The AI clipping and reframing struggles with reaction-style videos that use mixed layouts, such as split-screen (reactor on one side and source video on the other) and occasional full-screen face shots. The system often incorrectly focuses only on the reactor’s face, ignores the original video, or duplicates the same subject in both frames during splits. It seems the model lacks proper multi-subject and layout awareness, especially when transitioning between different scene formats. Improving dual-object tracking and context-aware framing (e.g., maintaining both subjects in split-screen and adjusting appropriately during full-screen speaking moments) would significantly enhance the output quality.