Lumen

MetaLoom // Lumen is the spatial media engine. You hand it video of a place; it reconstructs the place as a three-dimensional Gaussian-splat scene, keeps every scene in a versioned store, and lets you find places again — with a sentence, with a photograph, or with a new capture of the same location.

Lumen is a commercial offering. Unlike the open source MetaLoom platform, it is licensed per deployment and ships as a container image for your own GPU hardware — see the product page for what it does and whom to talk to.

The Lumen web console showing the use case overview tab with status badges
Figure 1. The web console’s use-case overview — each capability with its live, honest status

What Lumen is for

Video is the easiest way to capture a place and the hardest way to find one again. A folder of clips has no notion of where — two recordings of the same café share nothing a filename search could discover. Lumen’s answer is to make the place itself the unit of storage: captures are reconstructed into scenes, scenes are indexed, and every question — "where was that?", "have I been here before?", "does this footage show the same room?" — is answered against reconstructed geometry, not against file metadata.

Three properties distinguish it:

  • Real reconstruction. Scenes are trained radiance fields rendered as Gaussian splats, not photo carousels. A scene can be orbited interactively in the browser, rendered from any viewpoint, and described in words.

  • A store, not files. Scenes live in a write-ahead log with quantised primitives, spatial ordering, compaction, level-of-detail pyramids and published quality tiers. Any scene can be read at any point in its history.

  • Honest retrieval. Search is calibrated against measured score distributions: a place that is not in the corpus is refused, not approximated. Merging two captures of one venue requires geometric verification, never mere similarity.

Where to go next

Concepts

How a clip becomes a scene: the frame gate, pose recovery, optimisation, decomposition — and how the store keeps and serves the result.

The web console

A screenshot tour of the browser interface: search, the 3D splat viewer, storage insight and the ingest queue.

Running Lumen

The container image: GPU requirements, volumes, fetching the encoder weights and serving the console.