The AI Music Tool: Generate Original Tracks

The music tool generates original tracks from a text description — background music for videos, intros, ads, loops and mood pieces. This guide covers the engines and how to describe a track so you get something you'd actually use.

The music models

All three are from the ACE-Step family — the difference is speed versus richness:

• Automatic (recommended): a good all-round choice. Start here.
• ACE-Step Turbo: fast drafts while you're finding the right vibe.
• ACE-Step: the everyday choice for usable tracks.
• ACE-Step XL: more depth and texture for pieces that carry a video or an ad.

“Describe the feeling and the use, not just the genre. "Calm lo-fi for a study video" beats "lo-fi" every time.”— WorkCrafter

How to describe a track

You give the tool a caption (what the music is), optional lyrics, and settings like duration and tempo. The caption does most of the work — treat it like a brief for a session musician:

• Genre and mood: "upbeat indie pop, hopeful and bright".
• Instrumentation: "acoustic guitar, soft drums, warm bass".
• Use case: "background for a product demo" tells it to stay out of the way.
• Energy and tempo: state a rough BPM or say "slow and mellow" / "driving and energetic".

Caption: warm lo-fi hip-hop, relaxed and nostalgic, mellow piano, vinyl crackle, soft drums, background for a study video
Lyrics: [instrumental]
Duration: 60s · BPM: 75

For instrumental music, set the lyrics to [instrumental] so the tool doesn't invent vocals. If you do want vocals, keep the lyrics short and simple — dense wordplay rarely lands.

Duration and tempo

Ask for the length you'll actually use. Shorter clips hold together more reliably than long ones, and it's easy to loop a clean 30-second piece under a video. If a track drifts or loses the thread, shorten it and regenerate rather than pushing for a longer take.

Tips that help

• Name two or three instruments — it anchors the arrangement more than a genre label alone.
• Say what the music is for; "background" produces something that supports rather than competes.
• Generate a few short options and pick the strongest, then extend that direction.
• Keep vocals simple, or go instrumental — it's the more dependable path to a usable track.

A caption formula that works

Most disappointing tracks come from a caption that names a genre and stops there. "Lo-fi" describes thousands of records, so the model picks the average of all of them. Four elements, in this order, remove almost all of that ambiguity.

1. Genre and era — "90s boom-bap", "modern indie folk", "80s synthwave".
2. Mood in two words — "warm and nostalgic", "tense and restless".
3. Two or three instruments — this does more than any other single change.
4. The job it has to do — "under a product demo voice-over", "opening titles".

That last one is the piece people leave out, and it is the one that decides whether the track supports your video or fights it. A model told the music is background will hold back the arrangement and leave room in the middle of the mix. Told nothing, it writes something that wants to be listened to.

Working with lyrics

Leave the lyrics field on [instrumental] unless you specifically want singing — a vocal line the model invented will rarely fit what you had in mind, and it makes the track much harder to sit under narration.

When you do want vocals, keep them short and plain. Four to eight lines with a repeated hook works far better than a full three-verse song: the model has to fit the words to a melody it is also inventing, and dense wordplay is where that falls apart. Write the hook first, repeat it, and let the verses be simple.

Fixing the four common problems

• The track wanders or loses structure — it is too long. Halve the duration and loop it instead.
• It sounds thin — you named a genre but no instruments. Add two or three specific ones.
• It competes with the voice-over — say "background" in the caption and lower it to about a third in the edit.
• Every generation sounds the same — vary the seed and change one element of the caption, not all four.

Matching music to the format

Different jobs want different lengths and energy, and knowing the target up front saves regenerating.

• Social clip (15-30s): high energy from the first bar — there is no time for an intro.
• Product demo (60-90s): steady and unobtrusive, no dramatic changes that pull attention off the narration.
• Opening titles (10-15s): one clear idea that resolves cleanly, so the cut lands on the resolution.
• Long-form background: generate 30 seconds you like and loop it — a genuinely long generation drifts.

Licensing and using it commercially

Tracks you generate are yours to use commercially under the platform terms, including in monetised videos and client work. There is no per-use royalty and nothing to clear. Keep a note of the caption you used — if a client later asks for "more like that one", the caption is the recipe.

How long a track should actually be

Generate to the length of the edit, not to a round number. A 30-second bed under a 22-second video means you cut the music, and music cut without a musical ending is the most obvious tell in an otherwise good piece.

• Social clip or ad: 15-30 seconds. Ask for a defined ending rather than a fade.
• Explainer or product demo: 60-90 seconds, low energy, no strong melody to compete with narration.
• Loop for a background: describe it as seamless and keep the arrangement sparse — fewer elements loop more cleanly.
• Intro sting: 3-6 seconds. Name the ending explicitly, or you get a fragment.

When a track has to sit under a voice, ask for it plainly: sparse, mid-tempo, no vocals, no lead melody. Music written to be noticed and music written to be underneath are different requests, and the caption is where you say which one you want.

What it costs

A music generation costs 15 credits — about $0.30 at the $5 starter rate, and less on larger packs. Failed generations are refunded automatically.

Frequently asked questions

Can I use the music in monetised videos?

Yes — tracks you generate are yours to use commercially under the platform's terms, including in monetised content. Keep a record of what you generated.

How do I get an instrumental with no vocals?

Set the lyrics field to [instrumental]. Otherwise the tool may add a vocal line to fit the caption.

My track wanders. What do I change?

Shorten the duration and tighten the caption. Long generations are where coherence breaks down; a focused 30–60 seconds you can loop is usually more useful anyway.

Start composing

Open the music tool, write a caption with a mood, a few instruments and a use case, set a short duration, and generate a couple of options. Extend the one that feels right.