Foley sound design starts with an action, not a collection of impressive noises. Identify what touches, moves or changes on screen, decide which actions need sound, then align each chosen sound with the visible event. A short cue sheet helps you keep those decisions consistent as the edit changes.
For AI-generated video, this is especially useful when a visually convincing clip arrives without the particular footsteps, clothing movement or prop contacts you need. Oakgen can be part of the asset-making workflow through its audio tools and sound-effects generation guide. This article covers planning and editing the resulting assets; it does not claim that generating a sound automatically synchronizes it to your video.
Foley, ambience and designed effects do different jobs
Adobe's sound-effects guide distinguishes action-timed Foley from environmental ambience and specially designed effects. A foot landing, a jacket brushing a chair and a cup being placed on a table are useful Foley cues. A continuous room hum describes the space. An exaggerated cinematic impact may communicate emphasis rather than a physically literal action.
The boundary matters because each needs a different edit. You can move a room-tone bed without tying its beginning to a hand movement. Move a cup contact away from the visible landing and the relationship becomes confusing.
Before collecting sounds, choose a style: naturalistic, understated or deliberately exaggerated. Do not alternate those styles accidentally. A tiny ceramic cup does not need a cinematic explosion just because the sound is available.
Make a first pass with the music muted
Watch the locked or nearly locked picture once without adding anything. Write down the actions you actually see. If the original clip already contains useful audio, listen to that separately before adding duplicates.
On the second pass, mark three kinds of event:
- A contact with a clear moment, such as a heel touching the floor.
- A continuing movement, such as fabric shifting during a turn.
- A change that needs a supporting cue, such as a lid releasing.
A silent action is not automatically a defect. Keep the cues that help the audience understand weight, material, distance or a story beat. Leave out detail that competes with narration or makes a quiet scene unnaturally busy.
Copy this cue-sheet structure
The following is an illustrative plan for a fictional product clip, not a tested Oakgen output. Replace the times after checking your own edit.
| Cue | Picture time | Visible event | Sound requirement | Edit decision | Check |
|---|---|---|---|---|---|
| F01 | 00:02.200 | Hand lifts a leather pouch | Brief soft leather movement | Keep beneath narration | No zipper sound unless visible |
| F02 | 00:04.600 | Metal clasp closes | One small metal contact | Align contact transient | Avoid extra clicks |
| F03 | 00:07.100โ00:08.300 | Pouch slides across wood | Short textured movement | Match movement duration | Stop when object stops |
| F04 | 00:10.000 | Product is held still | No new action sound | Leave silent | Preserve room ambience |
Add the source filename, license note, take number and approval status to your working copy. Separate โrequestedโ from โapprovedโ: a generated file is a candidate asset until you have listened to it in context.
Use stable cue IDs when the picture changes. If a cut shifts F02 later, update its time rather than renaming every subsequent sound. That makes a handoff easier for an editor who already has the earlier cue sheet.
Choose how to obtain each sound
Record it when you have a suitable object and need close control over a sequence of movements. A simple hand performance may be easier to time than searching many unrelated files. Listen for unwanted room sound and handling noise before committing to the take.
Use a licensed library when an appropriate clean recording already exists. Check the actual license for your intended use and retain the source. A search result calling something free does not establish commercial permission.
Generate a candidate when you need a particular material or texture that is inconvenient to record. Describe one action at a time, then audition it. Generation can produce an unwanted second impact, an unsuitable room sound or a different sense of scale. Those are acceptance checks, not proof that another attempt will fix the problem.
For F02, an illustrative request is: โOne small metal clasp closing on a leather pouch, close perspective, dry indoor sound, no speech or music.โ It describes the needed asset without promising a precise contact time inside the returned file. Use the current model form and displayed credit estimate; generation uses credits, and model availability and limits can change. Do not generate repeated variants before checking whether a usable contact already exists.
Align the event inside the file
Put the picture and the sound on separate tracks in an editor that supports precise timeline placement. Find the important event inside the sound: the short contact in a clasp recording, rather than the beginning of the file's silence.
Move that event against the visible contact. Play across the edit at normal speed, then inspect a shorter region. If the recording contains a small preparatory movement, let it occur before the contact only when it fits the image. Trimming away everything before a transient can make the cue sound abrupt.
For continuing movements, match both the start and the stop. A scrape that continues after the pouch has stopped may be more distracting than a slightly imperfect material match. If stretching or heavy processing damages the texture, choose a shorter section or a different recording instead.
There is no universal offset that makes every sound feel right. Perspective, distance and deliberate stylization can affect the relationship. Begin with the visible event and judge the result in the actual scene.
Mix for the action's importance
Bring narration and ambience back before approving the cue. Set the action sound low enough that it belongs in the scene, then raise it only as far as the story needs. A cue that sounds satisfying in isolation can be intrusive under a spoken explanation.
Listen for three competing problems. First, duplicate sounds: existing clip audio and a replacement can make one contact sound like two. Second, inconsistent perspective: a close dry footstep can clash with a distant wide shot. Third, buildup: many individually quiet cues can crowd the mix when combined.
If music masks important speech, handle that as a separate audio-ducking decision. Do not keep increasing the Foley level to solve a narration-versus-music problem. Check the final export as well as timeline playback, because the delivered file is what the viewer hears.
Approve the cue sheet, not just the files
Before delivery, watch the whole sequence and confirm that every approved cue still describes the current picture. Check contacts, movement endings, narration clarity and overall consistency. Record rejected cues too: โF04 intentionally silentโ prevents someone from filling the gap later merely because it looks unfinished.
Keep the clean source, edited cue and final mix identifiable. If the video is recut, you can reposition the approved cue instead of generating or recording it again. For a broader asset-selection workflow, use the AI sound-effects guide; for picture planning, see the storyboard workflow.
To make supporting assets, create an Oakgen account, review the available audio tools and check current plans. Choose sounds against your cue sheet, then perform the timing and mix review in your chosen editor. Account access, generation credits and any editor export requirements are separate checks.


