VIDEOMONTAZHERDAVINCI RESOLVE · VIDEO · AI

FUSION · NODE REFERENCE

RepairFrame

Optical-flow frame reconstruction

RepairFrame replaces a damaged frame or region using that frame and two neighbors. It computes optical flow internally; a separate upstream OpticalFlow is unnecessary. Define the time of application and the spatial mask separately.

RestorationAdvanced

RepairFrame

Overview

What RepairFrame does

Repairing frame 59 requires accessible frames 58,59 and 60. The tool is not a simple clone of one neighbor: it estimates motion and warps the selected sources. A cut, an occluded area or damage in all three frames can leave no reliable material for reconstruction.

The manual explicitly states that RepairFrame computes optical flow itself and removes AUX channels after processing. Do not expect it to supply ready Vector/Back Vector data for another vector operation. If needed downstream, calculate new fields for the repaired sequence.

The archived Inspector has Forward/Backward rows and Prev/Next columns. The reference preserves Forward Prev, Forward Next, Backward Prev and Backward Next; the manual reverses the word order as Prev Forward and Next Forward. Prev/Next chooses the frame and Forward/Backward chooses the deformation field; these are not four playback speeds or directions. Enabled reconstructions are blended, so more checkboxes do not guarantee cleaner edges.

Fastest On Top is a starting hypothesis for a locked camera with a moving foreground object. When the camera tracks the object, the background may move faster; compare Slowest On Top instead. This orders overlap by motion magnitude, not physical Z depth or render speed.

Clamp Edges can fill a small transparent gap by stretching the border. The cost is stretched pixels, not recovered hidden imagery. Edge Softness appears only after Clamp Edges is enabled. The manual suggests around 0.01 with multiple reconstructions or around 0.03 with one; these are starting points rather than mandatory values.

Method=Advanced uses the GPU; Classic uses the earlier CPU algorithm. The Advanced group is collapsed in the capture; shared controls are documented under OpticalFlow. Settings for one method should not be mechanically transferred to the other. Half Resolution and Proxy reduce analysis resolution rather than final frame dimensions.

For a single defect, set Settings Blend=0 at 58,1 at 59 and 0 at 60 with Linear interpolation. On the integer frames used in this exercise only 59 is repaired; the value stays zero outside the outer keys. A separate soft Effect Mask restricts the area within that frame. Subframe processing or motion blur requires a separate temporal-interpolation check.

If neighboring frames differ in exposure or color, the borrowed region can stand out. First choose a usable continuous shot and bring neighboring frames to a comparable appearance; a soft mask helps the transition but cannot fix incorrect motion or missing data.

When to use it

  • Single-frame dust or a scratch absent from neighboring frames.

  • Replacing one damaged frame in a continuous shot.

  • Local repair using a soft mask.

  • Comparing overlaps for locked-camera and tracking shots.

  • Comparing CPU/GPU methods on the same defect.