Sound Design in Film: The Half of Cinema Your Audience Never Sees

There's an old line in post-production circles: picture is what the audience watches, sound is what they actually feel. It's a slogan, but it survives because it keeps being true. Cut a tense scene with a flat, honest recording of the room and it plays as a rehearsal. Cut the same frames with a low room tone, a distant dog, a chair leg that scrapes a fraction too loudly, and it plays as dread.

Nobody in the audience will ever say "great sound design." They'll say the film got under their skin.

This is the third piece in our post-production series, after the future of post-production and color grading, and it covers the part of the craft that clients most often discover they've underbudgeted. Here's what sound design in film actually involves.

What is sound design in film?

Sound design in film is the deliberate construction of everything you hear that isn't the original production audio, and, increasingly, the repair and reshaping of the production audio itself. A film sound designer decides what the world of the film sounds like: which sounds exist, how loud they are, how much space is around them, and, critically, what gets removed.

That last one is the part people miss. Sound design is subtractive as often as it's additive. A busy street scene with every real sound left in is noise. A busy street scene where the traffic drops away the moment the character sees the face they've been looking for, that's design.

It's also not the same job as sound mixing, though the roles overlap on small crews. The designer creates and shapes the material; the re-recording mixer balances it into a final soundtrack that holds together across a cinema, a laptop and a phone speaker. On a lean production, one person does both. On a feature, they're separate departments with separate deadlines.

The five stems of a soundtrack

Every finished mix is built from layers, "stems," in the trade, and it helps enormously to know which is which when you're giving notes.

Dialogue. Production sound from set, cleaned and edited, plus ADR (automated dialogue replacement) for any line that couldn't be saved. Dialogue is the priority stem: everything else is arranged around keeping it intelligible.

Foley. Human-performed sounds recorded to picture in a studio, footsteps, clothing rustle, a hand on a door, cutlery, a body hitting a floor. Named after Jack Foley, who developed the technique at Universal in the sound era. Foley exists because the microphone on set was pointed at the actor's mouth, not their shoes, and because a real footstep recorded in a real corridor almost never sounds as convincing as a performed one.

Ambience. Room tone, backgrounds, the continuous bed that tells you where you are. This is the cheapest layer to get right and the most damaging to neglect, a scene with no ambience under it feels vacuum-sealed, and the audience will feel it without knowing why.

Hard effects. The specific, synchronised events: a car door, a gunshot, a lift arriving, a phone. Sourced from libraries, recorded specially, or built from unrelated ingredients.

Music. Score and source music. Sound design and score have to be planned together or they fight, the most common cause of a muddy mix is a composer and a sound designer both filling the same frequency range in the same scene without knowing the other was there.

Foley: why we still record footsteps by hand in 2026

Of all the layers, foley is the one clients question most often. Why perform sounds that already exist in a library?

Because libraries give you a footstep, and foley gives you this character's footstep, the hesitation, the weight shift, the fact that they're carrying something in their left hand. Foley is a performance, not an asset. A foley artist watches the actor and moves like the actor, and the result syncs to the body language in a way a placed library file can't. Audiences don't consciously register it. They register its absence as "something feels off."

There's a practical argument too. A foley pass gives you a clean, isolated version of every physical sound in the film, which is exactly what you need when the film is dubbed into another language and the dialogue stem is replaced wholesale. Any film with international distribution ambitions needs an M&E (music and effects) track, and you can't build one from production audio alone.

Diegetic and non-diegetic sound and why the line matters

Diegetic sound exists inside the world of the film: the characters can hear it. A radio in the kitchen, a passing tram, a scream from the next room. Non-diegetic sound exists only for the audience: score, narration, a stinger under a cut.

The reason to keep the distinction clear isn't academic. The most effective moments in modern sound design come from deliberately crossing the line. A song that starts as score and turns out to be playing on a car stereo. A heartbeat that begins as the character's and continues after the cut to another room. Each of these lands because the audience has an unspoken model of which sounds are "real," and the film breaks it on purpose.

You can't break a rule the audience doesn't feel. Which is why we map the diegetic world of a project during pre-production, alongside locations and schedule — the same stage covered in our pre-production checklist.

Sound design starts on set, not in post

The single biggest determinant of a film's final sound quality is decided months before anyone opens a session file: whether the production treated location sound as a real department or as something to fix later.

Things that cost nothing on the day and cost a fortune afterwards:

• Thirty seconds of clean room tone at every location. Non-negotiable. It's the glue for every dialogue edit that follows.
• Wild lines. If a plane went over on the best take, grab the line again immediately, with the actor still in the emotional register. It is far cheaper than an ADR session six weeks later, and it always sounds better.
• Wild tracks of anything location-specific. The particular hum of that factory, the particular door of that flat. You will not find it in a library.
• Turning off the fridge. Then writing "FRIDGE OFF" on a piece of tape stuck to the fridge, so nobody turns it back on at lunch.

"Fix it in post" is real, and modern tools have made it more real than it used to be. It is still, without exception, the most expensive way to get a worse result.

The post-production sound pipeline

Once picture is locked, post production sound follows a fairly stable sequence:

• Spotting session. Director, sound designer and composer watch the cut together and mark every moment that needs sound, needs music, or needs silence. Silence is a decision, and it gets marked like any other.
• Dialogue edit. Every line is chosen, cleaned, de-noised, de-clicked and smoothed. Room tone fills the gaps. This is unglamorous and it is most of the work.
• ADR. Only what survives step 2 as unusable, plus crowd and non-sync additions.
• Foley. Recorded to picture, in passes: feet, moves, specifics.
• Design and effects. Building the non-literal material — the things that don't exist and have to be invented.
• Pre-mix and final mix. Balancing the stems, riding levels, placing sounds in the stereo or immersive field, and delivering to the loudness spec the platform requires.

That last point catches people out. There isn't one standard. European broadcast delivery generally targets -23 LUFS under EBU R 128; streaming platforms normalise closer to -14 LUFS; a theatrical mix is calibrated to a reference monitoring level in the room rather than a single number. A mix that's correct for a cinema will feel thin and quiet on a phone if nobody's made a separate pass for it. We budget for the deliverables list before we start mixing, not after.

Where AI genuinely helps — and where it doesn't

Our position on AI in sound is the same one we take across the studio and set out in AI as the assistant, not the artist: it's outstanding at the grind and useless at the judgment.

What it does well now. Dialogue isolation is the headline. Tools like Fairlight's Voice Isolation and Dialogue Separator in DaVinci Resolve can lift a usable performance out of a take that would have been a mandatory ADR call five years ago, traffic pulled out from under a line, air conditioning removed without the vocal turning to glass. Automatic transcription and speech-to-text search make finding one word across forty hours of rushes a ten-second job. Automated de-clicking, de-essing and dialogue level-matching handle the repetitive passes competently.

The compound effect is real: it moves hours out of the technical stage and back into the creative one. Which is the whole argument of our post-production series.

What it doesn't do. It doesn't decide that the room should go silent one frame before the door opens instead of one frame after. It doesn't know that this particular character's footsteps should be slightly too loud in this scene because she's pretending not to be afraid. It doesn't have taste, and sound design is almost entirely taste applied to technical material.

There's also an honest caveat about generative audio specifically. It's improving quickly and it's genuinely useful for texture, drones and abstract beds. It is not yet reliable for anything that has to sync to a body on screen, and we don't use it for foley.

How we approach it

We treat sound as a department from the first budget draft, not a line item added when picture locks. In practice that means a sound plan in pre-production, a location sound standard we don't negotiate down, and a post schedule where dialogue editing gets the weeks it actually needs rather than the days that are left.

It's not the expensive choice. It's the choice that stops the film needing to be rescued.

Working on a film, a music video or a branded piece where the sound has to carry the emotion? That's the work we like most, and if you're weighing up what cinematic production is worth to a brand, our breakdown of branded content ROI is a good place to start. Get in touch and tell us what you're making.