Build your request

Fill in the music you want, add your own lyrics, then copy the style or download a YuE2 request. This tool prepares text; it does not generate audio.

Request previewJSON

Ready to use with the official Python pipeline.

Runs in your browser. Your text is not uploaded.

Use your request

Save the downloaded file in the YuE repository, activate your environment, then run:

TERMINAL / FILE
yue2 generate --request request.json --output outputs/my-song

Write a brief a musician could follow

Start with a genre, a vocal character, a language, a tempo and two or three instruments. “English indie pop, warm female vocal, 96 BPM, piano, rounded bass, restrained drums” gives a more coherent direction than a long list of incompatible genres.

A tempo in the prompt is an instruction to the model, not a timing guarantee. If you need a specific composition or rhythm, inspect and edit the score after a first generation.

Let the lines breathe

Use [Verse], [Chorus], [Bridge] and [Outro] to mark the structure. Keep neighboring lines close enough in length to sing naturally. A chorus works better when its central phrase is easy to repeat.

Write production notes in Style. A sentence such as “add a bigger drum sound here” in the lyrics can be interpreted as words to sing. For a first run, a short verse and chorus make pronunciation and structure easier to assess.

Change one thing per take

Keep the lyrics, planning mode and seed fixed while comparing two style briefs. Then try a lyric revision. Save each output with its request so you can identify which change helped.

The same seed does not guarantee identical sound across hardware or software versions. Score editing also renders a fresh recording; unchanged notes need not produce the same waveform.

Sources & further reading

Documentation checked against the linked sources. We have not benchmarked generation on our own hardware.