Watch What Happens at the Point of Contact
A useful physical clue in polished AI-generated video is a mismatch between touch and consequence. A hand may appear to grasp a cup, for example, but the cup barely responds; a finger may press against a surface without the skin, object, or surrounding material reacting as expected. These moments can be subtle because the shapes look convincing while the action connecting them does not.
That connection is difficult to render. A video model generates a sequence of images that resemble examples it has learned from; it does not necessarily track a hand, cup, and tabletop as solid objects with fixed positions and forces. When the hand moves, the model must keep every part of the scene visually consistent from one frame to the next. At a point of contact, even a small mismatch can make an object seem to slide, soften, or pass slightly through another.
Why Motion Can Give It Away
A still frame may look perfectly plausible. The flaw becomes easier to notice across several frames, when viewers expect a grip to stay in place or a pushed object to move in response. Human perception is especially alert to these cause-and-effect cues: we know how a firm grasp, a bump, or a step should feel, even when we are not consciously checking the physics.
Malformed fingers can still be a warning sign, but they are no longer the only giveaway—and a strange-looking moment is not proof that a clip is AI-generated. The more revealing question is whether contact behaves consistently. In a convincing video, touching changes something; when the touch and the result drift apart, the scene can feel wrong before the viewer can explain why.
