Camera Shots for AI Video explained
Filmmakers describe a shot with three choices. Shot size is how much of the subject fills the frame: extreme wide or establishing shot, wide or full shot (the whole body), medium shot (waist up), medium close-up (chest up), close-up (the face) and extreme close-up (the eyes or a small detail). Camera angle is where the lens sits relative to the subject: eye level, low angle looking up (power, scale), high angle looking down (vulnerability), overhead or bird's-eye, and Dutch tilt for unease. Movement is how the camera travels during the shot.
The common moves are pan (turn left or right on the spot), tilt (turn up or down), dolly in or push-in and dolly out or pull-out (move toward or away), truck (slide sideways), crane or pedestal (rise or fall), orbit or arc (circle the subject), tracking or follow shot (move along with a walking subject), zoom (change the focal length without moving), handheld (slight natural shake) and static or locked-off. Video models learn from very large amounts of captioned footage, so these established terms correspond to motion patterns the model has seen. A vague word like 'cinematic' does not tell it what to do with the camera.
Put the camera language near the start of the prompt and keep to one move per shot: 'Medium close-up, slow push-in, a chef plating a dessert, warm kitchen light.' Asking for a pan, a zoom and an orbit in five seconds usually produces muddled motion. For a sequence, plan several short shots, each with its own size and move, and cut them together, which is how edited film works anyway.