# Sound Design Foley > Provider-independent sound design and Foley direction for generated media. Use when an agent must plan, prompt, source, generate, edit, mix, hand off, repair, or QA sound effects, Foley, ambience, impacts, transitions, UI/product audio, emotional sound design, or accessibility-safe captions for videos, ads, trailers, animation, avatar clips, explainers, games/trailers, product films, and social e… - **Type:** Skill - **Install:** `agentstack add skill-calesthio-generative-media-skills-sound-design-foley` - **Verified:** Yes — security-reviewed for prompt injection and unsafe behavior - **Seller:** [calesthio](https://agentstack.voostack.com/s/calesthio) - **Installs:** 0 - **Category:** [Developer Tools](https://agentstack.voostack.com/c/developer-tools) - **Latest version:** 0.1.0 - **License:** MIT - **Upstream author:** [calesthio](https://github.com/calesthio) - **Source:** https://github.com/calesthio/generative-media-skills/tree/main/skills/production/audio-craft/sound-design-foley - **Website:** https://github.com/calesthio/OpenMontage ## Install ```sh agentstack add skill-calesthio-generative-media-skills-sound-design-foley ``` Requires the [AgentStack CLI](https://agentstack.voostack.com/docs/cli). Works with Claude Code, Cursor, and any MCP-compatible agent. ## About # Sound Design and Foley Direction Use this skill to make generated media sound intentional instead of merely accompanied by music. Treat sound as story, timing, texture, space, and information. The goal is not to fill every frame with effects; it is to decide what the audience must feel, notice, understand, or believe at each moment, then design sounds that support that job. ## Evidence stance Documented facts below are drawn from professional and standards sources verified on 2026-07-10: - Berklee describes Foley artists as devising and recording everyday sounds for films, television, and video games, and notes that Foley work begins with a spotting session that catalogs needed effects before recording to picture. - ITU-R BS.1770-5 specifies algorithms for measuring programme loudness and true-peak audio level; many platform specs derive from or reference ITU loudness methods. - EBU R 128 recommends -23 LUFS for broadcast normalization in its domain, with EBU Mode momentary, short-term, and integrated metering. - W3C WCAG guidance for prerecorded captions says captions include dialogue plus meaningful non-speech sound information. - Section508.gov similarly says sounds needed to understand and enjoy synchronized media should be captioned, and that caption style should be consistent. - Creative Commons licenses can permit reuse under defined conditions such as attribution, noncommercial limits, share-alike, or no-derivatives. These conditions must be checked per asset. - YouTube Audio Library terms, verified on 2026-07-10, describe its music and sound effects as copyright-safe for YouTube use, while warning that only Audio Library assets are known to YouTube to be copyright-safe and that YouTube does not provide legal guidance for off-platform use. - Freesound, verified on 2026-07-10, hosts Creative Commons-licensed sounds; some sounds cannot be used commercially and many require attribution. Empirical observations are practical listening checks and workflow tests an agent can perform on the actual cut: muting layers, playing small speakers, measuring peaks, checking sync frame by frame, and comparing against references. Production heuristics are experienced working rules. They are useful defaults, not universal laws. Override them when the brief, distribution spec, genre, accessibility need, or director preference requires it. ## First pass: spot the cut before creating sounds Do a spotting pass from picture and script before prompting a sound provider, searching a library, or writing an edit plan. Create a spotting list with one row per audible event or bed: - timecode in/out or frame range; - visual trigger, narrative beat, or interaction; - sound family: dialogue-support, Foley feet, Foley cloth/moves, Foley props, designed effect, impact, whoosh/transition, ambience/room tone, UI/product, creature/vehicle/magic, musical sting, silence; - diegetic status: inside the world, subjective/internal, interface/brand layer, or editorial punctuation; - sync strictness: frame-locked, loose sync, rhythmic, background, or offscreen; - perspective: macro/close, normal, distant, offscreen left/right, occluded, through wall/glass/helmet, underwater/phone/radio; - material clues: shoe, floor, cloth, metal, paper, glass, plastic, liquid, skin, wood, dust, sand, snow, electronics; - emotional job: trust, danger, speed, luxury, intimacy, comedy, scale, surprise, calm, unease; - rights/source plan: record, synthesize/generate, licensed library, user-owned, public domain/CC-compatible, or needs clearance; - mix priority: foreground, midground, background, subliminal, or mute. Do not treat the spotting list as a shopping list where every visible action needs a sound. In short ads and social edits, too many literal sounds can fight the voiceover and music. Prioritize events that clarify causality, sell materiality, heighten rhythm, or carry emotional information. ## The sound families ### Foley feet Use Foley feet for character presence, weight, pacing, surface, and scale. Specify: - shoe or foot: sneaker, boot, heel, bare foot, paw, robot foot, tiny toy feet; - surface: concrete, wet asphalt, dry leaves, polished wood, marble, carpet, gravel, snow, hollow metal; - gait: cautious, hurried, confident, exhausted, comedic tiptoe, heavy landing; - distance and room: close dry step, hallway slap, warehouse echo, distant upstairs thump; - sync pattern: exact steps, partial sweeteners, or abstract rhythm. Heuristic: when generated visuals show walking but the soundtrack is mostly narration, choose two or three anchor steps rather than a full step for every footfall. ### Foley cloth and moves Use cloth/move sounds for human intimacy, performance, and animation believability. Include jacket rustle, backpack strap, cape sweep, leather creak, armor jingle, hair movement, sleeve brush, or body shift. Heuristic: cloth should usually be felt more than noticed. If the viewer says "nice cloth sound," it is often too loud unless the fabric is the product. ### Foley props and hand interactions Props sell contact. Specify object size, material, handling force, and hand state: - ceramic mug set down gently on wood; - phone lifted from desk, glass-on-wood tap plus tiny finger slide; - cardboard box opened with tape peel and flap flex; - premium pen click, cap twist, and paper drag; - sword draw, key turn, latch, zipper, dish rattle, tool clack. For product films, keep prop sounds clean and elevated. A luxury product usually wants fewer, more detailed sounds: micro-clicks, precise snaps, soft leather, smooth magnetic closures, controlled low-end sweeteners. ### Ambience and room tone Ambience creates place and prevents dead air. Design ambience in layers: - base bed: room tone, city air, office HVAC, forest bed, vehicle cabin, distant crowd; - local details: neon buzz, bird, elevator ding, printer, distant siren, wind against window; - time/weather: morning birds, night insects, rain roof, winter muffling, summer street; - perspective: open exterior, small room, cavern, car interior, phone speaker, underwater; - motion: static bed, evolving bed, passing element, entering/leaving space. Heuristic: keep a very low-level continuous bed under scenes unless the intended effect is silence. Generated video often feels synthetic when every sound is an isolated one-shot with no air around it. ### Designed effects Designed effects are not merely recordings; they are layered sounds that express a concept. Use them for sci-fi interfaces, magic, glitches, reveals, brand mnemonic sounds, abstract data flows, power-ups, drones, monsters, trailer moments, and emotional transitions. Build a designed effect as layers with jobs: - transient: click, snap, hit, spark, consonant-like start; - body: tonal hum, motor, roar, whoosh, texture, pressure; - tail: reverb, delay, debris, shimmer, reverse, decay; - sub/weight: low thump or boom used sparingly; - detail: grit, servo, sparkle, air, particles; - perspective filter: muffled, phone, underwater, through helmet, far away. Heuristic: if a sound is meant to feel huge, do not only add low end. Add contrast: small pre-click, silence before impact, wide air, debris tail, and a short drop in competing layers. ### Impacts, hits, and trailer punctuation Impacts must follow picture energy and edit rhythm. Specify size, material, frequency range, and tail: - soft UI tap, dry and close; - medium punch, skin slap plus cloth movement, no exaggerated gore; - product reveal hit, tight low thump plus elegant glass shimmer; - trailer impact, sub drop, metal slam, air burst, long reverb tail; - comedic impact, cartoon pop, hollow boing, or deliberately undersized knock. Heuristic: reserve the largest impacts for the few moments that truly need them. A 30-second spot with 12 maximum hits becomes fatiguing and loses hierarchy. ### Transitions and whooshes Transitions connect edits and motion. Match the whoosh to what the viewer sees: - camera whip: airy stereo whoosh, fast attack, short tail; - object pass: tonal Doppler or material-specific air; - graphic slide: clean synthetic swish; - glitch cut: digital tick, bit-crush, data tear; - luxury dissolve: soft breath, silk sweep, tonal shimmer; - horror reveal: reverse swell into low impact. Heuristic: transition sounds should often start before the visual cut and resolve just after it. That anticipates movement and makes edits feel motivated. ### UI and product sounds For app demos, explainers, SaaS ads, device launches, and product films, UI sounds must be informative without becoming toy-like. Decide: - interaction class: tap, confirm, error, hover, drag, snap, completion, loading, notification; - brand personality: warm, clinical, playful, premium, secure, fast, human; - frequency zone: avoid masking narration; keep harsh 2-5 kHz ticks controlled; - system consistency: related actions share a palette but differ in pitch, envelope, or texture; - accessibility: do not rely on sound alone to convey essential state in the final video. Heuristic: a brand UI palette can be built from three base gestures: a short affirmative tick, a soft negative/blocked tone, and a completion flourish. Vary them instead of inventing unrelated sounds for each screen. ### Emotional and subjective sound Use subjective sound for memory, fear, wonder, focus, intoxication, stress, relief, or point-of-view changes. Common moves: - narrow bandwidth, heartbeat, tinnitus tone, breath, muffled world for panic or impact; - warm room tone, soft cloth, close breath, low noise floor for intimacy; - high shimmer, airy rise, harmonic bloom for wonder; - brittle ticks, distant metallic groan, reversed texture for unease; - silence or near-silence before revelation. Heuristic: emotional design works when it is motivated by the character or edit, not pasted over the whole sequence. Use it as a lens that turns on and off. ## Prompt planning for sound generation providers Provider tools differ, but good sound prompts usually include the same production facts. Provide the generator or sound designer with: - exact duration target and whether it may loop; - sync cue and visible action; - source identity and material; - performance: gentle, violent, playful, precise, clumsy, hurried, heavy; - perspective: close mic, medium, distant, left/right, stereo width, occluded; - environment: dry studio, tiled bathroom, office, forest, warehouse, street, underwater; - layer role: transient, body, tail, ambience bed, UI cue, impact, sweetener; - emotional adjectives tied to audible properties: "tense" should become low rumble, brittle high ticks, sparse air, or similar; - avoid list: no music, no dialogue, no crowd, no recognizable brand jingle, no copyrighted melody, no clipping, no excessive reverb, no tonal pitch if it should be neutral; - file handoff: sample rate, mono/stereo, loopable, stems, handles, loudness target if known. Weak prompt: "Add sci-fi sounds." Better prompt: "Create a 1.2-second close, dry, non-musical sci-fi confirmation sound for a premium medical app. It should start with a tiny glassy tap, bloom into a soft warm electronic chime, and end cleanly with no long reverb. Avoid gamey arcade tones, alarm beeps, voices, and melodies recognizable as a tune." For long scenes, do not ask for one giant mixed sound unless the provider is specifically built for full-scene audio. Ask for separate assets or stems: ambience loop, key Foley hits, transitions, impacts, UI cues, and designed sweeteners. Stems are easier to sync, license, caption, and repair. ## Layering and perspective Every important sound should answer three questions: 1. What physically caused it? 2. Where is it relative to the camera/listener? 3. What story emotion should it add? Use material and distance to avoid generic results: - Glass is bright, short, brittle, and can ring. - Wood is warm, hollow, knock-like, and varies by size. - Plastic is lighter, duller, sometimes squeaky. - Metal can ping, scrape, rattle, ring, thud, or scream depending on mass. - Fabric is broadband, soft, noisy, and useful for human movement. - Liquid has bubbles, splashes, droplets, viscosity, and container resonance. - Snow, sand, gravel, leaves, and mud are defined by granular texture and compression. Perspective controls credibility: - Close: more transient detail, dry texture, less environment, stronger stereo placement if appropriate. - Medium: natural room reflections and less microscopic detail. - Distant: less high end, more reverb/air, softer transients, often mono-ish or diffuse. - Occluded: low-pass filtering, muffled attack, room or barrier coloration. - Macro/product: exaggerated tiny detail, very low noise floor, precise timing. ## Edit and mix handoff When handing off to an editor, mixer, or automated mix stage, provide: - spotting list with timecodes and priorities; - asset manifest with filename, source/license, prompt or recording notes, duration, sample rate, channels, and intended use; - stems grouped as DX/dialogue/narration, MX/music, FX, Foley, ambience, UI/product, designed sweeteners, and optional M&E when localization is expected; - sync notes for frame-locked events and acceptable slip range; - loop instructions for ambience beds; - alternate versions for risky sounds: dry/wet, loud/soft, realistic/stylized; - loudness and true-peak target requested by the platform or delivery spec; - accessibility notes: which non-speech sounds require caption labels; - known risks: temporary library audio, unverified rights, possible masking, possible harshness, or client approval needed. For localization, keep dialogue/narration separate from music and effects. If the final may be dubbed, avoid baking language-specific UI narration or verbal sound effects into the FX stem. ## Loudness, safety, and accessibility Do not invent one universal loudness target. Ask for or infer the delivery context, then follow that platform, broadcaster, client, or festival spec. If no spec exists, make a conservative web/social rough mix and document that it is not a formal delivery master. Use ITU-R BS.1770-style loudness and true-peak measurement when possible. EBU R 128 is relevant for many broadcast workflows; Netflix and other platforms publish their own delivery specs. These are not interchangeable. Safety checks: - avoid clipping and intersample peak issues; leave true-peak headroom appropriate to the spec; - avoid sudden painful high-frequency spikes, especially in headphones; - avoid excessive sub-bass that collapses on phones or distorts small speakers; - check dialogue/narration intelligibility with music and FX active; - check headphone and phone-speaker playback, not only studio monitors; - avoid using alarms, sirens, emergency alerts, or medical-device-like beeps unless the story requires them; - avoid startle spikes in content for children, wellness, education, or accessibility-sensitive contexts unless explicitly intended. Caption meaningful non-speech audio. Include sounds that carry story, timing, warnings, speaker identity, music mood, offscreen action, or emotional information. Keep caption labels concise and consistent, for example: "[door unlocks]", "[soft notification chime]", "[distant thunder]", "[music swells]". Do not caption every tiny Foley rustle if it adds no meaning. ## Rights-safe sourcing and generation For each sound, record provenance. A useful rights ledger includes: - filename and short description; - source type: recorded by project, generated by provider, purchased library, user-owned, platform library, CC/public-domain source; - license or terms URL; - attribution text if required; - commercial-use status; - modification/adaptation status; - territory/platform limits; - verification date; - whether the asset is approved for final or temporary only. Prefer: - original recorded Foley when feasible; - generated sounds whose provider terms permit the intended commercia … ## Source & license This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code. - **Author:** [calesthio](https://github.com/calesthio) - **Source:** [calesthio/generative-media-skills](https://github.com/calesthio/generative-media-skills) - **License:** MIT - **Homepage:** https://github.com/calesthio/OpenMontage Install and usage instructions live in the source repository linked above. ## Pricing - **Free** — Free ## Security capabilities Automated source analysis of v0.1.0 — what this tool can access: - **Network access:** no - **Filesystem access:** no - **Shell / process execution:** no - **Environment & secrets:** no - **Dynamic code execution:** no *"Yes" means the capability is present in the source — more access means more to trust, not that it is unsafe.* ## Versions - **0.1.0** — security scan: passed — Imported from the upstream source. ## Links - Listing page: https://agentstack.voostack.com/l/skill-calesthio-generative-media-skills-sound-design-foley - Seller: https://agentstack.voostack.com/s/calesthio - Browse the marketplace: https://agentstack.voostack.com/browse --- Listed on AgentStack — the marketplace for AI agent skills and MCP servers. Every listing is security-reviewed. Creators keep 70%.