Marble World-Generation Audit: What I Actually Measured
World Labs' Marble turns a single photo into an explorable 3D world. Vendors make claims about commercial usability and dimensional accuracy — I ran the actual trial: two real generations, a byte-level provenance scan, and a real physical measurement. Here's what held up and what didn't.
The question this demo answers
“If a vendor tells you their AI-generated 3D reconstruction is commercially usable and dimensionally accurate, do you take that on faith — or do you check?”
Commercial rights & ToS
Checked against the actual terms, not a summary of them.
Run the audit to reveal.
Provenance scan
Byte-level scan of every exported file for C2PA/XMP/EXIF/glTF-extras/PLY-comment markers.
Run the audit to reveal.
Metric-scale accuracy
A real physical measurement, compared to the reconstruction — not a spec-sheet number.
Run the audit to reveal.
Same audit, second vendor: NVIDIA Cosmos
Cosmos makes physics-aware video for robotics, not explorable 3D worlds, so this isn’t a like-for-like product comparison. It tests whether the audit method carries over to an unrelated, open-weights vendor.
Run the audit to reveal.
Every number on this page came from an actual run against World Labs' World API, not a mocked or illustrative scenario — 2 real generations (~$2.40), a byte-level provenance scan of every exported file, and a real two-point measurement against a physical reference. The test harness itself had 3 real integration bugs along the way (API response fields didn't match assumed names) — all caught and fixed by running it, not by reading documentation alone. The Cosmos panel (added Sept 28, 2026) reuses the same scan method on NVIDIA’s own published Cosmos outputs and compares terms quoted from NVIDIA’s primary documents. It is a terms review, not legal advice.
Want the full picture for your environment?
A 20-minute intro call puts you in touch with the person who builds these engagements — not a form, a conversation.
Start a conversation