Apple released VisionOS 2.8 in mid-July, opening access to two developer-facing APIs that shift spatial computing from prototype territory into production-grade performance. Foveated rendering (rendering only the user’s point of gaze at full resolution, the periphery at lower quality) is now a public API, and hand occlusion prediction lets developers render virtual objects with correct depth ordering when hands pass in front of them — no more hands passing through solid objects.
Watch: Apple Vision Pro — Hands-On (YouTube)
Foveated rendering cuts rendering costs in half
Spatial computing demands continuous rendering at 120 FPS in two eyes at 2K resolution each, which is computationally brutal. Foveated rendering exploits how human vision actually works: we only perceive fine detail in a narrow cone centered on our gaze, and the periphery is much blurrier. VisionOS 2.8’s new API tracks where the user is looking and renders only that small area at full quality, dropping periphery rendering to half or quarter resolution. Results: complex scenes that would have run at 60 FPS now hit 120 FPS without quality loss in the area the user is actually watching.
Hand occlusion makes interaction feel solid
Virtual objects that pass behind users’ hands without visual precedence feel wrong — like the space isn’t real. VisionOS 2.8 predicts hand position 20 milliseconds in the future (compensating for camera latency) and automatically renders virtual geometry with correct depth ordering relative to hands. Developers no longer need to manually fiddle with layering or use heuristics to guess where hands are.