Sound for video

This lesson gives a silent clip its sound. You'll make a single effect from a description with ElevenLabs Sound Effects v3, let Sonilo V1.1 Video to SFX watch a locked clip and generate effects timed to it, and then see how the tracks come together in one mix.

Before you start

• Know which kind of sound you need. A single effect (a door, footsteps, wind) comes from a description. Sound that has to follow a moving picture comes from the video itself.
• Lock the picture first if you start from a clip. Every trim after generation moves the picture away from the sound.
• Scenario makes the tracks; you mix them in your audio or video editor.
• Takes are cheap compared with video. Make two or three of anything you'll use. The cost shows on the Generate button.

Models

Take a separate track when you'll mix, add narration or adjust levels; take the video with sound in it for a single clip that goes out as is. A music cue with an exact length comes from a music model (Generate music).

Step 1: Pick an effects model

Open Create › Audio, click the model name, then View all models. Tick Audio under Output and Sound Effects under Capabilities.

• ElevenLabs Sound Effects gives you a set length and a loop switch. This lesson uses v3.
• Sonilo V1.1 Text to SFX gives many variations, in lossless formats.

Step 2: Describe the sound

The main field is called Sound Description. An effects model reads physical facts, so give it, in this order:

1. The source and material: "A heavy wooden door with iron hinges."
2. The action, in stages: "creaks open slowly, then closes with a solid thud."
3. The space: "inside a stone hallway with a short echo." The space changes the sound as much as the source.
4. What to leave out: "No music, no voices." Effect models can add both.

A heavy wooden door with iron hinges creaks open slowly, then closes with a solid thud, inside a stone hallway with a short echo. No music, no voices.

Describe the sound, not the reaction. "Massive", "cinematic" or "scary" won't help. Write what makes it scary instead: "a slow low creak, then silence".

Step 3: Set the length and guidance, then generate

Below the description:

• Duration: set the length with a little room after the last sound, so the tail can fade. Here 5 seconds.
• Guidance: how closely the take follows the description. It opens at 0.3; here it's raised to 0.5. If a take ignores part of your description, raise it. Lower it to let the model vary the sound.
• Loop: leave it off for a one-shot like this door.
• Output Format: MP3 by default.

Click Generate. The take arrives at 00:05, with the creak, then the thud, in a stone hallway.

Here is the same description on ElevenLabs Sound Effects 2, at 5 seconds:

oak-door-creak-and-thud-elevenlabs-sfx-2.mp3Audio: oak-door-creak-and-thud-elevenlabs-sfx-2.mp3

Loops

For rain, a room tone or an engine, describe a steady sound with no beginning or end event, such as "steady light rain on a tin roof, constant, no thunder", and turn Loop on. Play the take twice back to back and listen for a click at the join.

Layers

Real effects are often layers: a body (the heavy door closing), a detail (the iron latch clicking) and a space (the stone room's tail). Make each one on its own, so you can balance them later and redo a bad layer cheaply.

Step 4: Start from the video

When the sound has to follow the picture, start from the clip. Lock the cut first: every trim afterwards moves the picture away from the sound, and you'd have to generate again.

1. Pick Sonilo V1.1 Video to SFX in the model menu.
2. Under Video, add the clip.
3. Leave Prompt empty to let the model describe the clip, or name the key sounds in one line:

Footsteps on paving stones, a busy market crowd, fabric flapping in a light breeze

4. For a clip with distinct scenes, click Add Segments and give each part its own description.
5. Under Settings, set Audio Format to WAV if you'll mix it.
6. Click Generate.

Then play the clip with its new track and check the hardest moments: impacts, footsteps, doors. They should land on the frames where they happen.

Step 5: Music, and dialogue that's already there

• For music that follows the clip, use Sonilo V1.1 - Video to Music the same way. It reads the clip's pacing and mood and returns the music as its own track.
• If the clip already has dialogue, protect it. Pull the audio out with Audio Extract, keep only the voice with ElevenLabs Voice Isolator, then add the new sound around it. The Video to Video models can also keep existing speech.

Step 6: Mix the tracks

Scenario makes the tracks; you mix them in your editor, in this order:

1. Dialogue or narration first, at a steady level.
2. Effects on their events. Nudge a few frames if an impact lands early or late.
3. Music under the voice, lower while someone speaks.
4. One loudness target for the whole piece: about -16 LUFS integrated for online video, checked with a loudness meter.

For quick fixes without an editor, three tools run in Scenario. Find them by name in the model menu:

For a series, keep a sound list (time range, sound, prompt, kept asset) and run it through one node in a workflow. The trailer playbook runs the whole process on one piece.

Common problems

What's next

Next, Transcribe and subtitle turns speech into a transcript, an SRT you can fix, and subtitles burned into the video. To make the score for a clip, see Generate music.