render3d() run six times over, once for each face of a cube, with the camera sitting in the middle of the volume looking outward. Together the six views cover every direction at once.
The faces join without a seam, and not because anything blends them. Each face covers exactly a quarter turn, and adjacent faces share the directions along their common edge. A ray leaving the edge of one face is literally the same ray entering the edge of the next, so the pictures line up exactly by construction rather than by correction.
Lighting and gamma are inherited from render3d(), so what you get is a lit object floating in space, viewed from inside.
If you want the raw field with no lighting applied at all, use renderCubemapSurface() instead.