5 Constraints for Better AI Video Prompts

1. RECORDING SYSTEM Put the camera language right at the beginning: "12-second vertical 9:16 iPhone video, one continuous handheld macro shot, harsh phone flash, low-light compression, autofocus breathing." Include format, framing, optics, and camera behavior before the action starts. Avoid vague terms like "cinematic macro shot." 2. ANCHOR POINTS Anchor motion to precise starting locations. Example: "Two finger-widths above the wrist crease is a two-inch faded gecko tattoo." Pick another fixed point in the scene (e.g., a freckle). Throughout the shot, these anchors stay exactly where they started. Describe movement relative to something that isn't supposed to move. 3. INCLUSIONARY FRAMING Describe what should actually be in frame instead of listing what you don't want. Example: "Only his forearm, empty hand and shirt edge enter frame." Avoid negative phrasing like "Show only the arm and not the whole body. No full body shots." Tell the model exactly what the camera can see and leave the rest out entirely. 4. LAYER SEPARATION When something moves out of a detailed surface, maintain surface continuity. Example: "Pores, fine hairs and skin sheen continue across the ink." This gives the surface its own continuity and the moving element its own motion. Use for fabric, scratches, wood grain, rust, dirt, etc. 5. ACTION LOOPS Write out the entire sequence instead of using quick verbs. Example for a tattoo coming alive: - flat faded tattoo - gains depth - peels from skin - becomes a living gecko - catches the fly - returns to its original coordinates - presses flat - becomes faded ink again Include a small consequence for the final state: "a slightly rounder inked belly and one fly wing." Writing the full loop ties in an end state and improves realism and prompt adherence. When a generation keeps going wrong, look at what you left too open and need to remove rather than what else you can add. Subtract before adding.