A lot of image to video clips look fake for the same reason: the motion design asks the model to invent too much. After testing a bunch of workflows, these are the patterns I keep coming back to: Portraits usually need less motion, not more. A subtle blink, slight head turn, or tiny hair movement is often enough. Big motion makes identity drift show up fast. Camera motion and subject motion should not compete. If the face is moving, keep the camera calm. If the camera is pushing in, ask less from the subject. Source image quality matters more than people expect. Compression artifacts, messy backgrounds, and unclear edges make motion worse. Clean inputs give the model less guessing to do. Old photos work best when you stay conservative. The best results are usually small expressions, soft eye movement, and minimal environmental motion. Trying to make an old photo feel cinematic often breaks the illusion. Aspect ratio changes the feeling of motion.…