How to Test AI Film Generation for Shot Continuity

Choose the model that produces a usable sequence for your edit. A three-shot test shows what survives a cut between a wide view, an insert and a medium shot, including details that can look convincing in a single clip.

Worked example

Three shots with one repeated prop

Fixed details: yellow raincoat, blue umbrella with wooden handle, light from the shop window at frame left.
Wide: the subject walks left to right and stops at the shop door.
Insert: the right hand closes the umbrella at the door.
Medium: the subject sets the closed blue umbrella beside the door.

Score the cut from the wide shot to the insert

Blue umbrella and wood handle match: pass / fail
Right hand and movement stay consistent: pass / fail
Window lighting stays on the same side relative to the scene: pass / fail
Takes used: ___

Steps

  1. Write fixed details for wardrobe, props, light direction and movement. Prepare one reference still for each planned angle, using the same subject and location.
  2. Use the same approved references and shot brief for each model. Where a model accepts an input image, describe the action and camera movement separately in the text prompt. Log the number of tries you allow per shot.
  3. Cut the three takes together at the same frame rate. Score prop identity, screen direction and lighting across each cut, then count how many takes you needed to get an acceptable sequence.

What to check

  • Check the frames on both sides of each cut. If a test misses the hand or handle detail, record that as a failure even when the motion looks smooth.
  • Track the accepted output count and the total attempts at the same time. This makes the cost of reshoots visible.

Compare video generation models

References