Scene Editor

No scene yet

Add photos of a space and build a 3D scene you can segment, edit, and walk through.

drag look · scroll zoom · WASD move · Q/E up-down · ←→ roll (⇧ yaw) · ↑↓ pitch · F recenter — touch: 1-finger look · 2-finger pinch & pan

00 Capture photos → 3D

Choose photos or a video 2+ photos of one scene, or a video clip · JPG / PNG / MP4 / MOV Sharper reconstruction — process photos at 1024px (heavier GPU) HD splat — train a solid walk-through model (+1–3 min) Use all views — keep every image's points (fuller, more floaters)

Photos are processed in memory and never stored.

01 Segment pick an object

Tap the object in a photo to select it.

02 Lift to 3D mask → points

union intersect

Selected 0 points · masks in none

03 Edit the selection

Delete or move, then Fill hole to stitch the gap with the surrounding surface.

infer new geometry (Depth Anything)

Detects everyone in every view and removes them in one step. The person masks stay selected, so you can Fill hole after for a clean plate.

Recolor

Move

Rotate

Scale

◆ Scene

Point size

Shading Solid surface (eye-dome lighting)

Surface Mesh view (smooth lit surface)

Noise filter

Higher removes more floaters (rebuilds from the raw scan, so it resets edits).

Cross-checks the scan against monocular depth (Depth Anything) and drops points floating in front of surfaces. Undoable.

One click: removes floating blobs, then speckle, then smooths surfaces. Undoable in one step. Or run them individually below.

Manual

blobs = disconnected islands, speckle = lonely stray points, smooth = thins fuzzy surfaces, adaptive thin = merges flat regions but keeps detailed items dense.

Point-cloud editor. Reconstruction runs MASt3R on the server (~20–60 s); segmenting, editing and navigating happen in your browser. photos never stored