If you’ve spent any time looking at AR and MR headsets, you’ve probably noticed that they work in two completely different ways. On one side you’ve got devices like the Apple Vision Pro and Meta Quest 3, where you’re essentially watching a live video feed of the world through high-resolution cameras mounted on the outside. On the other, you’ve got optical see-through displays — HoloLens, Magic Leap, and waveguide-based smart glasses — where you’re looking through actual transparent optics with digital imagery layered on top.

Both approaches are called “AR.” Both blend digital and physical worlds. But the trade-offs they make are dramatically different, and if you’re evaluating hardware for a deployment or trying to make sense of the market, understanding the distinction matters a lot.

How video passthrough works

Video passthrough — sometimes called “video see-through” — is conceptually straightforward. Cameras on the outside of the headset capture the physical world in real time, and that video feed is displayed on the screens inside. You’re not actually looking through anything transparent; you’re watching a high-quality video of your surroundings.

The Meta Quest 3 does this with its colour passthrough cameras, which are good enough that most users find the transition between the virtual environment and the real world surprisingly seamless. Apple Vision Pro takes it further — its 12-camera array and dedicated R1 chip process the passthrough at 12ms latency, low enough that the disconnect between head movement and visual response basically disappears.

The key advantage here is flexibility. Because the “real world” is just a video input, the software can do anything with it — augment it, replace it, dim it, pause it. Mixed reality becomes truly mixed: a virtual object can cast a shadow on your physical desk, or a digital window can be anchored precisely to a wall. Tracking quality, depth sensing, and occlusion (where real objects correctly obscure virtual ones) are all far easier to implement well.

How optical see-through works

Optical see-through takes the opposite approach. The lenses are waveguides — thin layers of glass or plastic that bounce light using total internal reflection, projecting imagery into your field of view while remaining largely transparent. You see the physical world directly, with digital overlays on top.

Microsoft HoloLens 2, Magic Leap 2, and the more recent Xreal Air 2 Ultra all use this approach. It’s also what most smart glasses — from Google Glass onward — have used when they’ve included any display at all.

The appeal is obvious: you’re seeing the real world with your own eyes, at full resolution, with natural depth of field. There’s no latency, no camera artefacts, and no uncanny valley effect when looking at people. For applications where you need to maintain full situational awareness — surgery, warehouse picking, field maintenance — that direct visual connection matters.

Here’s the thing, though: waveguide optics come with real limitations that have proved difficult to engineer away. Field of view is typically 40–70 degrees in current commercial devices, compared to 100+ degrees for video passthrough headsets. Outdoor visibility in bright light is a persistent problem. And producing photorealistic, high-brightness digital imagery through a waveguide without compromising transparency is genuinely hard — the physics push back.

Why video passthrough is pulling ahead in 2026

If you look at where the market has gone, video passthrough has clearly taken the lead in the consumer and prosumer space. Apple Vision Pro, Meta Quest 3, and PlayStation VR2 all use it. The reason is mostly resolution and field of view: camera and display technology has improved faster than waveguide manufacturing, so the video passthrough experience on a £3,500 Vision Pro is genuinely impressive in a way that current optical see-through at a comparable price point isn’t.

There’s also a software ecosystem argument. Because video passthrough devices are essentially running everything in software, developers can iterate far faster on passthrough quality and spatial tracking. The gap between where passthrough AR was in 2022 and where it is today is significant. Waveguide hardware improvements are slower and more expensive.

That said, optical see-through still has real advantages for specific deployments. A warehouse worker wearing a HoloLens for eight hours needs to maintain full peripheral vision and spatial awareness in ways that make video passthrough genuinely uncomfortable for prolonged use. Industrial inspection, surgery, and any application requiring extended wear in safety-critical environments still favour optical designs.

What this means if you’re evaluating hardware

The practical question is always: what’s this for, and who’s wearing it?

For immersive consumer experiences, creative applications, and knowledge worker productivity — video passthrough is the better bet today. The hardware is further along, the software ecosystem is more mature, and the experience for typical 1–3 hour sessions is better.

For industrial deployments involving extended wear, full situational awareness, hands-free operation in bright outdoor environments, or staff who need to interact with the physical world while glancing at digital data — optical see-through remains the right choice. Magic Leap 2 and HoloLens 2 are both purpose-built for these scenarios, and they show.

The convergence bet is that waveguide technology eventually catches up to video passthrough in resolution and field of view, at which point you get devices that offer both modes or that are genuinely good enough at optical see-through to obsolete the video approach. Several labs are working on exactly this. But for now, it’s two different tools for two different jobs — and knowing which is which saves you from a very expensive mistake.