How to generate sound effects and ambience

Describe a sound and get it back as an audio file. Three engines sit behind the Sounds tab and they are built for different lengths: one runs to a minute for ambience, one to thirty seconds and is the one to loop, and one does short punchy hits between three and ten seconds. The Mix tab then layers up to six of them, with speech, into a single track. What is not here is music, and this page says exactly where that line falls.

Three engines, three lengths

Pick by length first and price second, because the envelopes barely overlap. Anything past thirty seconds only Sonilo will do; anything under three seconds has to be Sonilo or Mirelo, because Kling will not go shorter.

Sound effect engines. All prices are in ✦ gold credits.
EngineLengthPriceBuilt for
Sonilo SFX1 to 60 seconds✦1 per started 15 secondsAny effect or ambience — the default, and the only long one
Mirelo SFX 1.61 to 30 seconds✦0.34 a secondRich, loop-friendly ambience
Kling Audio3 to 10 seconds✦1.2 flatShort punchy effects

Which plan: The Audio Studio opens on the Starter plan. That gate lives on the server: every generating audio endpoint answers 403 to a free account, so it is not a greyed-out button you can work around. Everything behind these tabs runs on a paid provider, which means it is billed in ✦ gold credits — ⚡ green credits, the ones that refill daily, cannot pay for any of it.

What it costs: Sonilo bills per STARTED 15-second block, so 16 seconds costs the same as 30. Mirelo is linear per second, which makes it cheaper for very short hits and dearer for long beds. Kling is flat: the same ✦1.2 whether you ask for 3 seconds or 10.

How to generate a sound

  1. Open the Sounds tab. Go to /audio and pick Sounds in the rail. One prompt box, one engine picker, one duration control.
  2. Describe the sound, not the scene. "Rain on a car roof, steady, no thunder" gets you rain. "A lonely night" gets you something, and it will not be repeatable.
  3. Pick the engine by length. Sonilo for anything over thirty seconds, and for a first attempt. Mirelo when you want a bed that will loop cleanly. Kling for an impact, a whoosh or a door.
  4. Set the duration. Each engine clamps to its own envelope, so asking Kling for 30 seconds gets you 10 rather than an error. The quote updates as you change it.
  5. Generate and audition. The result is an mp3 in your assets library. Effects are cheap enough that generating three variations of a prompt is usually better value than refining one.
  6. Take it to Mix if it needs company. A single effect is often not the sound you wanted — rain plus distant traffic plus a door is. That is what the Mix tab exists for.

Layering: the Mix tab

Mix builds one audio file out of several generated tracks. Each track is either speech, with its own voice and engine, or a sound effect with its own prompt and duration. Every track carries a start offset and a gain, so a line of narration can land eight seconds in, over ambience that has been running from the start.

Mix limitThe number
Tracks in one mix6
Total length, and the furthest a track can start600 seconds
Characters on a speech track5,000
Per-track gain0 to 2 — 1 leaves the track untouched
Non-looping tracks requiredAt least 1

Sound tracks can be set to loop, which stretches them to fill the mix. Speech never loops — repeating a sentence under itself is never what anyone meant, and it would leave the mix with no defined length. For the same reason a mix made entirely of looping tracks is refused, with a message explaining why rather than a rendered guess at how long it should be.

What it costs: The mix itself is free. What you pay is exactly what each track would have cost generated on its own — the speech at its per-1,000-character rate, each effect at its engine's rate — because the layering is done by local ffmpeg on our own server. The quote shows the total before you commit.

A worked example, to make the shape concrete. A thirty-second vertical cut with narration over rain: one Sonilo track of steady rain at 30 seconds, set to loop and pulled down to a gain of 0.4; one Mirelo track of distant traffic at 20 seconds, offset 5 seconds in, at a gain of 0.3; and one speech track of about 600 characters on Flash, offset 2 seconds, at full gain. That is ✦2 for the rain, ✦6.8 for the traffic and ✦3.4 for the voice — one press, one file, and the layering adds nothing on top.

Tracks are rendered one at a time rather than all at once, so a six-track mix takes about as long as generating its six pieces separately would. The wait is the only thing the layering costs you.

Prompts that work

There is no music generation here

Worth stating plainly, because it is the thing people most often arrive expecting. The studio makes speech, and it makes sound effects and ambience. There is no music model, no tab, and no plan that unlocks one. If you need a score, bring your own track and layer your generated effects and narration around it.

What these engines are good at instead is the layer under a voice: room tone, weather, traffic, machinery, a crowd at a distance. On a narrated cut that bed does more work than a music track does, and it costs a few credits rather than a licence.

Open the Sounds tab — Opens the studio on Sounds. Prompt, engine and duration are set there; the cost is quoted before anything runs.

How long can a generated sound effect be?

Up to 60 seconds on Sonilo, 30 on Mirelo and 10 on Kling Audio. Kling also has a three-second floor, where the other two start at one second.

What do sound effects cost?

Sonilo is ✦1 per started 15 seconds, Mirelo is ✦0.34 a second, and Kling is a flat ✦1.2 whatever length you pick. All in ✦ gold credits, on a Starter plan or above.

Can I layer effects under narration?

Yes — that is the Mix tab. Up to six tracks across 600 seconds, each with its own start offset and gain, with speech tracks capped at 5,000 characters.

Does mixing cost extra?

No. A mix costs the sum of its tracks and nothing more, because the layering runs on local ffmpeg. You pay the same as if you had generated each track separately.

Can I make a backing track with this?

No. There is no music model on this platform — no tab and no picker row, on any plan. The Audio Studio makes speech and sound effects, and nothing else.

Why is my looping mix refused?

A mix needs at least one track that does not loop. Looping tracks stretch to fill the others, so a mix made only of loops has no defined end and is refused rather than rendered at an arbitrary length.