Interpolation is worth keeping only when it makes motion easier to follow without adding distracting duplicate shapes. Compare the untouched clip and interpolated version at the same playback speed. Look closely at hands, object edges, and places where one thing passes in front of another. If new ghost images are obvious, keep the original: interpolation estimates in-between images; it cannot recover an accurate frame that was never visible.
Tell source motion from added-frame ghosts
Choppy movement may already be present in a generated clip: motion can jump between positions or feel uneven from frame to frame. Motion blur can smear a moving hand or edge in the original image. Interpolation artifacts are a different concern: an estimated in-between image may show a doubled outline, a warped shape, or a translucent trail, especially when movement is quick or one object occludes another. These appearances can overlap, so inspect the untouched clip before deciding what caused them.
Checkpoint 1: If the same smear or jump is already visible in the untouched clip, do not credit interpolation with creating it. If the doubled shape appears only in the processed version, treat it as a likely interpolation artifact.
Compare the clips without changing the viewing conditions
Keep the original file unchanged and compare the same short section in both versions. A speed or crop difference can make one seem smoother even when the underlying motion has not improved.
- Choose a section where something moves across the frame, ideally with a hand or edge passing in front of another shape.
- Play both versions at the same playback speed and with the same crop. Watch once at normal speed to judge whether the movement reads naturally.
- Replay the section and pause at several moments around the crossing. Inspect fingers, outlines, and the boundary between the foreground object and what it briefly hides.
- Note what changes in the processed copy: smoother transitions, new duplicate edges, or shapes that appear to melt together.
Checkpoint 2: Keep interpolation under consideration only if the motion is easier to follow and the new outlines do not distract at normal playback. Smoothness alone is not a pass if important shapes visibly split.
Fictional planning example: Lena’s paper-cup handoff
This is a fictional planning example, not a generation experiment. Lena is reviewing a short clip in which a person passes a paper cup from one hand to another. She compares the untouched and interpolated versions at the same speed. Her observable checks are whether the cup rim stays a single clear outline, whether each visible finger has one readable edge, and whether the cup remains distinct as the hands cross.
If the processed copy adds a second rim or a finger-shaped trail at the crossing, she keeps the untouched clip for that passage. If the original already has a smeared hand, she records that separately rather than treating it as proof that interpolation caused the blur. The acceptance test is visual and specific: the handoff should remain readable without new duplicate contours.
Decide whether to keep, limit, or skip it
Interpolation estimates motion between available images; it does not reveal hidden, accurate frames. When one object covers another, the missing area may not be present in either visible frame, so an estimate can invent a plausible-looking but inaccurate edge. Keep the original as a fallback and use the processed version only where its visible motion is clearer without distracting ghosts. If the issue is confined to one passage, an external editor can handle final video assembly when needed.
Imagild’s Film workspace is /en/film-studio. Its current Skill catalog at /en/skills includes source/result examples; check which controls are currently available instead of assuming a particular interpolation option. This procedure is advice for comparing versions, not evidence that a particular control was tested.
Does interpolation restore frames that the clip never showed?
No. It estimates in-between images from available visual information and cannot recover a hidden, accurate frame.
Should I judge ghosting on a paused frame or during playback?
Use both. Playback shows whether motion reads naturally; pausing around a crossing helps reveal doubled edges that are easy to miss at speed.
What if the original is already blurry?
Keep the original as your reference and separate its existing blur from any new trails or doubled shapes in the processed copy.