dMerge

A Deep Image is not simply flat RGBA with one Z-buffer/AUX channel. A true Deep EXR may store multiple color, alpha and depth samples per pixel. dMerge operates on that multi-sample structure.

Blackmagic identifies the Deep Image Compositing Toolset as Studio Version Only. Check the Resolve edition and version as well as the presence of actual Deep source data.

There is one orange Background input for the base deep stream. The Foreground input is white and dynamic: adding another Deep connection creates an additional incoming Foreground socket, similarly to Merge3D.

The dMerge entry in the Deep Image Nodes reference does not specify a blue Effect Mask input. Do not copy an ordinary Merge wiring diagram and invent a third “Effect Mask” socket here.

Every source feeding this node must supply Deep Image data. Add an ordinary 2D MediaIn through Image to Deep instead of plugging flat RGBA directly into a deep Foreground.

Image to Deep gives a flat element a Deep representation at a specified depth. This allows 2D graphics into a deep scene, but cannot reconstruct hidden multi-sample geometry from an ordinary photo.

dMerge combines Z samples; it does not invent missing depth information or repair mismatched cameras, Z-axis conventions, near/far ranges or scene scale.

“Foreground” describes an input connection, not an instruction to “always paint on top.” Depth samples and their compositing interpretation govern the occlusion once flattened, not merely cable order.

dMerge has no creative Merge tab with Apply Mode/Size/Center and no Controls tab with a Blend slider. The captured Inspector shows Settings because the work is combining deep data, not mixing two flat RGBA layers.

The real dmerge-settings.png screenshot in the supplied project confirms Hide Incoming Connections, the Layers group, three dropdowns, Comments and three Render Scripts sections. Dynamic ports cannot be verified from this Inspector capture and are sourced separately.

Layers / Process Layers selects the named image layer(s) to be processed. The screenshot shows Default Layer, a reasonable starting state for streams without deliberate multilayer packaging.

Layers / Background Layer selects which named image layer to request from the Background input. Auto requests the current Process Layers choice and falls back to the default layer if it is missing.

Layers / Foreground Layer similarly selects the named image layer supplied by Foreground inputs. It routes image layers; it is not a depth/Z offset or a command to force an object in front of the camera.

Blackmagic describes Auto, Match and explicit layer selection for multilayer Input Layer controls. Match fails if a required layer is absent rather than falling back. Verify the exact dMerge dropdown contents on the target Resolve build.

If a Deep EXR contains several named image layers, keep two diagnoses separate: first verify which named layer the input requests, then inspect the actual Z samples belonging to that layer.

Hide Incoming Connections helps organize node graphs: with it enabled, incoming links can be assigned via Inspector fields and their lines remain hidden until the tool is selected. It is a Flow-display mechanism, not input bypass.

Comments documents source origins, camera, depth conventions and the expected flattening point. Its text does not alter the image, but it makes handoff and debugging significantly easier.

Frame Render Script, Start Render Scripts and End Render Scripts are common Fusion scripting facilities. They may run custom pipeline code but are not dMerge algorithms. Leave them unused in ordinary compositing.

The output of dMerge is still a Deep Image. Feed it to DeepToImage for a normal 2D preview. That is the explicit boundary between Z samples and flat RGBA.

DeepToImage offers Volumetric Composition for overlapping semi-transparent volume samples. This belongs to flattening, not dMerge itself; when testing fog, smoke or glass, keep track of which stage actually changes the outcome.

Do not feed dMerge.Output directly to normal 2D Merge, Blur or MediaOut and expect Deep samples to survive. Flatten through DeepToImage intentionally, then use ordinary 2D nodes.

If Deep passes differ in dimensions, pixel aspect ratio or camera solve, geometry may not line up in the same pixels. Match the source geometry first, then judge dMerge.

Separate “which surface is in front in Z” from transparency: per-sample alpha and Z ordering both matter to the final result. Reordering foreground connections is not a substitute for correct Deep alpha and sample placement.

Keep Deep passes in compatible linear color/alpha conventions. Converting or changing premultiplication on only one stream can create flattening fringes even when dMerge itself is wired correctly.

For a baseline test, use two known Deep EXRs with a clear object intersection. Compare DeepToImage after dMerge with ordinary 2D Merge of separately flattened inputs: partial overlaps should demonstrate why Deep ordering matters.

To test the dynamic Foreground design, connect a third Deep source to the same dMerge. You should still have a single dMerge instead of a chain of flat two-input Merges; the node grows its own Foreground inputs.

Evidence boundary: the screenshot verifies visible Settings labels and defaults; the documentation establishes the topology and semantics. Actual `.comp` port identifiers, dynamic add/remove UI and rendered results still require live Resolve verification.