No coding required. Any AI image tool will do: submit one image, one prompt, one source, and a scene returns to the map.
How
Claim: open a claim issue, or click a hollow dot on the map.
Research: find references (copperplates, album, old photos, Lei drawings) per the scene's listed basis.
Make: write parameter-rich prompts, generate with references, self-check the physics list.
Submit: attach image + metadata to the issue; Git users open a PR, others we merge for you.
Checklist
Image: 2940×1912 PNG or WebP ≥1600px.
Reference source: link or holding info, public domain or licensed.
Full prompt + model name/version.
One line: what is sourced, what is inferred.
Credit: your name/ID for the contributor wall.
Rules: no fabricated “sources”; never present AI images as photographs or surveys; references must be traceable; images shared under CC BY-SA 4.0.
Three-dimensional restorations
Blender, SketchUp or scripted modelling — all welcome. 3D ranks equal to images: renders alone count; add a glb and the scene becomes rotatable in the gallery and viewable in AR on a phone.
One hero render plus front, side and oblique views, WebP ≥1600px.
Optional glb, ≤10MB; metres, origin at terrace centre, Y-up. Trim geometry first — Draco shifts decoding to an overseas CDN.
State element by element what is sourced and what is estimated.
Procedural or public-domain textures only; no unverified scans or images from the web.
3D is checkable geometry, not a survey. It stands only if the column grid, bays and dimensions trace back to the drawings; materials and painted decoration remain inference and must be labelled.