Photo-to-3D-Model Scanner — Phone Photogrammetry

Closer, and turned a little. Measured against reference scenes with known geometry, a head close-up matches the true shape at 0.85–0.95, a figure at three-quarters at 0.72–0.86, and a figure straight on at only 0.1–0.5 — head-on, a person has almost no depth variation and the relief comes out smooth. For a full-length shot, turn the subject 30–45° and step closer.

AI Depth for reliefs — medallions, coins, faces, keepsakes, lithopanes. Turntable for full 3D objects — figurines, toys, small sculptures. If you only have one photo, AI Depth is your only option.

Any modern phone. For the turntable mode, keep the distance, lighting and focus constant between shots; use a plain white or green background so the silhouette is easy to segment.

A single photo carries only the front, so depth is built from the visible geometry. The “Cut to the subject” shape removes the background and stands the relief on a plinth; “Figurine” mirrors it onto the back, which reads as a statue from across the room — but that back is a mirror of the front, not a scan of it. Real volume comes only from the turntable.

AI Depth is good for art and keepsakes (±1 mm on fine details, up to ±5 mm on overall shape). Turntable is more accurate for silhouette-dominant objects (±0.5 mm) but concave details can be lost.

After the AI model finishes downloading once, the service worker caches it together with three.js and the page. Subsequent visits work even without an internet connection until the cache is cleared.

Yes — the exported STL is a closed, printable mesh. We run a watertight check and show a warning if it has open edges. Slice it in Cura, PrusaSlicer or Bambu Studio.