Sound effects
SFX v3 for Video: From Script to Sound Effects
Turn a short video script into a timed sound-effects cue sheet. Write prompts, generate separate layers and align the results in your editor.
For a video, start with a sound-effects cue sheet: a short list of what the audience should hear and when it should happen. Generate candidates for those cues, then align them in your editor. A prompt can describe a sequence, but it does not replace the edit timeline.
This walkthrough uses an invented 15-second product film to explain sound prompts and timing. Bring the descriptions into the audio workspace to generate and audition variations.
Turn the script into a cue sheet
Imagine a small desk lamp being unpacked, assembled and switched on:
| Edit position | Picture | Sound cue | Asset approach |
|---|---|---|---|
| 0–3 seconds | A box is opened | Cardboard flap and light paper movement | Separate short foley clips |
| 3–6 seconds | Lamp base touches the desk | One contained contact | One-shot |
| 6–9 seconds | Lamp arm is adjusted | Small hinge movement | Short mechanical detail |
| 9–12 seconds | A finger presses the switch | Tactile click | One-shot aligned to contact |
| 12–15 seconds | Final product frame | Restrained reveal cue | Optional transition layer |
These times belong to the edit. Do not expect a model to put an event at exactly the third second because it is written in a prompt. When timing matters, separate the assets and place them yourself.
Decide which sounds deserve attention
Choose one foreground event for each key action. A close-up switch may need a clear click; opening packaging can stay quieter beneath narration. Leave room for dialogue or voiceover instead of filling every visible movement with a loud effect.
A stable room background can join cuts together. Keep it separate from object sounds so you can control it without changing the click or hinge. Background music is another layer and should be selected through an appropriate music workflow.
Write prompts for individual cues
Packaging:
A small cardboard box flap opening once, followed by light tissue paper rustling, close perspective in a quiet furnished room.
Desk contact:
A compact metal lamp base placed gently on a wooden desk, one soft solid tap with a very brief tail.
Switch:
One small tactile rocker switch clicking into place, crisp but quiet, close and dry.
Keep the room and perspective consistent across prompts. If the packaging sequence is hard to align, split the cardboard and paper into separate requests. The prompt guide explains how to simplify missing or competing events.
Generate a candidate, then test it against picture
Review the estimate before each request. Include unused alternatives in your credit budget; a completed file that does not fit the edit is still a completed generation.
Start by asking whether the audible action matches the picture. Then check the transient, pace and tail. If a click is right but the tail feels too spacious, try a drier description or trim a working copy. Changing several adjectives at once makes the next result harder to judge.
Align the sound in your editor
Place the useful attack at the visible action, then adjust by ear while watching the shot. Trim leading silence, make deliberate fades and check the sound across cuts. A two-part generated sequence can be cut into regions when the file allows a clean separation; otherwise generate individual assets.
Keep effects, ambience, narration and music on separate tracks. Start with intelligible narration, then bring effects into the balance. Avoid clipping when layers combine. Listen to the final mix on the kinds of playback devices your audience will use.
Downloading a WAV does not guarantee that a result is free of noise, and converting an MP3 does not restore information lost in the source. Use the actual formats available from your selected model and editor.
Keep a small handoff package
Save the cue sheet, chosen files, unedited sources and a note of model, prompt, date and relevant usage terms. Give files descriptive names such as lamp_switch_click_v02. For a client, distinguish the finished film from separately delivered source sound assets; those can raise different licensing questions.
Read the commercial-use guide before publishing. If files are missing, results fail or the wrong model appears, use troubleshooting. For a background layer, see the ambience guide, or choose another idea from the prompt library.