No sign-up. No install.
AI music prompt
Give the genre, the vocal and what the track is for. Lyrics get their own section so the style field cannot swallow them.
This is what comes out
Make a ten-minute ambient track for a meditation appContent for the style field — paste this comma-joined block into the style field:
ambient meditation music, ten-minute continuous track, 60 BPM, D major, no lead vocal, soft felt piano, bowed glass, warm analog pads, low-volume granular texture, sparse brushed percussion, sustained tones, slow harmonic movement, breath-spaced phrasing, open intervals, gradual layering, gentle release, quiet dynamics, wide stereo field, soft high frequencies, warm low-mid range, no abrupt transitions, no dramatic climax, seamless-feeling ending, intimate near-field mix
User-facing direction:
Keep the arrangement spacious enough for meditation rather than foregrounding a memorable song hook. Let the piano establish the recurring harmonic point; introduce texture gradually, then remove layers before the ending. Each transition should be caused by a change in density or register, not by a sudden effect. Keep the pulse implied rather than prominent.
Use the style block only for musical characteristics. Keep all sung words, if any, in the lyrics field. If the track is intended to remain instrumental, preserve the empty-vocal direction and do not add lyrics.Full instruction | What the engine reads it as | 2 sections |
|---|---|
| Left for the engine to decide | Nothing — the request and the instruction settled all of it |
| Set by the instruction | 4/4 |
| Lines you can drop | 1 line |
| Same subject on a second run | No |
The instruction went in as-is; this is what came back, unedited.
One run. Paste it again and the wording will differ.
This is what Mephify is programmed to produce, so it is AI-authored output, published exactly as it came out.
Make yours
Other media documents: AI image promptAI video prompt
Paste it into whichever generator you use. No tool names are baked in.
We use only the facts you provide — nothing invented, and your text is never stored.
Nothing filled in yet — fill at least one field so the prompt has material.
Ultimate prompt
We turn the choices above into an instruction you can paste into your AI.
The same instruction gives different wording on Gemini. Compare on the example page.
Common questions
- Can I use it without an account?
- An account is not needed. Give the genre and the vocal you want, press the button, and the result appears here.
- Where do I paste what comes out?
- Anywhere a music generator takes a text prompt. The style block and the lyrics come out as separate sections, so each goes in the field that asks for it.
- Does what I type stay anywhere?
- What stays is a count, such as how many characters you typed. The text itself goes to the AI provider once and does not stay on our server. Save the file from the finished screen if you want it later.
- Do the lyrics survive the style field?
- They do, because they are never written into it. The lyrics are a section of their own, which is why a style block full of genre, tempo and mood words cannot squeeze them out. Paste the lyrics into the lyrics field and the style block into the style field. Read more
- Will the same voice come back twice?
- Not on its own. In the breakup-ballad run the stand-in reported that the same voice is not guaranteed without a reference input or a voice saved in the service, and the prompt says so itself. Use that feature where your service has one.
- Who decides the length?
- You do, when you say it. The ten-minute track above defaulted none of its four parameters because the request gave the length; in the ballad run, length was the one thing left to the engine. Put the total length in the request and the sections are timed to it.
More ai music prompt examples
From the operator's verification log
What to watch for
Reproducibility is the only open point: the instruction leaves the first render's timbre, pad texture and stereo width to be recorded after generation and reused — so the second render matches only if you actually do that.
Where not to use it
A stand-in check, not a rendered track. Ten minutes may exceed a service's single-render limit; the instruction does not know your service's cap.