Most AI video prompts fail before generation even starts. Not because the model is bad. Because the prompt says: “Cinematic man running through a city. Realistic. 4K.” That describes a vibe. It does not direct a shot. A professional AI filmmaking prompt should answer: → What should the audience notice? → Who is in the frame? → What exactly are they doing? → Where is the camera? → What lens are we using? → How does the actor perform? → Where does the light come from? → What physically reacts to the action? → Where does this shot begin and end? The framework I use is: INTENT → SUBJECT + ACTION → FRAME → OPTICS → CAMERA → PERFORMANCE + PHYSICS → LIGHT + SPACE → CONTINUITY Here is what that actually means. 1. INTENT Before writing anything, decide what this shot is doing. Reveal information? Create tension? Show scale? Capture a reaction? If the shot has no purpose, camera language will not save it. 2. SUBJECT + ACTION Do not write: “Man is scared.” Write the behavior: “He freezes at the doorway, eyes tracking toward the sound, shoulders tightening before he slowly steps backward.” Prompt visible behavior. 3. FRAME Define what the camera sees. → Wide = geography and scale → Medium = action and relationships → Close-up = emotion and reaction → Insert = important detail Then add angle, height and composition only when they matter. 4. OPTICS Lens choice changes the image. → 24mm = environment, perspective, speed → 35mm = natural spatial energy → 50mm = balanced human perspective → 85mm = compression and intimacy → Macro = tactile detail Do not add “35mm” because it sounds cinematic. Know why it is there. 5. CAMERA Use one motivated camera behavior. Static. Push-in. Pull-back. Tracking. Handheld. Pan. Tilt. Crane. Orbit. The camera should move because the story gains something. 6. PERFORMANCE + PHYSICS This is where many AI shots collapse. Direct: → eyelines → breathing → hesitation → posture → weight shifts → foot contact → momentum → clothing response → object interaction → balance and recovery “Runs fast” is weak direction. “Accelerates, loses traction for half a step, catches his balance and keeps moving” gives the model physical information. 7. LIGHT + SPACE Describe motivated lighting. Not: “Beautiful cinematic lighting.” Instead: “Warm sodium light from camera-right, cool ambient rain fill, hard reflections across wet asphalt.” Then describe only the environmental details that affect the shot. 8. CONTINUITY Every shot exists between two states. Define: → start position → screen direction → costume → props → injuries → scale → geography → end position This matters even more when generating multiple clips. And one major distinction: Text-to-video and image-to-video should not be prompted the same way. Text-to-video needs more visual information. Image-to-video already has the character, composition, lighting and environment. So stop describing the reference image again. Prompt the motion: → camera motion → subject motion → performance → environmental motion → timing → continuity The biggest rule: One shot. One camera setup. One clear beat. Do not cram five actions, three camera moves and two emotional changes into an 8-second generation. PS:) Follow for more! . .