New Eye Node

New Eye warps one of the source views pixel by pixel based on the disparity, to synthesize a view at any position between left and right. Without this capability, changing the interaxial after the shoot would require a reshoot.

A classic use: reduce the 3D effect of a shot that feels too strong (Position 0.7 instead of 1.0 = a new "right view" closer to the left = reduced depth). Or the opposite, exaggerate the depth with a Position outside [0,1].

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

New Eye uses the disparity map to know where each pixel of one view sits in the other. To generate a view at Position t:

Position = 0

The node returns the left view unchanged. No interpolation.

Position = 1

Returns the right view unchanged.

Position = 0.5

The middle view: each pixel of the left view is moved by half of its disparity.

Position > 1 or < 0

Extrapolation: a view beyond the source eyes. Increases the perceived interaxial = more depth. Risky: artifacts if the extrapolation is too strong.

• Left (orange) - the left source view.
• Right (green) - the right source view.
• Disparity (blue) - the disparity map (output of a Disparity node).
• Effect Mask - optional.

Reducing the interaxial (less depth):
LeftCam ─┐
         ├── Disparity ──────┐
RightCam ┘                   │
                             ↓
LeftCam (Left input) ─── New Eye (Position = 0.7)
RightCam (Right input) ──┘
                          → a new right view, closer to the left
                          (reduced depth)

Increasing the interaxial (extrapolation):
[same setup] → New Eye (Position = 1.3)
              → a view beyond the right one, more depth

Autostereoscopic multiview generation:
LeftCam ─┐
         ├── Disparity
RightCam ┘
[Several New Eye nodes in parallel:]
   Position = -0.5 → View 1
   Position = 0.0  → View 2 (= Left)
   Position = 0.5  → View 3
   Position = 1.0  → View 4 (= Right)
   Position = 1.5  → View 5
   = 5 views for a multiview screen

Position (default: 0.5)

Position of the view to generate between left (0) and right (1). Outside [0,1] = extrapolation. Often animated for stereoscopic zoom effects.

Method (default: Optical Flow)

The interpolation method. Optical Flow = the best quality, through vectors. Frame Blend = a simple crossfade (ghosting). Forward / Backward = uses only one source view.

Holes Fill (default: On)

Fills the holes (occluded areas revealed by the interpolation) by inpainting from neighboring pixels. Essential for heavily offset views.

Edges (default: Smoother)

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

• Reducing depth - a shot filmed with too much interaxial, viewers complain: Position 0.7-0.8 = a reduced right view, softened depth.
• Convergence shift in post - move the screen plane onto a different subject than during the shoot.
• Autostereoscopic multiview - some glasses-free 3D screens need 5-9 views. Several New Eye nodes at different Positions = the generated views.
• Fixing problems on one camera - only one of the two cameras has a flaw (focus, dust). Regenerate the faulty view from the other one with New Eye.
• 2D → 3D stereo conversion - combined with an AI-estimated disparity (MiDaS), it lets you generate a stereo view from 2D.
• Depth animation - animate the Position from 0 to 1 over time = a stereo zoom effect (depth that intensifies).

• A clean disparity is mandatory - artifacts in the disparity map carry over into the generated view. Clean it up before New Eye.
• Position within [0,1] for quality - extrapolation (outside [0,1]) generates more artifacts. Stay within ±0.3 beyond the limits to keep it acceptable.
• Optical Flow method = standard - the other methods are fallbacks for degraded cases.
• Holes Fill on - avoids black edges at occlusions. Always enable it unless you need to see the holes (debugging).
• Cache the output - New Eye is heavy. If several nodes consume it downstream, cache it so it is not recomputed.
• Combine it with the Combiner - New Eye produces one view, and the Combiner can assemble it with a source view to produce a new stereo pair (e.g. the original Left + the generated New Eye instead of Right).
• Check with an anaglyph - test with an anaglyph output to see whether the generated view fuses correctly with the source in the brain.
• For massive interaxial shifts, combine a partial New Eye with a Stereo Align to correct the remaining artifacts.