ZToDisparity

ZToDisparity performs the conceptual inverse of DisparityToZ: it starts from existing Z AUX and calculates stereo disparity, the X/Y image-space offset describing correspondence between left and right views.

Technical Z AUX must be distinguished from an ordinary grayscale depth-looking image. The tool expects an image stream carrying a Z channel; a pretty black-and-white image by itself is not equivalent to Z AUX.

ZToDisparity remains an image processor. It does not accept a raw Merge3D scene in place of an image with Z; render the scene through Renderer3D with Z enabled or load an EXR/source that already carries Z.

In None/unstacked topology the node exposes physical Left Input and Right Input paths. They carry separate eye images from one stereo pair and each output can receive corresponding Disparity AUX.

In Horiz, Vert, and Layers modes both eyes are already packed inside one representation. The primary Left Input carries the full packed stream, a separate Right Input is unnecessary, and the packed result continues through the primary Left Output.

Stack Mode is primarily a topology and routing control. It does not set stereo strength or repair bad Z; a wrong mode makes the tool interpret the structure of the input incorrectly.

Swap Eyes changes left/right interpretation. It corrects actual eye order; it is not a generic Z inversion, a disparity inversion switch, or a depth-strength control.

External Camera Mode bases the conversion on stereo camera geometry. For physically meaningful disparity, the camera must correspond to the scene or footage rather than being an arbitrary Camera3D.

Stereo Camera is conditional: it is required by External mapping. Artistic mode can map Z into controllable disparity without an accurate camera rig, but the result should not be treated as physically measured stereo geometry.

The Stereo Camera source can be a Camera3D with valid stereo geometry or a camera-containing 3D scene stream. Camera selects the intended camera model when multiple cameras are present.

In External mode, disparity scale depends on camera projection and stereo separation. The same Z field can therefore produce a different image-space offset under different valid camera geometry.

Artistic mode is useful when visual depth is needed without reliable camera calibration. It creates a creative mapping between Z and disparity rather than reconstructing capture geometry.

Convergence Point in Artistic mode defines the Z plane of zero disparity. Objects on opposite sides of that plane move in front of or behind the screen plane according to the eye convention.

Background Disparity defines the far-background or infinity disparity used by the Artistic mapping. The corresponding opposite-eye offset uses the matching opposite direction in a consistent stereo pair.

Output Disparity To RG does not enable disparity calculation. Technical Disparity AUX is created independently; the checkbox only copies disparity components into ordinary color channels for viewing or RGB processing.

The diagnostic color copy represents disparity as {x, y, 0, 1}: horizontal disparity is written to R, vertical disparity to G, B is zero, and alpha is one.

The RG representation uses float32 because disparity can be negative and extend well outside the normal 0-1 display range. A clipped normal Viewer can therefore look strange without the underlying data being wrong.

For production stereo pipelines it is often better to keep Output Disparity To RG disabled so the eye RGB remains untouched while downstream stereo-aware tools read hidden Disparity AUX directly.

Refine Disparity runs after the base Z-to-disparity conversion. It uses RGB structure to make disparity edges follow visible object boundaries more closely.

Refinement cannot repair fundamental input errors such as missing Z, a mismatched stereo camera, reversed eyes, or a wrong Stack Mode. Fix those causes before enabling Refine Disparity.

Enable activates the refine stage, while HiQ Only can restrict the refinement to High Quality processing. Compare interactive preview with the actual High Quality state during QC.

Strength controls the amount of RGB-guided correction. Excessive values can imprint texture, grain, skin detail, or lighting edges into disparity as false geometry.

Radius sets the spatial scale of refinement. A radius that is too large can smear narrow depth transitions and create halos around silhouettes.

A CG or EXR pipeline is a natural ZToDisparity use case: a render already carries accurate Z, and disparity can be derived using the same stereo Camera3D that defines the scene.

ZToDisparity does not synthesize newly exposed surfaces behind occlusions. Disparity describes motion of known pixels but cannot know what should appear when a new eye view reveals hidden content.

A downstream warp or new-eye synthesis can therefore require occlusion repair, inpainting, or an additional CG render, especially with large foreground disparities.

A Z-to-Disparity-to-Z round trip is useful for diagnostics but is not guaranteed to be lossless. Precision, camera assumptions, mapping, clipping, and refinement can introduce differences.

Far surfaces deserve particular care because very small disparity changes at large Z can correspond to large depth differences, making round-trip error appear amplified.

Alpha remains part of the eye image and is not automatically repurposed as disparity. AUX data and the RG preview are separate representations; do not use alpha as a substitute for Z or a second eye.

When caching or exporting, verify that the chosen format actually preserves Disparity AUX. An RGB render can look correct while the technical channel has silently disappeared.

Final validation should include playback, not only a still frame: noisy Z, motion blur, reflections, and occlusion boundaries can become temporal disparity instability that is particularly visible in stereo.