StringSong

AI & Content Notice

Last updated: 2026-10-17 · Version 1.1

We don't make StringSong to replace people. Music created by human artists carries a warmth and a story nothing can replace — we respect, value and enjoy their work. StringSong is simply a different experience for your free time: a change of mood, stories from virtual singers, or a melody to play along with from the score. If a style here speaks to you, please go and support the human artists who make it.

The short version: every song, every singer and every picture in StringSong is made with generative AI. The singers are fictional characters. We do not clone or imitate real people's voices.

Songs

  • The music and vocals are generated with ACE-Step, an open-source AI music model, from a style description (genre, mood, tempo) and lyrics.
  • Lyrics are written with the help of AI language models and edited by us.
  • Our style descriptions name genres and moods, not real artists or existing songs. We do not prompt the model to imitate a particular performer.
  • Songs are credited to a singer only after an automatic voice-similarity and voice-type check; earlier songs were regrouped by voice using the same check.

Virtual singers

  • Each StringSong singer (for example 林霜 Lin Shuang or Wren Calloway) is a fictional character with an invented name, story and look. Any resemblance to a real person is unintended.
  • Each singer's voice is AI-generated. To help a singer's songs sound alike, the model may be given that singer's own earlier AI-generated vocal as a reference. No recordings of real singers are used as voice references, and no real person's voice is cloned.
  • Instrumental "producers" (such as 慢潮 Slow Tide) are fictional too.

Artwork

Singer portraits and album covers are generated with AI image models from written descriptions of each character and album. Where a singer has no finished portrait yet, the app says so instead of using someone else's picture.

Music scores

Music scores are produced automatically from the recordings: the vocal melody is separated, its pitch is measured and turned into notes on a simple grid. They are approximate: pitch comes first, rhythm is simplified (4/4, sixteenth-note grid, estimated tempo), and occasional wrong notes are possible. We have not measured an accuracy figure on real songs, so we do not claim one.

How we label it

The app states that all songs in the library are AI-generated, and this website and our Terms of Service say the same. Pictures of singers and albums are labelled as AI-generated where they appear on this site.

Using the content

StringSong content is for personal, non-commercial listening and practice; see the Terms of Service. AI-generated music may not be protected by copyright in every country, but that does not give permission to reuse it outside these terms.

Report a concern

If you think a song, lyric, voice or picture sounds or looks too much like a real person or an existing work, or is offensive, email [email protected] with the song or singer name and what concerns you. We review every report and remove or change content where appropriate. For formal copyright claims, see Copyright & Takedown.

Version history: 1.0 (2026-10-17) first public version. 1.1 (2026-10-17) our stance on human-made music.