Can ChatGPT Make Music?

Yes. ChatGPT can help write lyrics and suggest melodies and chords. It can also give instructions for making music. To turn those instructions into a recording you can play, ChatGPT needs a music-making tool. Soundroo connects to ChatGPT: ChatGPT gives the instructions, and Soundroo’s engine makes the sound.

The distinction matters because notes and audio are different. A melody written as note names describes what to play; a recording is the sound you hear. ChatGPT can contribute to both parts of the process when connected to the right tools, but a prompt alone does not guarantee a finished song with singing.

What ChatGPT contributes to a song

ChatGPT can help develop a theme and shape lyrics. It can also suggest a chorus or describe a chord progression. It can also produce musical instructions in text. In a research project connecting GPT-4 to music software, the model suggested notes and rhythms that the software converted into MIDI, a format that tells an instrument or music program what notes to play and when. The program, rather than the text itself, supplied the sound (JAMMIN-GPT).

Research also shows a gap between writing valid musical instructions and making strong music. In one study, software could often read GPT-4 outputs as ABC notation, a text format for written music. The researchers found weaker results than with a specialized music model on music reasoning and listener preference (ChatMusician study). Treat generated melodies as drafts: listen to them, check whether they fit the song, and ask for revisions.

ChatGPT’s exact capabilities depend on the tools connected to the conversation. OpenAI’s capabilities overview describes general ChatGPT features, while a separate music service may handle sound generation. For singing in particular, do not assume that a ChatGPT conversation will produce a finished vocal recording.

How to make a track with ChatGPT and Soundroo

Soundroo is a free, open-source tool that lets an AI agent, such as ChatGPT, make music. ChatGPT connects to Soundroo through a remote connector, a link that lets the two communicate. You add the connector once and sign in to Soundroo with Google, GitHub or an email link. Follow the Soundroo connection steps, then ask ChatGPT for a track.

For example, you could ask for “a slow, warm instrumental with electric piano, soft drums, and a rising melody.” Soundroo’s custom prompt builder also lets you choose styles, track count, length, and spoken words at soundroo.com/make/custom.

The process works like this:

1. Connect ChatGPT to Soundroo and sign in.
2. Tell the agent what sound you want, or ask it to surprise you.
3. Listen to the private preview and request changes in plain language, such as “faster,” “less piano,” or “keep the bass line.”

ChatGPT writes the track as a score, a structured set of instructions for tempo, instruments, notes, patterns, and sections. Soundroo’s JavaScript engine creates the sounds electronically and mixes them into a stereo track. Its instrument types include guitars, basses, and drums, along with keyboards and synths. It also includes strings and bells. Piano and string sounds are electronic approximations.

ChatGPT cannot hear the result for itself. The agent checks for technical problems such as clipping, which causes distortion when sound levels are too high, or silence. You listen and decide what should change. ChatGPT and Claude conversations generally produce one to three shorter, simpler pieces. Each conversation returns a private preview link, which expires after a day unless the track is published.

Soundroo does not make sung vocals or accept audio uploads. It also has no drag-and-drop editor. Short spoken phrases can become rhythmic chops, textures, or a vocoder-style robot voice, but the words are spoken rather than sung. You describe what you want and ask the agent to revise the score.

What you can do with the finished track

In ChatGPT, each track comes back as a private preview link that plays in any browser. When you like a track, ask ChatGPT to put it on soundroo.com. It then stays on your artist page, where you can make it public, unlisted or private. Open tracks can be downloaded as WAV, MP3, MIDI and score, and other people can remix them with credit under CC BY 4.0. See the Soundroo documentation for details.

If you use a coding agent like Claude Code or Codex instead of a chat, Soundroo runs on your own computer and opens a small music player in your browser. From there you can export WAV audio, the score, MIDI and separate instrument tracks.

Soundroo is free, but the AI agent you use is a separate service with its own pricing. The Soundroo code repository also describes the toolkit for coding agents that run commands on your computer.