Most AI prompts describe what. Cinematic prompts describe how it feels. Here's the framework I use to control mood, light, and pacing in every generation.
The difference between a forgettable AI generation and a cinematic one rarely comes down to the model — it comes down to intent. When you approach a prompt the way a director approaches a shot list, everything changes.
Start with the emotional target
Before writing a single word of a prompt, I ask: what should the viewer feel in the first three seconds? Dread? Awe? Tenderness? That answer determines everything else — the lighting language, the camera movement, the color temperature, the pace of motion.
Most people start with nouns: “a woman walking through a city.” I start with adjectives and cinematographic intent: “melancholic, slow-motion, golden-hour backlight, shallow depth of field, 35mm grain, the kind of scene that plays over a film's ending credits.”
Build a language of light
Light is the primary language of cinema and it's just as powerful in AI generation. The prompt doesn't just need to describe light sources — it needs to describe what the light reveals and what it conceals. Hard rim light isolates a subject. Diffused overcast creates intimacy. Practical lights in the background create depth.
I keep a personal library of lighting descriptors that reliably produce specific moods in each platform I use. What works in Kling doesn't always translate to Runway — each model has its own aesthetic biases that you learn to work with, not against.
Control pacing through motion description
Pacing is time — and in AI video, you control time through explicit motion language. “Slow push-in” vs “handheld verité movement” vs “locked-off wide master” are fundamentally different emotional experiences. Most prompts ignore this entirely.
Camera movement carries narrative meaning. A slow dolly toward a character implies revelation or threat. A pull-back implies isolation or context. These are the grammar of cinema — and they work just as well in prompts as they do on set.