Following one prop end to end makes the pipeline concrete. Take a wooden supply crate destined for a third-person game.
Block-out
A box at the intended dimensions, placed in the level. The only questions being answered are: is it the right size to read as a crate, can the player vault it, does it block the sightline it needs to block.
Reference
Photographs or concept. Establishes proportions, plank layout, hardware, and — most importantly — the wear story. Where does this crate get scuffed, and why.
High poly, or generation
Either sculpt the detail — chamfers, plank gaps, dented corners, nail heads — or generate a detailed mesh from the reference. Both produce the same thing: a dense mesh carrying detail that will not ship.
Retopology
Rebuild at the target budget. For a mid-distance prop in a third-person game, that is typically low hundreds to low thousands of triangles depending on the project. Silhouette-critical geometry — the chamfered edges — survives; flat plank faces do not need density.
UVs
Unwrap with seams along the plank edges where they are already visually broken. Even texel density across the crate; the top and front get slightly more, because that is what the player sees.
Baking
Transfer normal, ambient occlusion and curvature from the high poly to the low poly. This is what makes a few-hundred-triangle box read as planks and hardware.
Texturing
PBR channels driven by the baked maps. Wood grain in base colour, roughness variation where hands touch it, metal hardware masked by the curvature bake.
Export and integration
Correct scale and pivot at the base. A box collider — not the render mesh. Three LODs. Materials wired up and, if it is instanced heavily, added to the shared atlas.
The point
Every stage answers a question the next one depends on. Skipping one does not remove the question; it moves it later, where the answer costs more.