AI music generation
Generate a track from a description, or a genre variant as a new version on an existing track. Lyria through OpenRouter, ElevenLabs Music 2.5 with ELEVENLABS_API_KEY, Stable Audio 3 with STABILITY_API_KEY. Admin preview billed to Aden.
What it is
Describe a piece of music and a model plays it for you. On the generate page the result becomes a new track in your team, with the audio as its first version. From a track, you can also generate a new version in a genre or style (house, drum and bass, acoustic). In a version's comments the result comes back as a reply with the audio attached.
Each generation is priced per song or per clip. The price is on the button before you click it. While the feature is in admin preview, Aden carries the cost.
Using it
On the generate page
- Open Generate from the create-track page header (admins only for
now), or go to
/app/tracks/generate. - Describe the music. Genre, mood, instruments and tempo all help.
- Write lyrics, leave the box empty for the model to write its own, or turn on Instrumental.
- Give it a title if you have one, pick a model, click Generate.
- A minute or two later the track shows up in the list on the right, with a
play button and a link to its page. It is a normal track: status idea,
tags
aiandai:<model>, the prompt as its description.
On a track (genre version)
- Open the track, open the chevron on Add version and pick Generate a version (admins only for now). The empty state of a track with no versions still shows it as its own button.
- Type a genre or pick one (house, drum and bass, acoustic, and so on).
- Lyrics start from the latest version that has them. Edit or turn on Instrumental.
- Pick ElevenLabs Music or a Google model, then Generate.
- The new version appears in the list with a name like
ElevenLabs · house. It does not replace the current version. Play it like any other upload. Waveform is computed on first play, same as other generated audio.
The same page also mounts under /org for organisation workspaces.
In a version's comments
- From your draft. Type what you want in the comment box and click the sparkles button next to Comment (or next to Reply in a thread). Pick a model and generate. The draft is the prompt; the audio arrives as an AI reply in the thread.
- From someone's comment. Hover a comment and click its sparkles action. That comment's text is the prompt, and the reply lands under it, pinned to the same timestamp when the comment had one.
An AI reply shows an AI · model badge. Its audio plays inline like any attachment, and Add as stem lays it onto the version's timeline.
Finding AI-generated work later
Every generated version and its track get the tags ai and ai:<model>, for
example ai:lyria-3-pro or ai:elevenlabs-music. Genre versions also get a
slug tag such as house or drum-and-bass. Group the tracks board by tag or
filter the list on ai to see everything a model touched.
Who can generate
Admins only, while the feature is in preview. For everyone else the controls are not there at all: no Generate link on the create-track page, no Generate version on a track, no sparkles action in version comments. The generate page itself still explains the preview if someone opens the URL directly. Guests on share links cannot generate.
Sample clearance follow-up
Track who you are waiting on to clear each sample in a track, keep the contact and next chase date on the sample, research the owner with an AI chat, draft the request email and hand the rights holder a listen-only share link.
Player & queue
The persistent music bar at the bottom of the app. Transport controls, progress, volume, a dockable/floating mode, and a queue panel listing everything lined up. Each queue row and the track currently playing carry a single "…" actions menu (go to track, comment at the current playhead, jump to any version-card tab, copy link, share, download, all versions, remove from queue) so track actions are reachable without leaving the page you're on. In the bar that menu also holds the player's own options under an "Audio player" heading (progress bar style, speed, shuffle, clear queue, float/dock, minimize), so there is only ever one "…" button.