Ask ChatGPT for a song and something odd happens. It writes Python, runs it, and hands you a WAV file. Can ChatGPT make music? Technically, yes. Our 30-second test track came with chords, drums and a "female vocal." Press play, though, and the singer turns out to be a stack of sine waves humming vowels.
A real song needs a music model behind the chat. We connected one through Atlas Cloud and, on a free ChatGPT account, made four things in one project: a sung track, an instrumental bed, a voiceover and a 12-second video with the sound mixed in. Here is every step.
One ChatGPT project: a song, an instrumental, a voiceover and the finished video.
Key Takeaways
- In our test, ChatGPT coded a 30-second WAV without plugins, but its synthesized singer only hummed vowels.
- Add the Atlas Cloud plugin and the same chat returns finished MiniMax Music 3.0 songs with real vocals.
- One request turned ChatGPT's own lyrics into a 79-second sung track, and one song can run about five minutes.
- ChatGPT's Read aloud only plays replies inside the app, while an ElevenLabs v3 voiceover arrives as an MP3 file.
- ChatGPT then mixed the music and voiceover onto a Seedance 2.5 clip with ffmpeg, inside the same project.
Can ChatGPT Make Music on Its Own?
Sort of, and the "sort of" is interesting. In the ChatGPT desktop app we asked for a 30-second lo-fi song with a female vocal, as a playable audio file, and told it not to use any plugins. The first Python run errored. The second one worked, and ChatGPT presented the result as a track called Moonlight Dreams.
Then we read the code. Every sound in the file is built from math: sine-wave chords, a synthetic kick and noise bursts for the snare, all rendered at 22,050 Hz. The "singer" is a tone shaped to sound like the vowels oh, ay, ee and oo. ChatGPT did write four lines of lyrics, but they only appear as a comment in the code. No word is ever sung.
To its credit, the reply says so: the vocal is "synthesized to resemble a soft female singing voice, rather than a recording of a real singer." Here is exactly what it made. We reran ChatGPT's own script and the output matched its file byte for byte.
The native skill that is actually useful came next. We asked ChatGPT, still without plugins, to export the melody and chords as MIDI. It returned a four-track MIDI file at 82 BPM built on a Dmaj7, Bm7, Gmaj7 and A progression, ready to drop into a DAW and assign real instruments to. It even warned that the melody is an approximation of the audio, not a transcription.
So ChatGPT is a good sketchpad for lyrics, chord charts and MIDI. A finished song with a human-sounding voice is another matter, and OpenAI isn't promising one soon.
When a user asked for a native music module on the OpenAI forum, OpenAI support replied on September 8, 2026: "We don't have a timeline to share." OpenAI did build music models years ago, MuseNet in 2019 and Jukebox in 2020, but neither is a ChatGPT feature.
How to Use ChatGPT as a Music Generator With the Atlas Cloud Plugin
The Atlas Cloud plugin hands ChatGPT a catalog of 400+ models, including music and voice models, and returns their files to the chat. Below is where this article ends up: a finished video with a song bed and a voiceover, playing inside the ChatGPT thread that made it.
Install the plugin once from its Plugins Directory page and sign in with your Atlas account in the browser. Our launch walkthrough covers each click. We worked in Work mode inside a ChatGPT project, so every file the plugin returned was saved to a folder on our computer.
From there, each song or voiceover takes three moves:
- Write the request in the message box. Name the plugin and the model, describe the sound, and ask to see the request before anything runs.
- Send it. ChatGPT writes any lyrics you asked for, checks the model's current options, and replies with the exact request it plans to submit.
- Reply "confirm submission" once the request preview looks right. The plugin runs the job, waits for it and saves the file under the name you gave.
The model in that request is MiniMax Music 3.0. It takes two inputs that do different jobs: a style prompt that describes the music, and a lyrics field that carries the words.
Can ChatGPT Make a Song With Vocals From Your Own Lyrics?
Yes, and this is where ChatGPT's writing finally gets a voice. We let ChatGPT draft the lyrics and pass them straight to the music model in one message:
ChatGPT wrote nine lines, with a chorus that opens on "Rooftop cinema, stay a little longer." It put the section tags on their own lines, which is how MiniMax Music 3.0 reads the arrangement. After one confirmation, a 79-second MP3 landed in the project folder at 44.1 kHz stereo.
The track opens with about 16 seconds of wordless vocal humming. The verse comes in around the 20-second mark, the chorus follows, and the outro line "Keep the picture where we are" closes it near 68 seconds. We ran the file through a speech-to-text model to check, and every line ChatGPT wrote is sung in order.
Got your own lyrics? Paste them in place of the writing instruction and keep the tags. The structure you give sets the length: three short sections gave us 79 seconds, while a full verse, chorus and bridge layout runs longer.
Use ChatGPT for Music Beds Without Vocals
A song with vocals fights a voiceover. For video work you usually want a bed: the same mood, no singer. In the same chat we asked MiniMax Music 3.0 for "the same summer-night rooftop mood as mellow lo-fi indie, 90 BPM, soft Rhodes piano, muted electric guitar, light vinyl crackle, brushed drums" with no vocals.
The plugin set the model's instrumental switch, which tells it to ignore lyrics entirely. Back came an 81-second track, long enough to cut any short video from.
Does ChatGPT Text to Speech Give You a Voiceover File?
ChatGPT can talk, which is why so many people search for ChatGPT text to speech. Read aloud, which OpenAI added in March 2024, speaks any reply. Voice mode holds a spoken conversation in one of nine voices. Both are built for listening in the moment, and we found no way to save their audio.
| Option | What you get | A file to keep | Best for |
|---|---|---|---|
| Read aloud | The reply spoken inside the app | No option found | Listening to a long answer |
| Voice mode | A live spoken conversation in one of nine voices | No option found | Hands-free questions |
| Atlas Cloud plugin with ElevenLabs v3 | Your script read in one of 21 voices | MP3 in your project folder | Voiceovers for video, ads and demos |
For our rooftop clip we needed a narration track, so we asked the plugin for ElevenLabs v3 with the Sarah voice at stability 0.4. Lower stability gives a more expressive read. The script was one line of copy:
Friday, nine p.m. Bring a blanket. The city goes quiet, the projector hums, and for one night the skyline is our screen. Rooftop Cinema.
ChatGPT looked up the voice ID for Sarah on its own and showed it in the request. The result was a 10.2-second MP3, and a transcription matched the script word for word. You can hear it in the finished video below.
OpenAI does sell text to speech to developers through its API, but that runs outside the ChatGPT app and returns audio to your own code.
Put Your ChatGPT Music and Voiceover on a Video in the Same Chat
Here's the thing about a bed and a voiceover: they only matter once they sit under pictures. So we stayed in the project and asked for the picture next, with sound switched off because the soundtrack was already made. The model was Seedance 2.5 in text-to-video mode:
The clip came back at 1280×720 and 12 seconds, silent as requested. It opens on the projector reels in the foreground, then drifts past the beam to her profile as she laughs into the blanket. Then came the mix, in plain words:
ChatGPT wrote and ran an ffmpeg command on our machine. It delayed the voice by one second, dropped the music 12 dB from that point, faded the tail, and copied the video stream untouched. The output was trimmed to exactly 12.04 seconds. Turn your sound on for this one.
People searching for a Suno video generator usually want this exact step: moving pictures under a finished song. Any MP3 in the project folder works the same way, not only one made in this chat. For other ways to start a clip, see our tests of video from text, photos and references. A multi-shot music video with lip sync is a bigger build, covered in our MiniMax H3 music video walkthrough.
Which Music Model Should ChatGPT Use for Your Project?
The plugin draws on the same Atlas catalog the platform uses, so the real choice is which model to name in your request. These are the ones worth knowing for music and voice work, with the limits each model page lists.
| Model | What it makes | Limit per request | Pick it when |
|---|---|---|---|
| MiniMax Music 3.0 | A finished song from a style description, with your lyrics or lyrics it writes | About 5 minutes of audio, 3,500 characters of lyrics | You want a complete track with vocals, or a clean instrumental |
| ElevenLabs v3 | Expressive speech in 21 built-in voices | 5,000 characters | The voiceover carries emotion, like a trailer or a story |
| MiniMax Speech 2.6 HD | Speech in 15 preset voices with emotion, pitch and speed controls | 10,000 characters | You need a long narration in one pass, or one of 16 languages |
| Gemini 3.1 Flash TTS | Speech in 30 prebuilt voices, styled by plain-language directions in the text | 10,000 characters | You want to steer accent and pace by describing them |
| Seed Audio 1.0 | A full audio scene from one prompt: voices plus sound effects | Up to 3 voice references | You want a radio-drama style clip rather than a clean voice track |
You don't need to memorize parameters. Name the model, describe what you want, and ask for the request first. In our runs ChatGPT checked each model's current options before it built the preview, so you can also ask it which voices or settings a model accepts.
FAQ
Can ChatGPT sing a song in voice mode?
Don't count on it. OpenAI's GPT-4o System Card from August 2024 says the team "instructed the model to not sing at all" during the limited alpha of Advanced Voice Mode, and built filters to block outputs containing music. OpenAI's current voice mode help page doesn't list singing as a feature. For a sung vocal you control, write the lyrics and have a music model perform them.
How much does a song made in ChatGPT cost?
The native WAV and MIDI cost nothing on a free ChatGPT account, and the plugin is free to install. Generations bill to your Atlas balance at the model's listed price. On October 12, 2026, our song and our instrumental each cost $0.15 with MiniMax Music 3.0, and the ElevenLabs v3 voiceover cost $0.01 at $0.10 per 1,000 characters. The 12-second 720p Seedance 2.5 clip was the big line at $3.62, which reflects a 20% discount from the $4.52 list price at the time of writing. The whole project came to $3.93.
Can ChatGPT make a full-length song or just a clip?
A full one. MiniMax Music 3.0 can return about five minutes of finished music in a single request, and length follows the structure you write: our three short sections produced 79 seconds. ChatGPT's own Python track stopped at 30 seconds because that's what we asked for. To fit a song to a video, generate it long and let ChatGPT trim it in the mix, as we did.
Can you copyright or sell a song made with ChatGPT?
It depends on how much of it is yours. The U.S. Copyright Office report of January 29, 2025 puts it plainly: "Copyright does not extend to purely AI-generated material." It adds that prompts alone don't make you the author. Lyrics you write yourself are still your work. Whether you can sell the track also depends on the terms of the service that generated it and of the distributor you upload to, so read both before release.
Conclusion
That first Moonlight Dreams file is a fair picture of ChatGPT on its own: clever with code and theory, stuck at humming sine waves when it needs a voice. The MIDI export and the lyric writing are real skills. The sound itself has to come from a model built for it, and once that model is one plugin call away, the same chat that writes the words can also hear them sung.
To try it, install the Atlas Cloud plugin, open a project in Work mode, and start with one song request that asks for the preview first. Add a bed, a voiceover and a clip when you need them. So, can ChatGPT make music? With a music model attached, it now makes the song, the narration and the finished video in one thread.
