Guide

Game stealth mechanics explained

Stealth gameplay is the art of moving through hostile space without being caught — and making that tension feel fair. Unlike action combat where success is measured in DPS, stealth success is measured in information asymmetry: the player knows more about guard routes than guards know about the player, until something goes wrong. That contract only works when rules are readable, feedback is immediate, and failure teaches rather than punishes randomly. NPCs sense the world through perception systems; this guide covers the player-facing side — visibility, sound, cover, disguise, detection UI, takedowns, and level design that rewards planning over twitch reflexes. Whether you are building a pure stealth title, optional ghost routes in an action game, or survival horror where hiding is the primary verb, these mechanics form the toolkit.

The stealth contract: readable rules, fair feedback

Players accept stealth when they can answer three questions at any moment: Can I be seen? Can I be heard? What happens if I am? Break any of these and frustration spikes — the guard who spots you through a wall, the footstep that is silent in one room and loud in the next for no visible reason, the instant game-over with no recovery window.

Good stealth systems expose their logic through world geometry and UI affordances: light pools on the floor, cover-highlights in a vision mode, a rising detection meter before combat triggers. The player should be able to predict outcomes before committing — that prediction is the puzzle. When perception is handled on the AI side, keep a clean boundary: stealth mechanics write stimuli (noise events, visibility samples); perception reads them and updates alert state.

Visibility: light, posture, and peripheral vision

Most stealth games model visibility as a function of exposure — how much of the player's silhouette is visible to a watcher — modulated by environmental factors:

  • Light level — discrete tiers (bright / dim / dark) or a continuous 0–1 value sampled from light probes or baked lightmaps. Standing in shadow reduces detection rate; carrying a lantern or muzzle flash increases it.
  • Posture — standing, crouching, crawling. Crouch typically shrinks the collision capsule, lowers eye height for line-of-sight checks, and reduces movement noise. Prone may grant full concealment behind low cover.
  • Movement speed — faster movement is easier to spot (larger animation silhouette, more motion in peripheral vision). Tie speed to both visibility and noise generation.
  • Peripheral vs central vision — detection accumulates faster in the center of an NPC's view cone than at the edges. Some games use separate cones: a narrow "instant spot" cone and a wider "suspicion" cone.

Day-night cycles shift the entire visibility landscape: routes safe at midnight become suicide at noon. Document light values per room in your level design bible so QA can verify consistency. For 2D games, visibility is often a simple ray from enemy eye to player center; for 3D, sample multiple points on the player mesh (head, torso, feet) so crouch behind a crate actually hides the head while feet peek out.

Sound: noise events, propagation, and budgets

Hearing is stealth's second axis. Every player action that could alert NPCs should emit a noise event with at least: origin position, loudness (radius or decibel equivalent), and a category tag (footstep, gunshot, glass break, voice). Guards within hearing range receive the stimulus; walls and materials attenuate it.

Common propagation models:

  • Radius falloff — simple circle; cheap, works for arcade stealth. Loudness = baseLoudness × materialAttenuation × distanceFactor.
  • Ray-bounced propagation — noise travels through open doorways; closed doors block or muffle. More believable for immersive sims.
  • Surface-dependent footsteps — metal grating vs carpet vs water. Players learn to read floor materials; inconsistency here destroys trust.

Assign each action a noise budget per encounter: sprinting across a metal catwalk should be a deliberate risk, not the default traversal speed. Distraction tools (thrown bottles, hacked speakers) create noise events at a remote origin — they work because NPCs investigate the sound, not the player, buying time for a flank. Sync audio playback with the event emission so players hear exactly what guards "hear."

Cover vs concealment: hard cover, soft cover, and occlusion

Military shooters distinguish cover (stops bullets) from concealment (hides you visually but not ballistically). Stealth games need the same clarity:

  • Hard cover — solid geometry that blocks line-of-sight entirely. Standing behind it means zero visual detection from that angle.
  • Soft cover — partial occlusion: waist-high crates, foliage, crowds. Detection rate is reduced but non-zero; moving while in soft cover may still alert observant guards.
  • Interactive cover — doors, lockers, vents, haystacks. Entering a hiding spot sets a IsHidden flag that perception skips until the player exits or the spot is searched.

Visual feedback matters: outline shaders on crouch-cover, icons for "press E to hide," or a vision-cone overlay in planning mode. Without feedback, players hug walls that look protective but are not tagged as cover in your collision layers. Tag cover volumes explicitly in the editor rather than inferring from mesh bounds at runtime.

Detection meters, alert states, and player feedback

The detection meter (question mark, eye icon, yellow-to-red bar) is the player's primary stealth UI. It translates continuous perception math into discrete anxiety. Typical alert tiers:

  • Unaware — patrol routine; no knowledge of player.
  • Suspicious — heard a noise or glimpsed movement; NPC investigates last stimulus location, not player position.
  • Alert / searching — confirmed contact or found body; active hunt using last-known position and squad communication.
  • Combat — full engagement; stealth phase ends (or shifts to escape/hide recovery).

Design principles for meters:

  • Ramp up faster than down — getting spotted should feel urgent; breaking line-of-sight should decay suspicion over seconds, not instantly.
  • Telegraph before commit — a filling yellow bar gives the player a window to break LOS, crouch, or use smoke. Instant red = feels unfair.
  • Per-NPC vs global — per-NPC meters support pack tactics (isolate guards); a global "alert level" suits base-wide lockdown scenarios.

Audio stingers (rising string, heartbeat) reinforce meter state without requiring the player to stare at UI. Pair with adaptive music stems that layer tension as alert rises.

Social stealth, disguise, and faction rules

Not all stealth is crouch-walking in shadows. Social stealth hides the player in plain sight: uniforms, forged credentials, blending with crowds, acting like an NPC (walking pace, no sprint in restricted zones). Hitman, Assassin's Creed, and immersive sims like Dishonored's masquerade missions use overlapping rules:

  • Outfit / disguise tokens — wearing guard armor grants access to guard-only areas; detected if you sprint, draw weapon, or enter officer-only zones.
  • Social zones — "employees only" spaces checked at doors by faction ID, not line-of-sight.
  • Incognito actions — list of actions that break disguise (running, climbing, carrying bodies, illegal items visible on back).
  • Witness memory — NPCs who saw you in the wrong outfit raise alarm even after you change clothes unless you eliminate or bribe them.

Social and physical stealth can coexist in one level: sneak through vents (physical) or walk through the front door in a stolen uniform (social). Offering both paths increases replayability and accommodates different player skill profiles.

Takedowns, non-lethal paths, and consequence

Stealth games often reward non-lethal completion — ghost ratings, bonus XP, narrative branches. That requires reliable takedown mechanics:

  • Melee from behind — requires proximity, correct angle (behind arc), and often an unaware target. Failed attempts should fail loudly, not silently whiff.
  • Choke / knock-out — timed hold with risk if target wakes early; body must be hidden or discovery triggers alert.
  • Sleep darts / gadgets — ranged non-lethal with ammo scarcity and wind-up time.
  • Body disposal — dragging bodies into closets, dumpsters, water. Unhidden bodies are stimuli that elevate base alert when discovered.

Lethal and non-lethal should share the same positioning requirements where possible so players are not punished for choosing pacifism with stricter geometry. Score and mission recap screens make consequences visible: "3 bodies found, 1 alarm raised."

Level design for stealth: routes, patrols, and pacing

Stealth levels are puzzles built from space and time. Core layout patterns:

  • Patrol loops — predictable paths with gaps the player can time. Vary loop length so waiting is never longer than ~8–10 seconds without alternative action.
  • Verticality — rafters, balconies, and vents create overhead routes that bypass chokepoints. Requires clear locomotion affordances (mantle, climb).
  • Chokepoints vs open yards — narrow corridors force timing; open areas force cover-to-cover movement or distractions. Mix both.
  • Optional objectives — side loot, intel, or secondary targets that tempt players into riskier paths.
  • Safe rooms — periodic low-tension spaces to plan the next segment; without them, constant alert fatigue sets in.

Blockout with placeholder guards early. Playtest with vision cones visible. If every route requires the same 12-second wait, add a vent, a distraction object, or a social path. Stealth that devolves into "save-scum until RNG favors you" is a level design failure, not a player failure.

Stealth in multiplayer and asymmetric modes

Multiplayer stealth (Dead by Daylight, Splinter Cell Spies vs Mercs, Among Us) shifts the problem: human opponents predict patterns AI does not. Key considerations:

  • Server-authoritative visibility — clients do not decide who is seen; server runs perception checks to prevent wall hacks.
  • Information fog — hide enemy positions until line-of-sight confirms; similar to fog of war in RTS games.
  • Role asymmetry — hunter vs hider kits with different sensors (motion trackers vs cloaking devices) keep roles distinct.
  • Communication limits — proximity voice or ping systems so stealth teams cannot omnisciently coordinate.

Accessibility and difficulty options

Stealth can gate players with slow reflexes or vision impairments. Common assists that preserve core fantasy:

  • Vision cone overlay — toggle to show guard sight lines (also a learning tool for first playthrough).
  • Extended detection decay — suspicion drops faster on easy mode.
  • Pause-and-plan — limited tactical pause to mark routes (not full turn-based unless genre demands it).
  • Colorblind-safe alert colors — do not rely on red/green alone; use icons and patterns.
  • Skip or simplify QTE takedowns — hold-to-complete instead of frame-perfect inputs.

See game accessibility for broader inclusive design patterns. Assists should be player-opt-in, not hidden penalties for using them.

Decision table: stealth model by game type

Game type Primary stealth axis Detection model Failure recovery
Pure stealth (Thief, Mark of the Ninja) Visibility + sound Gradual meter per guard Hide, escape, reload checkpoint
Immersive sim (Dishonored, Deus Ex) All axes + powers Tiered alert + squad radio Combat fallback, blink/teleport escape
Social infiltration (Hitman) Disguise + zones Incident-based suspicion Change outfit, blend crowd
Survival horror (Alien: Isolation) Sound + hiding spots Persistent hunter AI Locker, under-table, distract
Optional ghost route (action games) Simplified LOS Binary or short meter Instant combat if spotted

Anti-patterns that break stealth

  • Omniscient AI — guards run to your exact hiding spot without search behavior. Fix perception, not chase speed.
  • Inconsistent cover — visually identical props with different occlusion tags. Tag explicitly in authoring tools.
  • Binary fail states — instant mission fail on first glimpse with no meter ramp. Players need a recovery window.
  • Noise without feedback — loud actions with no visual/audio cue that guards might hear. Every noise event needs a player-facing signal.
  • Save-scum routing — only one viable path requiring perfect timing every time. Add alternatives or loosen windows.
  • Stealth as slow combat — waiting 30 seconds with no agency. Provide distractions, gadgets, or route changes.
  • Broken disguise rules — NPCs ignore you in restricted areas without explanation, then punish arbitrarily. Document faction rules.

Production checklist

  • Visibility function documented: light tiers, posture modifiers, peripheral falloff.
  • Noise event schema defined: origin, loudness, category, surface material.
  • Cover volumes tagged in editor; soft vs hard cover distinguished.
  • Detection meter ramp/decay tuned with playtest telemetry (time-to-spot histograms).
  • Alert state machine integrated with behavior trees or utility AI investigate/chase nodes.
  • Disguise / faction rules table for social stealth zones.
  • Body hiding and discovery stimuli wired to base alert level.
  • Stealth assists available in accessibility menu.
  • Level blockout reviewed with vision cones visible; every segment has 2+ viable routes.
  • Multiplayer perception server-authoritative if applicable.

Key takeaways

  • Stealth is an information game — readable rules and fair feedback matter more than raw difficulty.
  • Visibility and sound are the two primary axes; cover and disguise modulate them.
  • Detection meters should ramp before committing and decay slower than they rise.
  • Level design is half the system — routes, patrols, and safe rooms define pacing.
  • Separate player stimuli from NPC perception so both sides can be tuned independently.

Related reading