Three models live inside Lumen's AI Music tool, each with a personality:
- Eleven Music v1 — the highest quality, trained on licensed data. Pick this for commercial work. Strongest vocals.
- Suno v5.5 — the fastest. Catchy pop, rock and hip-hop structures. Cheap drafts.
- Udio v1.5 — the cleanest instrumental separation. Best for jazz, electronic, ambient, classical. Longest extensions.
The prompt
Describe the song in three layers: genre, mood, and references. Example:
A dreamy lo-fi track with soft female vocals about late-night trains and quiet cities, gentle jazz piano underneath, in the vein of early 2000s indie.
Style tags
Tags are the model's grammar. Pick 4-8 from the suggested list. Genre tag + 1 mood tag + 1 vocal tag + 1 instrument tag is a strong default. Avoid stacking too many — over-tagging dilutes the model's focus.
Instrumental or vocals
If you need underscore for video, generate instrumental — vocals fight narration. If the song is the product, generate with vocals.
Custom lyrics
When you write your own lyrics, use the structure tags Lumen ships with:
[Verse] Whispered lanterns drift along the wire Nobody waves but I am known [Chorus] Hold the light, hold the light Long enough to find your way home
Duration
30-60 seconds for ads and shorts. 90-180 seconds for songs you'll release. The model needs at least 90 seconds to deliver a proper verse-chorus-verse structure.