Attention
Unity has discontinued selling and supporting Deep Compositing effective February 7, 2024.
What's the workflow?
There are two typical Nuke workflows for deep compositing: Deep Holdout and Deepenate.
Deep Holdout Workflow
Here is a typical Nuke workflow. It reads RGB images and deep data separately, assembles them, and produces deep alpha maps. The DeepHoldout node computes a holdout matte:
![]() |
---|
Deep Holdout Workflow |
Deepenate Workflow
The holdout method works best when compositing layers back-to-front. Otherwise (in extreme cases,) results can differ if layer order changes. However you can eliminate any differences by including color information, using deep color maps — depth-sorted lists of Z,R,G,B,A values per pixel. You create deep color maps at composite time by merging deep data with the RGB rendered image. Specifically, you can run an operation that unpremultiplies by the alpha in the RGB render, then premultiplies by the alpha in each sample. This then lets you composite by merge-sorting passes, without having to modify the colors in each sample. And you can combine an unlimited number of renders into a single deep color map. The final output is still a regular (flattened) RGBA image.
The Deepenate workflow is more flexible than the deep holdout workflow, since the order of compositing is completely irrelevant. The depth-sorting operation in the merge node guarantees correct and consistent ordering. It is also a lossless workflow: combining multiple passes discards no information about the source passes. At any point (before resolving into the final flattened image), you can output a pre-composited deep color map — which you can later reload and interleave with other deep color maps.
Below is another example of a compositing script that includes a Deepenate node, which adds RGB color channels from a rendered image to the deep alpha map.
![]() |
---|
Deepenate Workflow |
The holdout-based compositing workflow is faster, but Deepenate is more flexible. Choosing which to use involves a tradeoff (depends on the scene).