Cameras lie about space

Field of view is a compositional choice, not a default. A narrow 30-degree lens flattens a product and reads as expensive; a wide 75-degree lens exaggerates perspective, makes rooms feel enormous, and is what almost every generated scene ships with. Aspect must match the canvas or everything is subtly stretched, and it must be updated on resize alongside the projection matrix, which is the line the model forgets.

Near and far are where the damage happens. Depth is stored non-linearly, so the ratio between them decides your precision: with near at 0.01 and far at 10000, almost the entire depth buffer is spent on the first few metres and distant coplanar surfaces flicker against each other. Push near out as far as your scene tolerates before you reach for anything cleverer. It is free and it usually fixes the flicker entirely.

An orthographic camera drops perspective divide altogether: parallel lines stay parallel. It is the right answer for isometric scenes, technical diagrams, UI rendered in 3D and every shadow map from a directional light, and it makes size comparisons honest in a way a perspective camera cannot.