Engine selection

> Pick tools that match delivery target, interaction model, and team constraints. Engines serve the game type — not the reverse.

Fit questions (ask first)

1. Platform: Web, mobile, PC, console, VR?
2. Primary loop: Action/physics, turn-based, narrative branch, management/UI, hybrid?
3. Presentation: Full-screen canvas, DOM/UI chrome, or both?
4. Toolchain: No-build / ESM OK, or bundler + editor OK?
5. Authoring: Code-only, or designers need Twine/Ink/Godot/Unity editors?

Architecture patterns

| Pattern | When | Notes | |---------|------|-------| | Full engine shell | Game is the canvas/scene | Phaser, Godot, Unity, Kaplay as app root | | Renderer + custom logic | You want draw power, own gameplay | PixiJS, Three.js + your systems | | Hybrid shell + guest | Dense UI/text + occasional skill-checks | DOM/app shell; mount canvas engines in modals/viewports only | | Narrative runtime | Branching prose is the product | Ink, Twine; host chrome separately | | Content-as-data | Levels/events authored as packs | JSON/YAML + thin loader; engine optional |

Web — decision tree

What type of game?
│
├── Mostly DOM / panels / forms / text UI
│   ├── + small arcade/spatial challenges
│   │     └── Hybrid: custom shell + guest
│   │         Raw Canvas/WebGL → Kaplay → Phaser → PixiJS
│   └── + branching story
│         └── Ink (inkjs) or Twine export → host in DOM
│
├── Full-screen 2D game
│   ├── Full gameplay features (scenes, physics, input)
│   │     └── Phaser 4  (or Kaplay if you want lighter/faster prototype)
│   └── Mostly rendering / custom systems
│         └── PixiJS 8  (or Raw Canvas/WebGL if tiny scope)
│
└── Full-screen 3D game
    ├── Full engine / physics / XR
    │     └── Babylon.js
    └── Rendering-focused / lighter
          └── Three.js

Quick comparison (web & common exports)

| Tool | Type | Best for | Watch-outs | |------|------|----------|------------| | Raw Canvas / WebGL | 2D/low-level | Tiny games, learning, no framework tax | You own everything | | Kaplay (ex-Kaboom) | 2D toolkit | Fast prototypes, jam games | Less “full product” structure than Phaser | | Phaser 4 | 2D engine | Complete 2D features | Heavier; often bundled | | PixiJS 8 | 2D renderer | Performance, custom game code | Not a full gameplay framework alone | | Three.js | 3D renderer | Visuals, lightweight 3D | You add gameplay systems | | Babylon.js | 3D engine | Fuller 3D + XR | Heavier than Three for simple scenes | | Ink + inkjs | Narrative | Complex branching prose | Weak for real-time multi-entity sims | | Twine / Twison / TweeJS | Narrative | Educator-friendly branches | Export/host glue; not a physics engine | | Godot 4 | Full engine | 2D/3D indie, open source | Web export iteration cost | | Unity | Full engine | Large teams, multi-platform | Heavy for simple web UI games |

Editor-first web shells (Construct, GDevelop) fit visual prototyping; weaker when you need versioned code-first content pipelines.

Non-web defaults (see also platform skills)

| Target | Lean toward | |--------|-------------| | PC indie / open source | Godot 4 | | PC large team / multi-platform | Unity | | Mobile | See game-development/mobile-games (touch, stores, battery) | | VR/AR | See game-development/vr-ar (+ Babylon/Three on web) |

Anti-patterns

| Don't | Do | |-------|-----| | Choose Unity/Godot for a form-heavy browser tool | Prefer DOM/hybrid | | Force Ink to run real-time concurrent simulations | Use narrative tools for branches; custom/sim code for clocks & entities | | Use Phaser as “the whole app” when the surrounding UI is HTML | Prefer a hybrid guest viewport | | Optimize for WebGPU on day one | Ship WebGL; add WebGPU + fallback when needed |

When to Use

Use when choosing or comparing game engines/frameworks before implementation, especially for hybrid DOM+canvas or narrative-first products.

Limitations

• Does not replace platform skills (game-development/web-games, game-development/pc-games, …).
• Final choice still depends on team skill and shipping constraints.
• Stop and ask for clarification if required inputs, permissions, safety boundaries, or success criteria are missing.