Adding sound to a silent video
Generated videos often come out silent. A bed described from the shot, a loop, and the Edit node lays it under the picture: the add sound to a video workflow does exactly that.
AI sound effect generator
Three sound effects made by ElevenLabs Sound Effects from text: looping rain on a tin roof, a creaking castle door, a stadium erupting. The full workflow, visible without an account.
Interactive demo
Real workflow, real results. Pan and zoom freely.
Use this workflowReal workflow: three Text nodes wired into three Sound nodes on ElevenLabs Sound Effects, with three different settings. Listen to all three.
At a glance
- What the workflow produces
- 3 audio tracks, 23 s in total
- Canvas structure
- 7 nodes wired by 3 connections
- Models wired in
- ElevenLabs Sound Effects
- Cost of one full run
- about 8 credits, roughly $0.10
The sound you are missing is never in the sound library: the door creaks, but not like that; the rain falls, but on a tile roof. ElevenLabs Sound Effects makes the sound you describe, in a few seconds and for a few cents. This workflow produces three, one per family: a looping bed, a one-off sound, a crowd.
A sound effect is described by what you hear, never by what you see. "A roof in the rain" is a picture; "heavy rain drumming on corrugated tin, water dripping into a puddle" is a sound. The three prompts of the demo are written that way, and they all end with "no music, no voices": without that ban, the model gladly adds them.
The three nodes do not share the same settings, and that is the lesson. The rain asks for ten seconds and the loop, to run seamlessly under a longer shot. The door asks for five seconds and a higher prompt influence. The crowd asks for a lower influence, which lets the model improvise. All three sounds cost ten cents in total.
How it works
- 1
Describe what is heard
Material, action, distance, surroundings: "heavy wooden door creaking on rusty hinges, then a dull thud against stone". Ban visual adjectives, they cannot be heard.
- 2
Set the duration
A one-off sound, three to five seconds. A bed, ten to twenty-two seconds, the model's ceiling. Too long for a short sound, and the model repeats it or drowns it.
- 3
Turn the loop on for a bed
Rain, wind, a murmuring crowd, an engine: the loop asks the model for a start and an end that meet. The edit then repeats it seamlessly.
- 4
Dose the prompt influence
High (0.7) for a precise sound you describe closely. Low (0.4 to 0.5) for an ambience where the model may invent the details. The demo shows both.
Variations and use cases
The same canvas, tuned for different needs: each variation is two or three lines of prompt away.
Sounds for a game or an app
A click, a notification, a coin picked up, a door opening: short sounds, high influence, two seconds each. Three nodes for three variants, keep the best.
The ambience of a podcast or an audiobook
Rain behind a scene, a murmuring cafe, a forest at night. Twenty-two seconds on a loop, at low volume under the voice. The prompt states the distance: "far away", "behind a window".
A sound impossible to record
A dragon, an explosion, an avalanche, a spaceship. Describe them with real sounds the model knows: roar, breath, rumble, cracking ice, turbine.
Sounds for a play or a show
Thunder on cue, the bell, the passing train. Each sound in its node, exported as MP3, ready for the sound desk. The whole set costs less than a coffee.
A sound effect, then cleaned
If the result has hiss or an artefact, the Enhance node in cleanup mode removes it. Rarely needed, but useful for a sound that will loop for a long time.
Common mistakes
Describing the picture instead of the sound
"A forest in autumn" cannot be heard. "Dead leaves crunching under footsteps, light wind in the branches, a crow far away" can. The model sees nothing, it listens to your sentence.
Forgetting to ban music and voices
The model tends to dress an ambience with a musical pad or a murmur of voices. One sentence at the end of the prompt is enough to stop it.
A duration too long for a short sound
A door creaking for fifteen seconds is fifteen seconds of creaking. Three to five seconds for a one-off sound, the duration is set on the node.
A bed without the loop
Without the loop on, the bed has a start and an end you hear when the edit repeats it. Turn it on for any sound that must run longer than its duration.
Recommended models
ElevenLabs Sound Effects
Seed Audio 1.0Frequently asked questions
How long is a sound effect?
Between one and twenty-two seconds, set on the node. For a longer ambience, turn the loop on and let the edit repeat the sound.
Does the model add music?
It does as soon as it is not told otherwise, especially on ambiences. Hence "no music, no voices" at the end of every prompt in the demo.
Can I describe a sound that does not exist?
Yes, that is even where it is best: a spaceship taking off, a creature breathing, a magic door. Describe it with real sounds: "turbine whine, deep rumble, metallic clicks".
How much does a sound effect cost?
Two or three credits depending on the duration, a few cents. The three sounds of the demo come to eight credits in total. The price shows on the node before you click.
What format does the sound come in?
MP3, downloadable from the node or wired straight into the Edit node. The file is a few hundred kilobytes.
Are the sounds royalty-free?
They are generated for you and exist nowhere else. ElevenLabs' terms allow commercial use on a paid account, which is what runs here.
What does the prompt influence setting do?
It tells the model how literally to follow your sentence. High, it produces exactly what is described, sometimes a bit stiff. Low, it improvises details around it, livelier but less predictable.
Can I do it from a phone?
Yes. The canvas runs on mobile, the prompt is typed in the Text node and the sound plays in the Sound node. The create on mobile guide explains the gestures.
AI sound effect generator
Use this workflowFree account, no card required. The workflow opens pre-filled in your canvas.