Compositing with depth channel data instead of flat images — adjusts focus, defocus, and atmospheric effects in post without re-rendering.
You're in the edit suite and realize: the focus isn't right, the depth of field doesn't match the story. In the past, that meant reshooting or a complete re-render — expensive, time-consuming, sometimes impossible. Deep Compositing solves exactly this problem by allowing you to carry over not just RGB images, but also depth data (Z-Depth) from the 3D render. This enables you to shift focus in the post-production process, adjust defocus afterward, or selectively control atmospheric effects by depth.
The practice works like this: from your 3D package (Cinema 4D, Maya, Blender), you export not only the beauty pass but also separate depth maps — Z-Depth or World Position passes. In your compositing setup (Nuke, After Effects with plugins), you use these depth values to precisely re-control the depth of field. You can make individual layers in space appear in focus without the original render's focus plane needing to be correct. A composited actor element can be made blurrier in post, while the background remains sharp — or vice versa. This not only saves render time but also gives you creative flexibility you might not have had in the initial pass.
This becomes particularly valuable in Atmospheric Compositing: you use the depth data to selectively place volumetric fog, smoke, or light rays in specific depth layers. A light ray can appear denser closer to the camera and thin out with distance — entirely controlled by the depth mask. It also helps with color correction: you can grade foreground and background separately without having to manually paint masks.
The workflow requires discipline: even during rendering, you must ensure your Z-Depth passes are consistent and that transparencies are handled correctly. Sometimes you need multiple depth layers (Near, Mid, Far) to achieve clean overlaps. With modern AI-based defocus (like in DaVinci Fusion), this becomes even more elegant — the software recognizes depth edges itself and interpolates missing data. This isn't fake bokeh, but physically plausible calibration.