A good reconstruction starts with a good capture. The reconstruction only knows what your frames show it — so the goal is a complete, consistent, well-covered set of views of a static, well-lit scene. These principles apply whether you capture in Photo, Video, or LiDAR.
Core principles
Cover the scene completely
Orbit the subject so every surface is seen from several angles.
Overlap consecutive views generously.
Vary your height — low, eye-level, and high.
Don’t skip corners, undersides, and concave areas.
Move steadily
Slow and smooth beats fast — sudden moves cause blur.
Keep the subject in frame and roughly centered.
Set a Minimum Shutter Speed when motion risks blur.
Keep it consistent
Lock exposure when lighting is steady.
Lock focus (AF Lock) or set a manual focus distance.
Avoid changing controls mid-capture.
Light it well
Even, diffuse light is ideal.
Avoid changing light during a capture.
Watch reflections, transparency, and shiny surfaces.
Patterns by scene type
The path you walk depends on what you’re capturing. Pick a scene type to see the pattern.
ObjectSingle object
Orbit it in 2–3 rings at different heights (low, eye-level, high).
Keep the object centered and roughly the same size in frame.
Close the loop on each ring.
Keep in mindAim for generous overlap so consecutive views share most of the object.
View distribution for Gaussian splattingSplatting rewards even angular coverage: surfaces seen from several directions reconstruct cleanly, while surfaces seen from one angle stay fuzzy or stretched. Spread viewpoints around the subject rather than piling frames into one vantage.
Scenes that work — and that don’t
✓ Works wellStatic, opaque, textured subjects and spaces.
In Photo and Video, the 0.5× ultra-wide captures broad context while the 1× wide captures detail — from synchronized viewpoints, both recorded at once. The added coverage and parallax from the two lenses generally helps reconstruction; you don’t have to choose between them.
Mode-specific notes
VideoBest for fast, continuous coverage. The steadier you move, the sharper the frames.PhotoUse Auto cadence for systematic walk-throughs, or Capture Once for deliberate, high-quality frames.LiDARKeep ARKit tracking healthy: move smoothly, keep textured surfaces in view, avoid blank walls. Only cleanly-tracked frames make the COLMAP model.