Multi-photo to 3D

EXPERIMENTAL

Walk around a still object. Bring the viewpoints together.

Local processing · No AI credits

MORE VIEWS, MORE EVIDENCE

One object.
Every perspective.

Add 3–40 overlapping photos of a stationary object.
We’ll recover cameras, combine depth and build a surface.

A complete model depends on overlap and visible surface detail.

A CONNECTED SET OF VIEWS

Keep it still. Move around it.

01 / OverlapTake small steps. Neighboring photos should share most of the object. Three is a minimum, not a guarantee.

02 / Surface detailUse sharp photos and soft, even light. Glossy, plain surfaces remain difficult without AI.

03 / CoverageInclude front, sides, back and higher views. We cannot recover surfaces that no photo sees.

YOUR PHOTO SET

Start with the photos

JPG, PNG, WebP · 20 MB each · up to 40 photos

Same camera, lens and zoom

Existing transparency is preserved. Opaque photos get an automatic background-removal attempt. Each cutout appears on a checkerboard. Check the result; adjust the subject box and update the cutout if parts are missing.

Automatic separation uses colors without AI. Objects touching the frame edges or matching the background may need correction. You can also use the original.

Keep the object and surroundings still. Original photos help camera alignment. The exact cutouts shown here limit the 3D surface reconstruction.

LOCAL SCAN ENGINE

Connecting…

Multi-view reconstruction uses a local companion service.

Local setup

From the res project folder, run:

npm run scan:serve

One object.Every perspective.

Keep it still. Move around it.

One object.
Every perspective.