Photo animation fails in a specific place, and it is rarely where people look. The uncanny feeling almost never comes from the motion. It comes from mismatch between what moves and what should move.
Depth decides whether it is a scene or a cutout
If the depth estimate is wrong, the background drifts with the subject. The image stops reading as a photograph of a space and starts reading as a card in front of another card. The brain notices immediately, even when individual frames are clean.
Identity decides whether it is the same person
Models that condition only on a first frame drift within seconds. No single frame looks broken, but the sequence does, because the face is subtly different every time it turns.
Edges decide whether it looks produced
Hair, fingers, thin structures and text are where generative filling invents detail, and invented detail is inconsistent across frames. That inconsistency is what people mean when they say something looks like AI.
The practical conclusion: a pipeline is judged by its weakest stage. Model and template pages organised by which stage each option addresses are here: https://pic2videoai.com/animate-photo
Top comments (0)