How to Write Cinematic AI Prompts
The difference between a flat AI render and a cinematic one is vocabulary. This is the prompt structure and the cinematography cheat sheet we use to direct generative models like a camera crew.
AI image and video models were trained on real film and photography, which means they already understand the language of cinema — most users just never speak it to them. Type "a man in a city" and you get a snapshot. Type the same subject with lens, lighting, and camera direction and you get a frame. This guide is the translation layer between what a director knows and what a model responds to.
The anatomy of a cinematic prompt
Every strong prompt follows the same skeleton, in order: subject, action, camera, lens, lighting, environment. Front-load the subject and action because models weight the start of a prompt most. Then layer the craft. Build the prompt the way you would brief a crew on set.
Build it in five moves
- 01
Lead with subject and action
State who and what first: "a lone astronaut steps onto red soil." Models weight the opening of the prompt most heavily.
- 02
Add the camera move
Specify movement explicitly — dolly, crane, pan, push-in — so the model animates intentionally rather than drifting.
- 03
Name the lens and format
Add "35mm," "anamorphic," "shallow depth of field," or a film stock like "Kodak Vision3" to lock the optical character.
- 04
Direct the lighting
Lighting carries the mood. "Low-key rim light," "golden hour," or "volumetric fog" change the emotional register more than any adjective.
- 05
Place it in an environment
Ground the shot: location, weather, time of day, and texture. End with the world the subject lives in.
Camera angles and movement
Translating physical camera work into prompt language is the highest-leverage skill. Use the real terms — the models know them.
| Criterion | Cinematography Term | What It Does | Use In Prompt |
|---|---|---|---|
| Dolly in | Dolly in | Camera physically moves toward subject | slow dolly in, steady push |
| Crane / jib | Crane shot | Sweeping vertical camera movement | crane shot rising over the scene |
| Whip pan | Whip pan | Fast horizontal blur between subjects | fast whip pan, motion blur |
| Rack focus | Rack focus | Focus shifts between two planes | rack focus from foreground to background |
| Low angle | Low-angle shot | Camera below eyeline, subject looms | low-angle shot, heroic framing |
| Dutch tilt | Dutch angle | Tilted horizon, unease | dutch tilt, off-kilter horizon |
Directing lighting
Lighting sets emotion faster than any other variable. Naming a lighting style does more work than a paragraph of mood description.
| Criterion | Lighting Term | Mood / Effect |
|---|---|---|
| Golden hour | Golden hour | Warm, soft, romantic low sun |
| Low-key | Low-key lighting | Deep shadows, high contrast, noir |
| High-key | High-key lighting | Bright, even, optimistic |
| Rim light | Rim / backlight | Glowing edge that separates subject from background |
| Volumetric | Volumetric fog / god rays | Visible light beams, atmosphere, depth |
| Chiaroscuro | Chiaroscuro | Dramatic light-dark interplay, painterly |
Film stocks and era styling
To pin a specific look, reference real capture media. "Kodak Vision3 500T" reads as warm, grainy filmic; "Alexa 65" reads as clean, high-dynamic-range digital; "Super 8" reads as soft, nostalgic, heavily grained. Era cues ("1970s film stock," "VHS aesthetic") layer period styling on top. These references are some of the most reliable steering tools in the whole toolkit.
Putting it to work
This vocabulary is the input to every generation stage — boards, stills, and motion alike. Apply it in the generative video workflow, and combine it with the consistency techniques in our character tutorial to direct both what the shot looks like and who is in it.