Photogrammetry — the process of reconstructing 3D geometry from a series of overlapping 2D photographs — was once the domain of specialized equipment and thousand-dollar software. Today, free tools like Meshroom and consumer-grade phone cameras can produce 3D models good enough for game assets, VFX references, and artistic applications. This tutorial covers the capture patterns that actually work, why lighting and image overlap matter, and how to process your photos into a usable 3D model.
Capture patterns: overlap, angle, and lighting
Photogrammetry software reconstructs 3D shape by finding matching points across photos and triangulating their position in space. The algorithm works by identifying features (corners, edges, texture) visible in multiple images and computing depth from the camera positions where each feature appears. This means good capture requires: high overlap between photos (each pixel should be visible in at least 3 images), images shot from varied angles around the object (a hemisphere of viewpoints, not a circle), and consistent lighting that doesn’t create specular reflections.
Why specular highlights break reconstruction
Photogrammetry struggles with shiny, reflective surfaces — the algorithm can’t distinguish between “this corner of the surface is at a different depth” and “this corner is reflecting light from somewhere else in the scene.” Objects with low specularity (matte painted surfaces, unglazed clay, fabric) capture beautifully. Shiny objects can be temporarily dulled with flour or spray matte (which washes off) during capture. Post-processing can remove specularities, but prevention is faster.
Meshing, decimation, and game-engine preparation
Raw photogrammetry output is millions of points with wildly high polygon density — beautiful for visualization but too expensive for real-time rendering. Decimation software (Meshlab, Instant Meshes) reduces the mesh to a workable triangle count. Normal maps (baked from the high-poly capture) preserve surface detail while running on game hardware. This tutorial covers that pipeline: capture → reconstruction → decimation → normal baking → import to engine.