When a camera moves past vertical lines, bending posts can look like a real rolling shutter. In a model test that requests a rigid scene, look for skew that changes direction or appears around a single post. Those failures affect whether a clip can match restrained live-action plates.
Build the test input
Draw an original empty warehouse with six columns, all equally spaced. Start with an image where the straight beams can be seen from top to bottom. Request one steady sideways camera track along the beam line, preserving the direction and apparent width of each column.
Run the proposed comparison
Run the same start image and motion brief in two image-to-video operations. Disable optional stabilization if only one result has it. Add a vertical guide on top of frames before, during, and after motion. Real perspective can tilt lines near a frame edge, so compare adjacent columns at similar placements.
Decide what passes
Log the top and bottom position for the same beam at three sampled frames. Mark local shape change separately from a coherent change across the scene. Reject a prominent column that buckles or divides in the middle, even if its start and end coordinates look reasonable.
Act on the outcome
A gentle, fixed slope across all columns may match the intended camera character. A jagged bend that moves to different columns is harder to conceal. Slowing the track or using a static plate with a separate crop moves the decision from model quality to an edit you can inspect.

