Engineering Dichoptic Vision: Why Project Phoenix is a Signal for VR Health
As a builder, you tend to look at leaked hardware specs through the lens of "what does this unlock for the software layer?" The recent Project Phoenix leaks from Meta—specifically the integration of dedicated prescription lens slots—might seem like a quality-of-life update for the average consumer. But for those of us engineering assistive technologies for visual coordination and amblyopia, it's a critical infrastructure signal.
For the 2–3% of adults living with lazy eye, the primary bottleneck in VR therapy hasn't just been the software, but the optics. To achieve effective dichoptic training, the brain needs high-fidelity, sharp separation between the two eyes. If a user has an uncorrected refractive error, the resulting blur compromises the neural stimulation required for neuroplasticity. When prescription support is native to the chassis, the compliance math changes: we move from "clunky workaround" to "seamless medical-grade practice."
At Seven Sports, we've been tackling this from the software side with Amblyotube. The core engineering challenge is implementing dichoptic rendering on the Meta Quest—essentially delivering two entirely different visual streams to each eye while maintaining a synchronized, YouTube-style content experience.
The Technical Approach
From an architecture standpoint, we aren't just splitting the screen. We are manipulating the visual pipeline per eye to force the brain to integrate information. We utilize Unity’s Universal Render Pipeline (URP) with single-pass stereo rendering, but we've implemented a custom shader graph to handle per-eye post-processing.
Our Dominant Eye Shader allows for real-time manipulation of contrast, brightness, and opacity. By applying digital occlusion (blurring the stronger eye), we create a digital patching environment that doesn't fully isolate the eye, encouraging binocularity rather than total suppression. The shader graph exposes these three parameters as a ScriptableObject that is updated each frame from a user‑configurable occlusion profile, so the same material can serve every session without a recompile.
To further stimulate the lazy eye, we've integrated AI-driven processing to identify human figures within the video stream. Once a figure is identified, the Lazy Eye Sharpener (MFBF) applies a high-frequency sharpening effect exclusively to the lazy eye's viewport. The detection runs on‑device and feeds bounding boxes into a ComputeBuffer; the sharpening pass reads that buffer and boosts edge contrast only inside the detected regions. We've also added a "flicker" or "jerk" stimulation to these figures—leveraging the human eye's natural biological bias toward detecting movement and light changes to trigger neural firing.
Because we stay on URP rather than migrate to HDRP, the per‑eye RenderFeature adds a single full‑screen pass per view, keeping draw‑call overhead low and preserving the single‑pass stereo path that Quest’s compositor expects. This trade‑off lets us ship a 72 Hz experience on the current hardware budget while leaving headroom for future eye‑tracking integration.
Solving the Fusion Problem
One of the hardest parts of amblyopia training is "fusion"—getting the brain to merge two disparate images. To solve this, we implemented a Magenta Focus Cue. We render a moving magenta circle for the lazy eye and a neutral grey cue for the dominant eye. Magenta was selected because it is statistically rare in natural video content, ensuring the cue remains visible across diverse YouTube backgrounds without being obscured. The cue is drawn as a full‑screen quad per eye; a sine‑wave animator drives its radius at the configured breathing frequency, so the cue pulses in sync with the rest of the visual rhythm.
To maintain user attention over a 30-to-40 minute session, we've added Visual Accents like a yellow-green highlight and red silhouettes around human figures. These aren't static; they operate on a programmable "breathing" rhythm (Hz), creating a pulse that prevents the brain from habituating to the stimulus. The accents are implemented as a second post‑process pass that reads the detection buffer, runs a Sobel outline filter, and modulates the outline color with a time uniform tied to the same breathing clock.
The Path Forward
We are building the software layer now so that when the hardware—like Project Phoenix—finally standardizes inclusive optics, the therapeutic ceiling rises. We're moving away from the era of red-blue glasses and toward a world where vision coordination is as simple as watching a video.
If you're building in the health-tech or accessibility space and want to chat about per-eye shader implementation or VR-based neuroplasticity, let's connect.
Experience the training tool here: https://www.meta.com/en-gb/experiences/amblyotube/25906906972338493/
Top comments (0)