A good BGM prompt can be one sentence. It does not need to be technical, but it should include the scene, mood, musical material, and things to avoid. The formula is simple: use case plus mood plus instruments plus density plus restrictions.
For example, 'cafe music' becomes 'quiet premium cafe BGM for rainy afternoon reading, warm Rhodes, muted guitar, soft brushed drums, gentle bass, natural loop, no vocals.' The second version is still readable, but it gives the model a much clearer target.
If you do not know instrument names, describe behavior. Say soft, warm, calm, bright, dusty, clean, dreamy, steady, sparse, or energetic. Mention whether the music should sit under speech, support a montage, loop in a store, or stay gentle for studying. Function is often more important than genre.
Avoid asking for famous artists or exact songs. That creates legal and quality problems. Describe the musical qualities instead: glossy synth pop, West Coast funk bass, smooth jazz cafe groove, mellow acoustic travel mood. The model can work with qualities without copying a person.
BGMFREE's prompt engine is designed to expand rough input, but good user intent still helps. If you write the scene honestly, the system can translate it into a better ACE-Step prompt. The goal is not to sound like a producer. The goal is to describe the moment clearly.