Skip to content

research · the field

A new way to capture the world

Capturing a real place and letting someone move through it, live, in a browser, is genuinely new. A few years ago it was not possible at all. The live demo is the first step. This page is where it is heading. It is direction, not a release calendar. What we have committed to next lives on the roadmap.

why it matters

The world becomes a place you can enter.

For a long time, the only way to put a real location in software was to rebuild it by hand or wrap a photo around your head. Capturing a place as it actually is, then driving through it, changes what software can hold. We think that is the start of a field, not a single product.

what is real today · the numbers

Measured, not promised.

Outcomes from the live demo, not a slide. These are the results. The methods that produce them stay in the lab.

5M+splats rendered live
~8scold link to driving
<2severy drive after
0installs, plugins, stores

in the lab right now

live today

One capture colliding with another

A car that is itself a scan drives inside a captured place and collides with it for real. Getting one capture to physically push against another, live in a tab, is hard. It works today.

live today

Street life that reads the space

People in the scene move with a real sense of the place: where they can go, what is in the way, when to get out of the car's path. It holds up while they walk and react.

in research

Captures that move

We have started research into captures that change over time instead of holding one frozen moment. Early, but it is the path to weather, traffic, and a day that turns.

benchmarks · the bar we hold to

Built for the machine you already own.

We do not benchmark on a render farm. The target is a laptop. Here is what we hold to, and how we measure it.

30 FPSthe bar, on integrated graphics
3hardware tiers, auto-detected
p50·p95·p99frame-time, logged every drive

Integrated graphics

Intel Iris / UHD, thin-laptop chips

Real-time drive at a reduced render scale, core detail held.

Apple silicon

M-series

Native resolution, high detail.

Discrete GPU

GeForce RTX, Radeon RX

Full detail, supersampled, with headroom to spare.

How we measure: every drive reports frame-time (p50, p95, p99), time to first pixel, and cold vs warm cache, segmented by GPU tier. We publish the bar we hold to, not a cherry-picked peak.

where this is heading

Places that are alive

Today a captured place is a held breath: still air, a fixed sun. The goal is a place that moves. Light that turns through the day, a street that feels lived-in. Driving through a moment, not a photograph.

A city, not a block

Right now you can drive a captured street. The aim is to keep widening that edge until it is a whole neighbourhood, then a city, and it still opens in a tab on the machine you already own.

Capture from anything

The real shift is when anyone can make a place. A walk on your phone becomes somewhere you can drive. The less skill the capture takes, the more of the world this reaches.

Places you can change

A capture you can only look at is a souvenir. One you can edit, relight, and build on is a stage. We want a scan to become something you can make things in, not just visit.

honest by design

Some of this is real in the demo today. Some of it is still ahead. I will not blur the line between the two. The roadmap says what is committed and when. This page is the longer view, and it is held to the same rule: no claims I cannot stand behind.

Working in this space, or want to build on it? I read every note.

[email protected]