Building Vision Therapy in VR: What a World Cup Final Taught Me About Dichoptic Training
Yesterday Madrid exploded. Spain beat Argentina 1-0 in extra time for the World Cup. Pure joy. But as a dev working in VR health, I watched Torres' winner differently: 120 minutes of elite visual processing. Saccadic tracking on through balls. Stereopsis under pressure as defenders closed in. Near-far accommodation flicking between a diving keeper and the far post. Most of us take this for granted. ~3% of adults don’t—they live with amblyopia, where the brain suppresses the weaker eye and depth, motion, and fine detail become a daily compromise.
Traditional patching still carries the baggage of low compliance and the old “critical period” narrative that left adults with few options. We built Amblyotube for Meta Quest (Unity + OpenXR) to reframe therapy as something people actually want to do. The core mechanism is dichoptic video playback: the lazy eye receives an alpha-blended version of the content while the dominant eye sees the complementary feed. Users simply select a video from a YouTube-style library—MrBeast clips, tech reviews, or last night’s football highlights—and the headset does the rest. Sessions run 30–40 minutes; no mini-games, no scoreboards. The therapy is a side effect of entertainment.
Delivering this required solving a stack of real-time challenges. We start with per-eye texture composition that never drops a frame; the content pipeline ingests a YouTube embed, streams it to a frame server, then renders dual-eye textures through custom shaders. Calibration covers individual IPD and suppression depth so the dichoptic blend is precise. For visual reinforcement we added a fusion cue system: a magenta circle for the lazy eye and a grey circle for the dominant eye, rendered on a fixed depth plane to drive vergence. An AI-driven MFBF pass uses on-device MediaPipe person detection to apply a lazy-eye sharpening filter when faces appear on screen. A Visual Accents shader layers a configurable yellow-green glow and red silhouette outline that pulses at 0.5–2 Hz, while the dominant-eye shader stack offers runtime sliders for Gaussian blur, contrast/brightness LUT, and opacity tuning. Safety is baked in: a hard 60-minute session cap, mandatory lazy-eye selection at first launch, and an age gate of 13+.
The bigger lesson is that compliance in VR health often hinges on content, not gamification. Users come for the videos they already love; the binocular training happens implicitly. Amblyotube is positioned as a wellness/educational tool that complements prescribed care; we are not a medical device. For fellow builders, remember that the headset’s dual displays are a largely untapped superpower for dichoptic and binocular interventions. If you’re curious, try it yourself:
https://www.meta.com/en-gb/experiences/amblyotube/25906906972338493/
Top comments (0)