Music Generation Guide
Compose original instrumental music from a text prompt.
What is Music Generation?
SonicVox Music Studio turns a written brief — genre, mood, instruments, vibe — into a finished stereo track (48 kHz) you can drop under a video, a podcast intro, a game loop, or a background bed. The engine composes; you never upload source audio.
The music engine is instrumental only. It does not sing, so the studio deliberately hides the vocals, lyrics and vocal-language controls when that engine is selected — a control the model would ignore is worse than a missing one, because you would pick it, pay, and get something else.
How to Compose a Track
Step 1: Open Music Studio
Navigate to Music in the sidebar (/app/music).
Step 2: Write the prompt
Describe the music as you would brief a composer:
- ✅ Good: "Warm lo-fi hip-hop bed — dusty vinyl crackle, mellow Rhodes chords, soft brushed drums, relaxed 78 BPM"
- ❌ Vague: "Chill music"
Starter prompts sit under the box if you want a working example to edit rather than a blank page.
Step 3: Set genre and mood (optional)
Two dropdowns narrow the direction without rewriting the prompt. Both default to Any.
- Genre — Cinematic, Pop, Lo-fi, Trailer, Ambient, Electronic, EDM, Acoustic, Folk,
Rock, Hip-hop, R&B, Jazz, Classical, Orchestral, Game Score, Latin, Reggae, Country, World
- Mood — Hopeful, Triumphant, Uplifting, Energetic, Romantic, Chill, Serene, Tense,
Dramatic, Mysterious, Warm, Adventurous, Melancholic, Playful, Dark
Step 4: Choose how many variations
Variations composes 1 or 2 takes from the same brief. Each take is billed separately.
Step 5: Compose
Check the Cost chip, then click Compose. The track is queued and the view switches to History.
Step 6: Play it back
Finished tracks appear under History (/app/music/history) with an inline player, the prompt, the genre and mood you chose, and the duration. Use the player's own menu to save the file locally.
Writing Effective Prompts
Name the instruments
- "Solo cello over sparse piano, close-mic'd, a lot of room"
- "Analog synth arpeggio, sub bass, gated reverb snare"
Say what it sits under
- "Background bed for a product demo — never pulls focus from narration"
- "Trailer sting for a 6-second logo animation, big hit on the last beat"
Give tempo and energy
- "Slow build from sparse to full orchestra over 30 seconds"
- "Driving 128 BPM four-on-the-floor, no vocals, radio-ready polish"
Describe the space
- "Recorded in a small wooden church, natural reverb tail"
- "Tight, dry, modern pop production"
Limits
| Music generation | |
|---|---|
| Length | The model picks its own length — about 30 seconds per track |
| Vocals | Not supported (instrumental only) |
| Variations | Up to 2 per request |
| Output | 48 kHz stereo audio |
| Reference images | Not supported |
Because the model chooses the length, there is no duration slider — only a note telling you what to expect. A slider there would let you change the price without changing the output.
Credits
Music is billed per track, not per second: the engine returns a fixed-length clip and the provider bills per clip, so a duration you cannot change must not move the price. Two variations cost twice one.
The exact figure for the model you have selected is shown in the Cost chip next to the Compose button before you confirm, and that quote is what gets charged — the price lives on the model row, so the studio always reads the live value rather than a number copied into a guide.
Troubleshooting
| Problem | Why | Fix |
|---|---|---|
| "I asked for vocals and got an instrumental" | The offered engine is instrumental-only, so the vocals, lyrics and vocal-language controls are hidden and vocal parameters are never sent. | Use Text-to-Speech for a sung or spoken layer and mix it over the track in Studio. |
| Compose button is disabled | The prompt is empty, or the cost exceeds your balance. | Add a prompt, or top up / upgrade — the chip turns red and reads "insufficient" when the balance is short. |
| The track is generic | The prompt named a category but not instruments, tempo, or role in the mix. | Rewrite it as a short production brief. Two focused variations beat one very broad prompt. |
| The track fights the narration | The brief asked for a full arrangement where a bed was needed. | Ask for "sparse", "background bed", "no lead melody", then duck it under speech in Studio. |
| Nothing playable in History | The job is still running, or it failed. | The History row shows the job status; a failed job is refunded, so check your balance rather than re-composing blind. |
Provenance and Rights
Generated music carries a SynthID provenance watermark — an inaudible marker that survives normal encoding. Your use of generated tracks is governed by the SonicVox Terms of Service.
Related
- Sound Effects — short, one-shot effects rather than musical beds
- Studio Editor — layer the track under narration and SFX on a timeline
- Text-to-Speech — the narration that usually sits on top