When this is running, the maker types one instruction into the prompt box asking to surface model the Mini in Tesselform, and the software builds a manufacturable surface model of the whole classic car shell from raw scan data. The AI does the tedious work while the user concentrates on checking the finished deliverable, which is the whole point of the setup.
The underlying scan was captured years ago with a Faro Quantum S and arrives as a blue polygonal mesh holding 9.6 million triangles. The model doing the work is Claude Opus 5.5 running at high effort, connected through Tesselform itself, and the whole job took about 30 hours with 32 prompts sent. The finished result is 143 panels, assembled one by one in an animated sequence, including a yellow rear bulkhead with stamped indents and cutouts. A scan check compares the original mesh against the completed model to confirm alignment, and zebra reflection stripes get projected over the surfaces to judge smoothness and curvature continuity.
Two details stand out to me. Everything the AI produces stays fully editable by hand afterwards, and the maker claims this is the first software doing design-intent modelling of complex scanned surfaces, arguing that existing auto-surface tools attach NURBS patches randomly and give unusable output on complete scans. The maker also estimates the same work would take a skilled person over 50 hours manually, so the time saving looks substantial.
One open question: higher effort levels on stronger models give better results but take longer, and it is unclear whether Chat models can handle this surfacing workflow at all, since the maker has not tried that yet.