id and returns; the node keeps making it.

The node makes one song at a time, oldest first, and keeps the queue across restarts: a song half made when the node stops carries on from its last finished stage.

$ cantor generate "a slow blues, brushed drums" -o blues.wav
a slow blues, brushed drums
  model     acestep:1.5-fast
  backend   cuda12 · …
  output    /home/you/blues.wav

   64%  diffuse  5 / 8
completed /home/you/.local/share/cantor/…/master.wav (exported to /home/you/blues.wav)

From a sound

Three ways to start from a sound instead of from nothing. Each needs a model that can do it; cantor list --can shows which.

A version of a song. --version-of SONG re-makes a song in the style you describe, keeping its tune and form. --closeness 0..1 is how far it may go: 0 stays close, 1 is free. --keep-voice keeps the source’s voice (needs timbre).

A longer song. --longer SONG --add SECONDS adds that much, at the end, or at the start with --at start.

A file of your own. --source FILE starts from any recording on this computer, with --closeness as above. Add --tune when the file is a melody you hummed, whistled or played rather than a finished song, and --tile SECONDS to repeat a short tune into a whole song first.

SONG is a song’s id: the job id that --detach printed, which is also the id of the song it made.

Songs from the terminal stay in the terminal

A song made here belongs to the node’s operator, not to a phone, so it does not appear in any phone’s field. Make songs you want on your phone from the app.

When it goes wrong

No model is installed. Pull one first.

The node is below its free-space reserve. The node keeps 2 GB free and refuses new songs below that. cantor rm a model, or free space elsewhere.

The job failed. The message says why; cantor logs has the rest.