Why does the number of products change during an AI multipack video?

Define the counted set

Record the actual pack quantity and distinguish merchandise units from props or empty display packaging. Choose a starting image in which the count is inspectable. Overlapping identical items can make it hard to tell whether one has disappeared or is merely hidden.

For an illustrative three-jar bundle, give the review a simple positional map: left, center and right. The map is for human inspection, not an CartVisual object-tracking capability. Use the actual supplied variants as the source of truth.

Review count at transitions

A temporary overlap is not automatically a defect. Follow the unit’s position before and after the overlap. If the relationship cannot be understood, simplify the action.

Keep the bundle motion restrained

In Product videos, plan one movement around an approved bundle first frame. Avoid asking every identical object to spin, jump and exchange positions in a short shot. A subtle camera move can preserve the pack explanation better than choreography. Read the proposed storyboard before confirming generation.

After generation, inspect the whole clip, not just sampled attractive frames. This method requires a reviewer to verify the actual delivered footage; it does not establish automatic inventory verification. Use actual bundle footage when a precise inventory demonstration cannot be reviewed confidently.

Check the final quantity message

Read captions and nearby purchase text with the clip. A decorative scene containing extra jars can contradict a two-pack listing even if its animation is smooth. Remove unsupported extras or choose a shot with an unmistakable supplied set. Do not fix a disappearing unit by changing the quantity claim.

Use bundle planning for the source image and occlusion review for hidden moments. Preserve the count map and review result with the released video.

Fictional three-jar set has A-left, B-center and C-right in the approved source. During a camera move, B becomes hidden behind A and later reappears. The positional names are review aids, not tracking output or unique markings added to the merchandise.

Three endpoint objects do not prove every transition was sound: two jars may briefly fuse or a label may switch. Inspect those relationships in the full proposed edit. An illustrative label cannot resolve a picture that contradicts what is supplied. The jar set and review observations are invented examples, not generated footage or verified inventory.