Skip to main content
Glass piano keys with music notes and colorful sound waves
AI Studio → Music includes Generate, Video and Edit. Generate makes a new track; Video makes music for an uploaded clip; Edit changes existing music.

Generate a song

1

Describe the sound

Name the genre, instruments, mood, energy and arrangement. Describe the result rather than asking for an exact copy of someone else’s track.
2

Choose the music version

The workspace offers Music v2.5 and Music v2. Use the version shown in the model selector.
3

Set the length and lyrics

Choose a length from 15 seconds to 5 minutes. Auto lets the model create lyrics, Custom uses the words you supply, and Instrumental requests a track without sung lyrics.
4

Generate and listen

Check the estimate, generate, then listen through the whole track before editing or downloading it.

Lyrics modes

Include musical direction in the prompt. Use the dedicated lyric editor for the words you want sung.

Score an existing video

In Video, upload the clip you want to score and describe its musical direction. This creates music for the video; it is not the same as generating a visual clip in Imagine. Review the soundtrack against the picture before using it in a production.
Glass film production objects representing video scoring

Finetunes

The finetune controls expose curated options and your own trained musical styles. Training accepts reference audio from 10 seconds to 10 minutes. Use audio you have the rights to submit, then select the resulting finetune where supported. Training and generating are separate operations: inspect their status and credit information rather than assuming training is free or instant.

Continue with editing

Use Edit songs for lyric targeting, exact-time insertion, extending or restyling sections and batching saved changes.
Choose Instrumental before generating.
Download or use the clip as your video input in Music’s Video mode.
Use the song editor and a targeted range instead of regenerating the entire track.
Music generation is marked Non-ZDR. Avoid sensitive prompt material and private reference audio. Generation and training use credits; AI Studio controls require Pro or Elite.