Stereo Align Node

Stereo Align analyzes the vertical disparity (which should be zero in ideal stereo) and locally warps the right (or left) view to bring it into perfect vertical alignment with the other. Any remaining vertical offset causes eye strain and prevents stereo fusion.

It uses a vertical disparity map internally (different from the classic horizontal disparity). If one is provided as an explicit input, it is used directly; otherwise the node computes it.

• Effects Library: Tools > Stereo > Stereo Align
• Quick add: right-click in the Node Editor → Add Tool > Stereo > Stereo Align

Global Align

A uniform correction (global transform). For mechanical rig flaws (cameras misaligned overall). First in the pipeline.

Stereo Align

A local pixel-by-pixel correction. For non-uniform optical flaws (differential lens distortion). Second in the pipeline.

Canonical pipeline: Global Align → Stereo Align. Don't reverse it: the local analysis is cleaner on views that are already roughly aligned.

• Left (orange) - the left view.
• Right (green) - the right view.
• Disparity (blue) - the disparity map (optional, otherwise computed internally).
• Effect Mask - optional.

Full alignment pipeline:
LeftCam ─┐
         ├── Global Align (rough)
RightCam ┘    ↓
        ┌──── Stereo Align (fine, local)
        └─── rest of the compositing → MediaOut

With Disparity computed separately (cache):
LeftCam ─┐
         ├── Disparity (cached)
RightCam ┘     ↓
LeftCam ──┐    │
         ├── Stereo Align (Disparity input)
RightCam ─┘
              → MediaOut

Full clean workflow:
LeftCam ─┐
         ├── Global Align → Stereo Align → Disparity
RightCam ┘                                   ↓
                                       Disparity to Z
                                          → MediaOut

Method (default: Classic)

The algorithm. Classic = fast, enough for most cases. Enhanced = more precise, slower.

Apply To (default: Right)

The view to warp. Right by convention (the one that moves most). Choose Left if the left view has the main optical flaw.

Smoothness (default: 0.4)

Smoothing of the warp. Higher = a more uniform warp that loses the fine corrections. 0.3-0.5 recommended.

Vertical Only (default: On)

When on, only vertical disparities are corrected (the horizontal disparity, which holds the depth, is preserved). Keep it on except in specific cases.

Edges (default: Smoother)

Behavior at the edges. Smoother = soft, Sharper = strict.

Maximum Vertical Disparity (default: 8)

The maximum correction. Limits extreme warps that could introduce artifacts.

• Differential lens distortion - the two lenses have slightly different distortions → Stereo Align corrects the resulting local drift.
• Cleanup after Global Align - Global Align does not capture non-uniform flaws. Stereo Align finishes the job.
• Footage from amateur rigs - DIY rigs often have complex micro-misalignments that only local alignment can correct.
• Pre-processing before Disparity - a clean Disparity analysis requires views that are perfectly aligned vertically. Stereo Align upstream = a quality disparity map.
• Refraction footage (underwater, through glass) - optical distortion from fluids or glass causes large local drift. Stereo Align compensates for it.
• 2D-to-stereo conversion - artificially generated views (by New Eye from an estimated disparity) can have local residues; Stereo Align smooths them out.

• Always after Global Align - the local correction works better on views that are already roughly aligned.
• Vertical Only on - otherwise the node also tries to correct horizontal disparities = it flattens the depth. Keep Vertical Only on unless you intend otherwise.
• Check with an anaglyph - an anaglyph downstream easily reveals remaining vertical drift (visible vertical ghosting). Adjust Smoothness / Method until it is clean.
• Disparity input for caching - compute Disparity separately, cache it and plug it into the Stereo Align input = no recomputation on each iteration.
• Suitable Smoothness - for fine flaws (subtle lens distortion), a modest Smoothness (0.3); for massive flaws, a high Smoothness (0.6) to avoid overcorrection.
• Apply To Right by convention - unless the left view clearly has more optical flaws (focus, lens, etc.).
• Cache the output - Stereo Align is heavy. Cache it if several nodes sit downstream.
• For very non-uniform flaws (a badly assembled extreme rig), you may need several Stereo Align passes in a row - but beyond 2 passes, consider a reshoot.