AI Music Generator

An AI music generator composes audio from a description. Instead of searching a stock library for something close enough, you describe the track you want, such as a lo-fi beat at a slow tempo with dusty piano and vinyl crackle, and the model makes a new piece to match.

On Mage that model is Seed Audio. One prompt can produce music, sound effects, voices or dialogue, and the MP3 comes back at exactly the length you choose: 5, 10, 30, 60 or 120 seconds. That exactness matters for video, where a track that ends on the cut is worth more than a longer one you have to trim.

Because Mage also makes the video, the track does not have to travel far. The Music Video app turns a track of up to 15 seconds into a performance by your saved characters, and video models such as Cherry, Plum, Lemon, Blueberry and MiniMax-H3 can take audio you upload on the Audio page as a reference.

The more a prompt sounds like a brief to a composer, the better the result. Cover five things: genre or style, tempo or energy, the lead instruments, the mood, and the shape over time. "Upbeat synth-pop at a quick tempo, punchy drums, bright arpeggiated synths, warm bass, optimistic, builds for the first half and hits a big chorus at the midpoint" gives Seed Audio a structure, not only a vibe.

Shape matters most on short clips. For a 10-second sting, ask for a quick rise and a clean ending. For a 30-second bed under a voiceover, ask for something steady and unobtrusive without a big drop. For a loop, ask for a steady groove that does not resolve, so it sits under a scene that runs longer than the track.

• Name the genre first, then the tempo
• List two or three lead instruments
• Describe the mood with a scene: "late-night drive", "sunrise run"
• Say how it starts and ends
• Ask for an instrumental bed under narration

Seed Audio is not only for music. Describe a sound the way a sound designer would: the source, the space and the distance. "Heavy rain on a tin roof, distant thunder, inside a small cabin" or "a crowd in a stadium rising to a roar, then applause" each come back as a clip you can layer under a video. Five or ten seconds is usually enough for a single effect; thirty or sixty for a bed of ambience.

For spoken words over the music, generate the voice separately so you can balance them. The AI voice generator covers narration and character voices.

The Music Video app takes a track of up to 15 seconds and one to three performers, which must be saved characters, plus up to three reference images, and makes them perform to it with Plum, Plum Max, Lemon or a Cherry model. Generate your track at 10 seconds for it, or trim a longer one before you upload. The AI music video generator walks through it, and the AI dance video generator covers copying a dance onto a character.