Sound design · Foley · Video SFX

How to Make Sound Effects That Work in the Mix

Foley, recording, synthesis, layering, picture sync, and generated SFX.

The fastest reliable way to make a sound effect is to define one event, choose a source, record or generate several clean variations, edit the strongest take, and judge it against the picture—not in isolation. Use Foley for performed movement, field recording for real objects and spaces, synthesis for controllable tones and noise, and generated SFX when you need fast candidates. Build only the layers that communicate contact, weight, material, distance and tail; preserve the raw source; then verify timing, dialogue space, rights and delivery format.
By
Sonilo Editorial Team
Published
Reading time
36 minutes
Foley artist records shoes on gravel with a shotgun microphone and field recorder
One real source, several takes, and a deliberate microphone position create more useful material than a chaotic finished layer.
Editorial and safety disclosure

Sonilo publishes this guide and appears in the generated-SFX section. Adobe did not sponsor or review it. DAW menus and generation features change. Never record impacts, weapons, vehicles, fire, pressurized objects, electricity, glass, chemicals, animals or public spaces in a way that risks injury, damage, alarm, trespass or unlawful recording. Substitute safe props, keep distance and hearing protection where appropriate, and use qualified production-safety supervision for hazardous work.

Last verified: August 21, 2026. Adobe Audition’s current documentation describes generated tones, sweeps, modulation and white, pink and brown noise as starting material for sound design. Sonilo’s current product and API pages distinguish text-to-SFX for standalone clips from video-to-SFX for effects aligned to visible events. Product capabilities, prices, plan rights and interfaces can change; verify them at use.

Choose the method from the job the sound must do

Foley

Performed to picture

Footsteps, cloth, hand contact and prop movement. Best when performance and micro-timing matter.

Field recording

Capture the real source

Doors, machines, rooms, weather and crowds. Best for authentic material and acoustic detail.

Synthesis

Build from signal

Tones, sweeps, noise, envelopes and modulation. Best for UI, sci-fi, transitions and controlled variations.

Generated SFX

Describe or supply video

Fast standalone or picture-aware candidates. Best when the brief is clear and every output will be reviewed.

These methods combine. A convincing heavy door might use a real latch for identity, a performed wood movement for timing, a pitched mechanical recording for mass and a short low-frequency synthesized layer for scale. The method is not the outcome: if one clean recording tells the story, do not add layers merely to look sophisticated.

01

Write a cue sheet before touching the recorder or prompt box

A useful cue describes what the audience should perceive, not only the object name. “Door” is vague. “Small interior wooden door, dry latch click, restrained close, three meters away, no slam, no music” defines material, action, intensity, perspective and exclusions. For video, add a time or frame reference and note whether the sound begins before motion, lands on contact or continues after the object leaves frame.

FieldQuestionExample
SourceWhat physically produces the sound?Leather sole on loose gravel
ActionWhat changes over time?One cautious step, weight transfer, lift
Material/scaleWhat is it made of and how large?Dry stones, adult weight, close detail
PerspectiveHow far and in what space?Two meters, small exterior courtyard
TimingWhere are onset, contact, peak and tail?Heel at 00:03:12:08; short debris tail
Mix roleWhat must remain clear?Dialogue has priority; no wide rumble
ExclusionsWhat should not appear?No second step, music, voice or birds
DeliveryWhat files are required?Clean mono source, designed mono, stereo room

Make a cue list for the whole scene and rank it. Hero cues carry story: a key turn, reveal impact, creature movement or product interaction. Support cues establish continuity: cloth, hand contact and subtle machinery. Beds establish a space. Optional sweeteners add scale or style. Intentional silence belongs on the list too; filling every cut with a whoosh makes important moments smaller.

02

Record useful source material, not a finished fantasy

  1. Choose a safe substitute. A heavy book, leather glove, celery, gravel tray or small metal mechanism can provide components without recreating a dangerous event. Never discharge, crash, break, burn or strike hazardous objects for realism.
  2. Control the room. Turn off avoidable fans and notifications, listen for traffic and electrical hum, and record room tone. A dead room gives flexible close detail; a real location gives perspective but less control.
  3. Place the microphone for the layer. Close placement captures detail; farther placement captures the object in a space. Aim to reduce unwanted reflections and handling noise. Record both when delivery may need perspective changes.
  4. Set conservative gain. Rehearse the loudest action and leave headroom. Clipped transients cannot be restored by lowering a fader. Monitor with headphones at a safe level and never normalize the raw file destructively.
  5. Perform variations. Record soft, medium and hard versions, changes in speed, object orientation and surface. Leave silence between takes and speak or log identifiers without contaminating the usable tail.
  6. Capture wild tracks. Record isolated latch, body movement, contact, debris and room decay. Separate components are easier to synchronize and remix than one chaotic performance.
  7. Back up and log. Preserve original files, sample rate, bit depth, channel configuration, microphone, recorder, prop, surface, performer, location, date, take and rights/consent notes.
Clean does not mean sterile.

A little friction, unevenness and room response can make a sound believable. Remove traffic, voices, hum or handling only when they interfere. Heavy denoising can create watery artifacts and erase the texture you wanted to record.

03

Turn a source into a designed effect

Work nondestructively on duplicates. Trim unusable setup, add short fades to prevent clicks and keep enough pre-roll or tail to reposition the sound. Align layers by listening and waveform, then make each layer answer a different perceptual question.

LayerPurposeTypical treatmentFailure to watch
IdentityRecognizable material or mechanismPrecise edit, light EQ, little processingLoses realism under effects
BodyWeight and scaleLower source, pitch shift, transient controlMud or speaker-dependent sub energy
DetailClick, debris, scrape, spark or textureHigh-pass, short envelope, close perspectiveHarshness or distracting repetition
MotionApproach, pass, acceleration or transitionEnvelope, pitch/filter movement, panningGeneric whoosh unrelated to action
SpaceDistance and environmentRoom recording, early reflections, reverb sendReverb printed too long or too wide
TailDecay and consequenceSeparate resonance, debris or ambience returnEnds at the cut instead of naturally

Shape before you stack. Use clip gain and envelope first. EQ should solve a reason: remove unusable rumble, make material clearer or create perspective. Pitch changes apparent size but extreme shifting exposes artifacts. Compression can control an unruly transient, yet over-compression removes impact. Reverb places a source in a space; render or print it only when the delivery requires a committed effect.

For synthesized effects, start simple. Adobe documents tones, frequency sweeps, modulation and multiple waveform shapes, plus white, pink and brown noise. A short tone with a falling pitch envelope can become a UI confirmation; filtered noise with a fast attack and curved decay can become air, steam or a transition; a low tone layered quietly under a real impact can add mass. Begin and end generated waveforms near a zero crossing or use micro-fades to avoid clicks. Avoid assuming that a named noise color automatically sounds like a finished natural event—it is raw spectral material.

04

Four practical sound-effect recipes

Footstep on gravel

  1. Record the shoe and surface that match character weight and movement; separate heel/contact from lift if timing is complex.
  2. Choose one performance with correct rhythm, then align contact to the visible weight transfer rather than the first leg motion.
  3. Add a restrained debris tail from another take only if the surface needs continuity.
  4. Adjust perspective with microphone distance, EQ and room—not only volume. Alternate variations so repeated steps do not machine-gun.

Door close

  1. Record latch, hinge/body and room tail separately. The pictured door may need a safe substitute with a similar material response.
  2. Place handle movement before the panel motion, body contact on the visible stop and latch click where the mechanism engages.
  3. Add low body only for perceived size; keep the recognizable latch near the center.
  4. Let the room return after contact. A tail cut at picture edit often sounds less real than a short overlap.

UI confirmation

  1. Generate or synthesize a short tone with a clear pitch relationship; keep the envelope compact.
  2. Add one quiet tactile click if the interaction should feel physical. Avoid musical phrases that compete with notification meaning.
  3. Test on phone speakers at low level and confirm it is distinguishable from error, warning and destructive-action cues.
  4. Export consistent variations and naming for hover, success, error and disabled states rather than randomly redesigning each event.

Cinematic impact

  1. Start with a sharp transient for timing, a midrange body for material and an optional low layer for scale.
  2. Move layers by milliseconds until they read as one event without smearing the contact.
  3. Reserve the biggest low-frequency and widest tail for the story’s largest moment. Loudness alone is not scale.
  4. Test in the full mix and on small speakers; if only sub-bass carries the effect, add audible midrange information.

05

Fit sound effects to picture frame by frame, then judge at normal speed

Mark four moments: onset when the cause begins, contact when surfaces meet, peak when the audience perceives maximum force, and decay when the world settles. They are not always the same frame. A swing begins before impact; a door creaks during movement; debris continues afterward. Align the most important transient precisely, then shape the lead-in and tail.

  1. Lock or version the picture so timing changes are traceable.
  2. Build separate tracks for Foley, hard effects, ambience and design sweeteners.
  3. Place hero cues first and listen to the scene without music.
  4. Add support only where it clarifies motion, material, scale, point of view or offscreen action.
  5. Check perspective across cuts: a close shot and wide shot should not sound identical by default.
  6. Restore dialogue and music, then automate level, frequency and space so each cue has a role.
  7. Review once frame by frame for placement, then at normal speed for perception and story.
  8. Watch the entire export outside the editor and inspect the first and last frames of every cue.

Do not synchronize every movement literally. A realistic scene contains continuous micro-sound, but a mix directs attention. If every hand gesture, camera move and cut receives a pronounced cue, the viewer hears the editing system instead of the world. Choose the few sounds that explain the image and let ambience connect the rest.

06

Use Sonilo where generation genuinely shortens the workflow

Use text-to-SFX when you need a standalone effect that you will place manually: “dry ceramic cup set on wood, close microphone, small room, no voice or music.” Use video-to-SFX when the visible timing is part of the request: Foley, ambience, impacts, UI events, transitions or motion cues aligned to the supplied cut. Sonilo’s AI-video sound-effects guide recommends naming visible sources, actions, materials, scale, perspective and exclusions, then reviewing timing before mixing.

Start fromUseReview before keeping
A written effect briefText-to-SFX candidateSource identity, duration, texture, noise, tail and similarity
A locked video clipVideo-to-SFX candidateEvent coverage, frame timing, invented sounds, perspective and silence
A repeatable product workflowVideo-to-SFX APIAuthentication, input authority, task state, downloadable asset, retries and records
A hazardous or impossible sourceSafe generation plus conventional designBelievability, context, client acceptance and no false provenance

Generated audio is a candidate, not proof. Preserve the prompt, authorized input reference, generation date, model/service information available to you, output identifier, chosen version, edits, reviewer and license evidence. Check current Sonilo licensing for personal, commercial, client, paid-media, product and distribution scope. Do not upload confidential or third-party footage without authority. Do not prompt to copy a living artist, protected character voice or identifiable recording; perform a creative and similarity review before release.

07

Quality control, naming and delivery

CheckPass conditionEvidence to retain
TimingOnset, contact, peak and tail support the visible action at normal speedPicture version, cue marker and reviewed export
Sound qualityNo clipping, unintended hum, clicks, denoise artifacts or phase cancellationRaw source, edit session and meters/listening notes
Story and mixHero cues read; dialogue remains clear; silence is intentionalFull mix, stems and reviewer approval
PerspectiveDistance, room, camera cut and point of view remain coherentNear/far variants or automation
Rights and safetySources, performances, locations, libraries and generated outputs are authorizedReleases, licenses, generation records and incident notes
Technical deliveryRequested format, sample rate, channel layout, handles and sync reference are correctManifest, hashes and playback test

Name assets for retrieval, not memory: PROJECT_SCENE_CUE_SOURCE_ACTION_PERSPECTIVE_TAKE_VERSION.wav. Keep raw, edited, designed and mastered stages separate. For video post, deliver clean SFX stems apart from dialogue, music and ambience when requested. Include handles so the editor can move a cue, and avoid lossy masters unless the destination specifically requires them.

08

Common failures and the smallest useful fix

ProblemCauseFix
Effect sounds genericObject-only briefAdd action, material, scale, perspective, timing and exclusions
Impact feels lateAligned to motion instead of contact/peakMove the primary transient; reshape lead-in and tail
Layered effect is blurryToo many transients or phase conflictChoose one timing anchor; mute redundant layers
Sound is huge alone, weak in mixEnergy sits under dialogue/music or only in sub-bassDesign audible midrange and arrange mix space
Repeated footsteps feel artificialOne sample with identical level/timingRecord variations and preserve performance rhythm
Noise reduction sounds wateryProcessing exceeds the contaminationReturn to source, use gentler cleanup or rerecord
Generated SFX invents eventsBroad prompt or ambiguous videoSpecify visible moments/exclusions, regenerate or edit manually
Export clicks at boundariesHard cut away from zero crossingAdd micro-fades and retain natural tail

Sources

Primary and first-party sources

Human review: A sound editor or Foley artist should reproduce the capture, edit, synchronization and mix workflow; production safety should approve risky props or locations; the project owner should verify releases and source licenses; delivery should inspect format, sync and playback. Sonilo should maintain current product, model, licensing, privacy, security, correction and generation-record evidence.

FAQ

Frequently asked questions

FROM A SOUND BRIEF OR A LOCKED CUT

Create sound-effect candidates, then shape them in your edit.

Use text for standalone SFX or start from video when visible timing matters.