Lilly Tech Systems
Best Practices Advanced
Shipping avatar systems to production requires careful attention to performance, memory, accessibility, and user experience. This lesson covers the essential optimizations and patterns that separate prototypes from polished products.
Performance Optimization
| Technique | Impact | Implementation |
|---|---|---|
| LOD Switching | 50-70% triangle reduction | Use Ready Player Me LOD settings + Unity LOD Groups |
| GPU Instancing | Reduced draw calls | Enable on avatar materials for crowd scenes |
| Texture Atlasing | Fewer material switches | Use 512px atlas for mobile, 1024px for desktop |
| Animation Culling | CPU savings | Set Animator culling mode to Cull Completely when off-screen |
| Object Pooling | Eliminates GC spikes | Pool avatar GameObjects instead of instantiating and destroying |
Memory Management
Each avatar consumes memory for meshes, textures, and animation data. In scenes with many avatars, implement a loading budget: only load full-detail avatars for nearby characters and use simplified placeholders for distant ones.
Cache Strategy: Configure the Ready Player Me SDK cache with a maximum size (e.g., 50 MB) and LRU eviction. Pre-warm the cache for frequently used avatars during loading screens.
Accessibility
- Provide text alternatives for avatar speech (subtitles/captions)
- Support high-contrast mode for avatar name tags and UI elements
- Allow keyboard and gamepad navigation in avatar customization screens
- Respect user preferences for reduced motion by offering simpler idle animations
Testing Checklist
- Test avatar loading on low-bandwidth connections (throttle to 3G)
- Profile memory usage with 10, 25, and 50 avatars in scene
- Verify animation retargeting with diverse avatar body types
- Test lip sync accuracy with different voice pitches and languages
- Validate multiplayer sync with 100ms+ simulated latency