How to prompt undetectable AI shots

1. Start with the narrative goal of the shot. Camera movement is not just movement, it is the storytelling. Define what the shot is trying to do. For example, in a 15-second one take, the goal could be: follow the female lead laterally while she runs, to build speed and tension, then briefly reveal the people chasing her, then let the camera hesitate for a moment and find her again. That small 'lose and recapture' moment adds spatial depth and makes the chase feel more intense. 2. Build a clear space for the camera to work in. Break the scene into layers so the model knows where everything belongs. Foreground: passing objects, environmental motion. Main subject layer: the woman running. Midground: cafe tables, pedestrians, the people chasing her. Background: the vanishing point of the street, and the entrance to the pedestrian area. Once the space is clear, the camera has a stage to move through. 3. Describe motion like a physical process. A good moving shot has to respect inertia. Instead of using broad words, describe a chain of actions the camera can actually perform: what accelerates, what slows down, when it adjusts, when it slightly overshoots, when it catches itself again. That little bit of imperfection is usually what makes it feel real. 4. Use focus as part of the storytelling. Design moments where focus shifts with intention. For example: focus briefly drifts from the chasers' faces, passes beyond them, lands on the woman in the distance, then quickly pulls back again. That kind of temporary focus loss and recovery adds a lot of immediacy and documentary feeling. 5. Build the sound space with the camera. Sound should move with the shot. When the camera turns toward the chasers, the woman's breathing should fall deeper into the sound field, while the chasers' footsteps and breathing move to the center. When the camera finds the woman again, her breath becomes the main sound again. 6. Use negative constraints to stop common AI mistakes. Add clear 'don'ts' at the end to stop the model from breaking the shot. For example: no cuts, no teleporting zooms, no sliding characters, no body fusion, no floating props, the travel bag must keep believable weight and inertia. These negative constraints act like guardrails to help keep the result inside a believable physical world.