Talk Like a Director: The Shot Language That Makes AI Video Look Cinematic
Type “a man walking in a city, cinematic” into any AI video tool and you will get… something. Vaguely film-like and completely forgettable. Now type: “low-angle tracking shot of a man's worn boots striking wet pavement, neon signs reflecting in puddles, shallow depth of field, slow motion.” Same man. Same city. Different universe.
The difference isn't the tool. It is the language. AI video models were trained on films, so they respond to film language. Here is the vocabulary that turns you from a prompter into a director.
Shot sizes: how close is the camera?
This is the single most powerful thing you can put in a prompt. Learn these seven:
- Extreme wide shot — the landscape swallows the subject. “Extreme wide shot of a lone figure crossing a desert at dawn.”
- Wide shot — the full scene with context. Establishes where we are.
- Full shot — the person head to toe. Shows body language and costume.
- Medium shot — waist up. The workhorse of dialogue.
- Medium close-up — chest up. Interviews, intimate conversation.
- Close-up — the face fills the frame. Emotion lives here.
- Extreme close-up — an eye, a hand, a detail. “Extreme close-up of a trembling hand hovering over a phone screen.”
Rule of thumb: beginners live in medium shots; directors live in close-ups and wides. If your AI footage feels flat, you are probably prompting everything at the same distance.
Camera angles: where is the camera looking from?
- Eye level — neutral, honest. The default; use it deliberately, not accidentally.
- Low angle — the camera looks up. Makes subjects powerful, imposing, heroic.
- High angle — the camera looks down. Makes subjects small, vulnerable, observed.
- Bird's-eye view — straight down. God-like, detached, graphic.
- Dutch angle — the frame tilts. Unease, chaos, a world off balance.
- Over-the-shoulder — dialogue scenes. Puts us inside the conversation.
- POV — we see what the character sees. Immediate and immersive.
Camera movement: is the camera alive?
Static AI shots look like moving photographs. Movement makes them cinema:
- Static / locked-off — the camera doesn't move. Calm, observational, tense.
- Push-in (dolly in) — the camera moves toward the subject. Intensity builds. My most-used move.
- Pull-out — the camera retreats. Reveals context; beautiful for endings.
- Pan — the camera turns left or right. Follows action or reveals space.
- Tilt — the camera turns up or down. Reveals height and power.
- Tracking shot — the camera moves alongside the subject. “Tracking shot following a runner through morning streets.”
- Handheld — slight shake, documentary energy. Urgency and realism.
- Crane / jib — the camera sweeps up and over. Epic reveals.
- Orbit — the camera circles the subject. Drama, confrontation, awe.
One honest warning: don't stack movements. “Drone shot pulling back while orbiting and tilting up” is how you get melted geometry. One clear move per shot — that is how real directors work too.
Light: the cheapest special effect
- Golden hour — warm, low sun. Everything looks expensive.
- Blue hour — the deep-blue minutes after sunset. Melancholy, romance.
- Candlelight / practical light — warm, flickering, intimate. Faces glow.
- Neon — coloured urban night. Energy, nightlife, edge.
- Overcast soft light — gentle, shadowless. Honest, documentary feel.
- Harsh midday sun — hard shadows. Tension, exposure, desert truth.
- Chiaroscuro — dramatic light-dark contrast. Caravaggio with a camera.
- Silhouette — subject black against bright background. Mystery, iconography.
- Motivated light — light with a visible source in the scene. The difference between “lit” and “believable”.
Texture words: the finishing touch
Sprinkle, don't pour: shallow depth of field (dreamy blurred background), anamorphic bokeh (those stretched horizontal lens flares), film grain (takes the digital edge off), slow motion (gravity and grandeur). One or two per prompt is plenty.
The formula
Build every prompt in this order:
Subject + Shot size + Angle + Movement + Light + Mood.
Watch it work. Vague: “A woman waiting at a bus stop at night, cinematic.”
Directed: “Medium close-up, slight low angle, of a young woman waiting at a rain-soaked bus stop at night — static camera, neon sign glow on her face, shallow depth of field, quiet loneliness.”
The model didn't get smarter between those two prompts. You did.
Your homework
Pick three shots from films you love. Pause them. Translate each into one prompt using the formula — size, angle, movement, light, mood. Generate them. Compare. That exercise, repeated for a month, will do more for your AI filmmaking than any new tool release.
And if you want a director walking you through that translation, shot by shot, for five focused days — that is my AI Content Creation Training.
Directing a full scene: watch the formula work four times
Theory is nice. Here is a full scene — a woman receiving bad news by phone — directed shot by shot:
- Shot 1 (establish): “Wide shot, eye level, static camera: a small kitchen at blue hour, a woman alone at the table, city lights through the window, quiet melancholy.”
- Shot 2 (turn): “Medium close-up, slight low angle: the woman answers her phone, handheld camera with subtle drift, cold practical light from a single lamp, growing unease.”
- Shot 3 (impact): “Extreme close-up, static: her eyes widening as she hears the news, shallow depth of field, face half in shadow, stunned silence.”
- Shot 4 (aftermath): “Slow pull-out from close-up to wide, crane-like: the woman motionless at the table, the phone slipping from her hand, blue hour deepening outside, grief.”
Four shots. Four sizes, three angles, three movements, motivated light throughout, one emotional arc. That is directing — and every word of it fits in a prompt box.
Mistakes that scream amateur
- Same shot size every clip. Eight medium shots in a row is a slideshow, not a film. Vary your distances.
- Stacking camera moves. One clear movement per shot. Two moves is a gamble; three is a mess.
- Using “cinematic” as a crutch. It is the vaguest word in prompting. Replace it with the actual size, angle, movement and light you want.
- Ignoring light direction. If your prompt says “neon glow on her face” but the scene has no neon source, the model will fake it badly. Motivate every light.
- Forgetting the cut. AI filmmakers prompt shots; directors prompt sequences. Always know what shot comes before and after.