DisparityToZ

What this node does. DisparityToZ does not reanalyze a stereo pair. It reads an existing Disparity AUX channel, usually generated by Disparity, and adds a computed Z AUX channel. A grayscale image in RGB is not a substitute for technical disparity data.

Pipeline position. A typical order is stereo preparation and alignment → Disparity → DisparityToZ → DepthBlur, fog, or another depth-aware process. Validate the stereo correspondence first, then convert it to Z.

Separate-eye inputs. With Stack Mode set to None, the left and right images use separate cables into Left Input and Right Input. The reference manual calls this topology Separate; None is the label shown in the captured Resolve 21 Inspector.

Packed stereo. Horiz, Vert, and Layers carry both eyes within one image stream. Feed that representation into Left Input and use a Stack Mode that matches the upstream node. An extra cable into Right Input is not needed.

Dedicated camera input. External Camera Mode exposes the magenta Stereo Camera input, which expects a Stereo Camera3D or matched camera setup as 3D-scene data. It is not an image, mask, or depth-map input.

Outputs preserve the imagery. Left Output and, in separate mode, Right Output return the corresponding eye images with an additional technical Z channel. RGB does not automatically turn into a grayscale depth image; that is the job of Output Z To RGB.

Why geometry matters. Disparity is a displacement between image projections, whereas Z denotes depth. The relationship depends on camera geometry. A pixel offset by itself cannot establish an accurate real-world distance.

External: camera-based reconstruction. Supply a stereo camera whose eye separation, focal geometry, and orientation match the shot. This provides camera-consistent depth, but it cannot guarantee accuracy when calibration or the incoming disparity is wrong.

Camera selector. The Camera tab can select a camera model from the attached 3D stream; the captured Inspector shows Default. If several cameras exist, confirm that the selected one corresponds to the actual stereo footage.

Artistic: depth without a camera. If no trustworthy stereo camera is available, Camera Mode = Artistic defines two disparity-to-depth anchor pairs, one near and one far. The resulting Z is useful for visual effects but is not a measurement of physical distance.

Disparity anchors. Pick Foreground Disparity and Background Disparity from the left-eye image, as the Inspector specifies. The first represents a chosen near region and the second a far region. Disparity for the right eye can have the opposite sign.

Depth anchors. Foreground Depth and Background Depth define the desired depths for those two measurements. Artistic depth fields use positive depth magnitudes in the UI; do not confuse them with raw camera-space Z, which normally becomes more negative with distance.

Artistic range limits. Disparity outside the range bracketed by the near and far anchors is clipped. Anchors that cover too narrow a range create large, flat areas in Z. Measure representative disparity values before defining the mapping.

Falloff shapes the curve. Hyperbolic more closely follows an inverse-depth relationship and emphasizes the foreground. Linear spreads the change more evenly and is intended for stylized work. Even Hyperbolic in Artistic mode does not make depth metrically accurate.

Z AUX versus RGB preview. The auxiliary Z channel is created whether or not Output Z To RGB is enabled. Turning it on also writes Z as {z, z, z, 1} into float32 RGB channels for inspection or conventional image processing.

Why the Viewer may look black. Raw Z is often negative and far outside the display range of 0–1. Viewer → Options → Show Full Color Range helps inspect it. Normalizing a preview image is not the same as altering the underlying Z AUX data.

When to use Refine Z. Enable applies color-guided depth refinement. Use it only after confirming that the raw conversion is plausible: it can improve silhouette boundaries for depth of field and fog.

HiQ Only, Strength, and Radius. With Refine Z enabled, HiQ Only restricts processing to High Quality, Strength increases smoothing and attraction toward RGB edges, and Radius sets the spatial smoothing scale. These dependent controls are hidden in the supplied screenshot because Enable is off.

The refinement trade-off. Excessive Strength can imprint fabric, skin, or foliage texture onto the Z field; a large Radius may erase narrow transitions or create halos. Compare raw and refined Z at the same edge, changing one setting at a time.

Refinement cannot repair stereo errors. Missing Disparity AUX, wrong eye order, a mismatched Stack Mode, or an unrelated stereo camera produces incorrect depth regardless of smoothing. Correct the upstream cause before refining edges.

Uncertain distant depth. As objects move farther from the camera, disparity changes become extremely small. Slight disparity noise can produce a large error in reconstructed far Z. Do not treat it as a precise distance measurement without independent validation.

Changes to stereo geometry. If eye alignment, scale, crop, or StereoAlign is changed after disparity was computed, the pixel correspondence may no longer be valid. Recheck and, when needed, recompute disparity before deriving new Z values.

Temporal quality control. Review moving frames, not just a single still: jitter, holes, and flickering disparity become unstable Z, especially when DepthBlur is used. Check foreground, midground, background, and difficult edges.

Delivering the result. For depth-aware effects, retain actual Z AUX together with the original eye RGB. When exporting EXR, distinguish raw float Z from a display-normalized RGB preview and from artistically mapped Z. They are not interchangeable deliverables.