A paper by Cutib has been accepted to the main track of ACM Multimedia 2026.
This research forms the technical foundation of OTACU, which automatically checks the quality of AI-generated video. Below we introduce the background and problem definition, the proposed method and the experimental results.
Background
As the number of cuts in generative AI video grows, so do errors in characters’ appearance, the state of props and the continuity between scenes. Checking every cut by hand increases production time and cost.
Method and results
We align the video, script and reference assets on a timeline and compare the elements expected in each scene with the actual generated result to find error candidates. Detailed experimental results are available in the full paper.

