Mubert reads tags, not prose
Mubert's text-to-music model assembles from a tag space: genres, moods, tempos, instruments. That means an entire useful prompt fits in one or two sentences, and every phrase in it should be something the model can actually match — a genre, a mood word, an instrument. Adjective chains and metaphor waste tokens.
Layer by layer
- Genre first. "melodic house", "lo-fi hip hop", "ambient drone" — one primary genre, the one you would tag the track with.
- Tempo as an explicit number. "at 124 BPM" pins the pace far better than "energetic" alone.
- One or two mood words. "rain-soaked", "driving", "weightless" — Mubert knows a bounded mood vocabulary; concrete beats decorative.
- Instruments with "featuring". Three or four max, most important first — "featuring dusty drum machine, upright bass, warm Rhodes, vinyl crackle".
- Continuity clause. "Continuous, loop-friendly arrangement" tells the model the shape you want — essential for Stream mode, harmless in Track.
- Exclusions at the end. "without vocals, without crowd noise" — state what must not appear; Mubert is instrumental-only, so vocals are already out, but "no drums" or "no abrupt transitions" are honored as direction.
Prompt or tags?
Both inputs reach the same model. Use the sentence prompt when you want instrument-level direction ("featuring upright bass and tape hiss"). Use the comma-separated tags — "lo-fi hip hop, nostalgic, slow, instrumental" — when you are iterating fast on genre/mood combos. Tag lists of 4–8 terms are the sweet spot; longer lists dilute.
Mubert prompt examples (copy-paste)
Melodic house · 124 BPM
Lo-fi hip hop · 84 BPM
Focus ambient · 70 BPM
Synthwave · 110 BPM
Tag-list form (fast iteration)
More ready-made prompts across moods and engines: AI music prompt examples.
Track vs Stream, and prompt length
- Track mode generates a finished piece with a start and an end — use the same prompt, but drop the loop clause if you want a resolved ending.
- Stream mode generates continuous, loopable audio — livestreams, focus sessions, background beds. Keep the "continuous, loop-friendly" clause in.
- Length: one to two sentences. Mubert's model is tag-driven; past two sentences you are adding words, not direction.
- Leave out artist names. They describe less than instruments and mood, and they risk content filters.
- Leave out contradictory genres. "Drum and bass ambient" pulls two tempos; pick a base and borrow one texture.
- Iterate on one layer. Close but dull? Fix the instrument list or the production clause — not the genre.
Generate a Mubert prompt from a reference track
The hard part of a Mubert prompt is the honest numbers: tempo, rhythm feel, brightness, dynamics, the instruments actually present. Music DNA Studio measures them from any MP3, WAV, FLAC or MP4 — locally in your browser, nothing uploaded — then outputs a Mubert-shaped prompt and its tag-list form (plus versions for Eleven Music, Stable Audio, Soundraw, Beatoven and AIVA). The spec underneath is measured; the wording is yours.
Describe The Sound
Drop in a reference, pick Mubert, copy prompt + tags. Free, no signup, nothing leaves your device.
ANALYSE YOUR AUDIO — FREE →FAQ
What is the best format for a Mubert prompt? ▾
One or two sentences: genre and tempo first, then mood, then instruments, then a continuity clause like "continuous, loop-friendly arrangement". Every phrase should be a tag the model can match.
Should I use a Mubert prompt or the tags field? ▾
Either — same model, two input styles. The sentence prompt is better for instrument-level direction; the comma-separated tag list is better for fast iteration on genre/mood combinations.
What is the difference between Track and Stream mode? ▾
Track generates a finished piece with a start and an end. Stream generates continuous, loopable audio for as long as you need it — livestreams, focus sessions, background beds.
Can Mubert generate vocals or a singer? ▾
No — Mubert output is instrumental. Direct the melody to instruments instead: "featuring a lead synth motif" rather than "with female vocals".
Can I write a Mubert prompt based on a song I like? ▾
Yes — describe the song's measurable properties (BPM, rhythm feel, brightness, moods, instruments) rather than naming it. Music DNA Studio analyzes any track in-browser and generates a Mubert-formatted prompt automatically.