Tutorial
How to get
a great scan
The model can only be as good as the photos. Five minutes here saves wasted credits — capture quality is 90% of the result.
Pick a good object
Photogrammetry measures surfaces by matching details across photos. Objects with visible texture reconstruct beautifully; surfaces that look identical from every angle cannot be measured.
- DOMatte, textured objects — toys, figurines, shoes, plants, sculptures, food packaging with print
- DOObjects between 5 cm and ~1 m, placed where you can walk all the way around
- DON'TMirrors, glass, transparent plastic — light passes through or reflects, so there is no stable surface to measure
- DON'TPlain shiny objects (polished metal, glossy white plastic) — reflections move as the camera moves
- DON'TVery thin structures (wire fences, fur, hair) and moving subjects (pets that won't sit still)
If your object is glossy, a light dusting of dry shampoo or chalk spray makes it scannable.
Set up the shot
- DOSoft, even lighting — an overcast day or a room with diffuse light is ideal
- DOPut the object on a surface with some texture (wood grain, a newspaper) — it helps the camera lock on
- DON'THarsh direct sun or a single bare bulb — hard shadows get baked into the model
- DON'TMove the object between shots — the scene must stay frozen; you move, it doesn't
Capture: orbit slowly, overlap a lot
Walk a full circle around the object, keeping it centered and roughly the same size in frame. Each photo should overlap the previous one by about 70% — small steps, not big jumps.
10–30 photos (PNG/JPG). One orbit at chest height; for tall or complex objects add a second, higher orbit looking down.
10–30 seconds (MP4/MOV), one slow steady lap. We automatically pick the ~25 sharpest, most diverse frames — you review them before reconstruction starts.
- DOMove slowly and keep your hands steady — motion blur is the #1 quality killer in video
- DOFill the frame: the object should take up most of the picture
- DON'TStand in one spot and rotate the object — walk around it instead
- DON'TZoom in and out mid-capture — keep the framing consistent
Upload and review
On the Create page, drag in your photos or video. Video uploads pause at a frame-review grid: drop blurry frames with the × button, and drag to reorder if needed. Then hit Start Reconstruction.
Optional: isolate the object with AI Detect
If the background is busy, type what the object is (e.g. "red toy car") and click AI Detect. We find the object in every photo and mask out everything else, so the reconstruction spends all its effort on what you care about. You can review and adjust each mask by hand before confirming.
View, download, print
Reconstruction takes a few minutes. When it finishes you get an interactive viewer — drag to rotate, scroll to zoom. Download as GLB (textured, for Blender, game engines, AR) or PLY (point/mesh data). Meshes are watertight and ready for slicing if you're 3D printing.
FAQ
Common questions
My reconstruction failed or looks broken. Why?
Almost always a capture issue: too shiny, too little texture, too few angles, or motion blur. Re-shoot following sections 01–03 — especially the 70% overlap rule. Failed jobs still consume a credit because GPU time was spent, so it pays to capture well.
How long does it take?
Typically 5–10 minutes for a 12-photo scan. Complex scans can take longer; keep the tab open and the progress bar will keep you posted.
How many scans do I get?
Free accounts include 5 reconstructions. Pro ($19.99/mo) adds 30 every month — see pricing.
Who owns the models?
You do. Models you create are yours to use commercially or otherwise.