A rondocode song is music written as code: synths, patterns and a mix, played as you type. Lolly can play a song under a video, and the Rondocode utility lets you write one and export it as audio. Both record the song in the file's Content Credentials, so the music says where it came from.
Play a song in Lolly
The asset picker takes a song as a .rondo or .rondo.json file, or as a pasted rondocode.com share link. A song is code that makes its own sound, so it plays at whatever length its clip asks for. It can be the music bed under a video export or play in Neurospicy Mode's ambient player.
The song's code runs in Lolly's sandbox, with no network and no access to your files. Any part a render cannot play, such as a live microphone, or singing without the voice models, is listed in the asset's details.
From the terminal, lolly mix takes a song as the whole soundtrack; see CLI rendering.
Write and export a song
Open Rondocode from Utilities to write a song in rondocode's live editor and hear it as you type. Export it as WAV, MP3, M4A, Opus or FLAC, as a ZIP of stems, as MIDI, or as a recording of what is playing. Export length sets how many seconds an audio export lasts. The song file itself (.rondo.json) is your code as written and carries nothing added.
A song can sing its lyrics with sing(). The singing is made on your device by downloaded voice models, and a file that holds synthesised singing always declares the singing as AI-generated (see AI features).
Songs in the credential
A video or audio export that plays a rondocode song records each song in its Content Credentials as a source:
- the song's name;
- the SHA-256 of its
.rondo.jsonfile; - how it was rendered: in Lolly's sandbox, the
vmexecution class, with the renderer's version and seed, and any part the render could not play.
Music computed from a song's code is marked as algorithmic media, so Verify does not flag it as AI. Before you export, a line under Download lists the songs the credential will record; after the export, the same line reports what a reading of the delivered file found. Untick C2PA Credentials to leave the record out. A still image records no song, because it carries no sound. lolly mix and the MCP server write the same record.
Files from the Rondocode utility
Most on-device utilities write no credential, because they transform a file you already own. Rondocode makes new audio, so every audio file it writes carries Content Credentials that record the song as its source: its name, the SHA-256 of its .rondo.json file and where its code ran. That covers WAV, MP3, M4A, Opus, FLAC, each stem in a stems ZIP and a recording of what is playing.
- The editor's own renders run in its private frame (the
frameexecution class) and record the editor's version. - A render made by the CLI or a share link runs in the
vmclass and records the renderer's version and seed. - A WAV also states the same in its RIFF comment.
- Music computed from the song's code is algorithmic media and is not marked as AI.
When a song sings, the file declares that singing AI-generated in a machine-readable form. The credential's digital source type is composite media made with trained models, the record lists the sung parts, and one C2PA AI disclosure per model identifies each model the singing passed through: Supertonic-3 text to speech, the wav2vec 2.0 phoneme aligner, the ContentVec encoder and the RVC voice, with the repository and commit Lolly hosts it from. A file that holds synthesised singing is never saved without that declaration.
A MIDI file has no place for a credential, so a text event in the file carries the statement instead, and the export panel tells you so. The panel describes all of this before you export, and after each export the editor reads the saved file back and reports only what that reading found.
For what Content Credentials record in general, see Export a file.