Images, video, voice and music
AI music generation
In short
AI music generation is software that creates music from a text prompt, a hummed tune, lyrics or other musical input. It predicts and assembles sound patterns to produce a new song, instrumental track or musical idea.
- Also called
- AI song generation, text-to-music, AI music
- Topic
- Images, video, voice and music
- Level
- Beginner
- First described
- Agostinelli and colleagues introduced MusicLM, a notable modern text-to-music system, in January 2023.
AI music generation uses a trained model to turn instructions or musical material into audio. A prompt might describe a genre, instruments, mood and tempo, or provide lyrics or a melody to guide the result12.
Many systems first compress music into small digital units, sometimes called tokens, and then predict sequences of those units. MusicGen, for example, uses a language model over several streams of compressed music tokens and can condition its output on text or melody2. MusicLM instead describes generation as a hierarchical sequence-to-sequence task and reported music at 24 kHz1.
The result is not a recording of a band playing in a room, and it is not necessarily a finished, mix-ready song. Models can lose musical structure, mispronounce lyrics or imitate a familiar style. The quality and permitted uses also depend on the model, its training data and the service's terms.
People meet the technology in song-making apps, background-music tools and creative software. It can produce a rough backing track, a jingle idea or an instrumental bed for a video, but a person still needs to check the output before publishing it.
Example
Someone types ‘an upbeat acoustic pop song about a train journey’ into a song-making app and gets several vocal and instrumental drafts to edit or replace.
Why it matters
It lowers the cost of making a musical sketch, even for people who cannot play an instrument. It also raises practical questions about copyright, consent, attribution and whether generated music is suitable for commercial use.
Questions people ask
How does AI music generation work?
A model converts musical audio into a representation it can predict, then generates a sequence from text, lyrics, melody or another prompt. MusicGen uses compressed music tokens, while MusicLM uses hierarchical sequence-to-sequence modelling12.
Is AI-generated music real music?
It is generated audio that can contain recognisable musical elements such as rhythm, harmony and melody. Whether it counts as a finished musical work, and what rights apply to it, depends on the output, the human contribution and the law that applies.
Can AI make a song from my humming?
Some systems can use a melody as a condition and transform it into a fuller arrangement. MusicLM demonstrated conditioning on hummed or whistled melodies alongside a text description1.
Related terms
Tools that use it
- SunoTypes a sentence or hums a tune and gets back a full song with vocals in about a minute.
- UdioAI song maker that settled with Universal and Warner and now keeps every song locked inside its own app.
- ACE-Step 1.5Open-source music generation model that writes full songs with lyrics and runs on ordinary consumer GPUs.
- ElevenMusic (ElevenLabs)ElevenLabs' song-making app, sold as licensed AI music; ElevenLabs itself is covered under voice and translation.