Advanced
Full-Body Avatar Best Practices
Optimization techniques, performance budgets, and production workflows for shipping full-body AI avatars in real-world applications.
Performance Budgets
| Platform | Polygons | Textures | Bones |
|---|---|---|---|
| Mobile VR | 10-15K tris | 1-2 textures, 1K | 50-75 |
| PC VR | 30-70K tris | 4-6 textures, 2K | 75-150 |
| Desktop game | 50-100K tris | 6-10 textures, 4K | 100-200 |
| Film/cinematic | Unlimited | Multiple 4K-8K | 200+ |
| Web (Three.js) | 10-30K tris | 2-4 textures, 1-2K | 50-100 |
LOD (Level of Detail)
- LOD 0: Full detail for close-up views
- LOD 1: 50% reduction for medium distance
- LOD 2: 75% reduction for far distance
- LOD 3: Billboard/impostor for very far distance
- AI tools like InstaLOD can automate LOD generation while preserving visual quality
Optimization Techniques
- Texture atlasing: Combine multiple textures into one to reduce draw calls
- Mesh merging: Combine clothing and body into fewer meshes where possible
- Animation compression: Use quantized keyframes and animation LOD
- GPU instancing: Render multiple avatars efficiently with instanced materials
- Occlusion: Don't render body parts hidden by clothing
Testing Checklist
- Visual quality at target platform's rendering settings
- Animation blending between states (idle, walk, run)
- Clothing penetration (body poking through clothes)
- Performance with target number of simultaneous avatars
- Loading time and memory footprint
- Cross-platform rendering consistency
Continue exploring: Learn about Real-Time AI Avatars for live face tracking and streaming, or AI Avatar for Video Calls for meeting applications.
Lilly Tech Systems