Photo to playable level: turning your desk into a race track

When to use it

It fits when the premise is "a real place, seen at an unusual size": a desk racer, a kitchen-counter platformer, a bookshelf climb, your own street as a top-down chase. Real places carry things that are hard to invent (wear, clutter, odd proportions), players recognise them instantly, and that recognition is the hook. It also works without a concept artist: the photos are your concept art.

It's the wrong tool for large spaces (a whole building is weeks of capture), for stylised games where photo textures clash with everything else, and for places dominated by glass, screens, chrome or blank walls, which photogrammetry can't reconstruct. A desk is close to ideal: small, measurable in ten minutes, full of distinct objects, and you control the light.

Three decisions come first:

1. Game scale. Physics engines are tuned for objects roughly 0.1–10 m in size. An 8 cm toy car at real scale jitters and falls through thin objects, so scale the world up. For a desk, ×50 (1 cm real = 0.5 m in game) puts the car at about 4 m.
2. Where each prop comes from. Model it (boxy, hard-surface objects), generate it from a photo with image-to-3D (irregular or organic objects), or scan it with photogrammetry (surfaces where the real wear is the point). Most levels mix all three.
3. How real. Photo-exact (scanned textures, matched lighting) or photo-faithful (real layout, proportions and palette with clean materials). The second is cheaper and ages better.

Kit: a phone, a tape measure, an A4 sheet or a ruler, Blender, an engine (Three.js, Unity, Godot), optionally a photogrammetry app (Polycam, RealityScan) and an image-to-3D tool (Tripo, Meshy).