Draft in low res first.
It's way cheaper, and you see what's wrong before you pay full price.
How to tell an AI video model what the camera should do.
If you don't tell the camera what to do, the model guesses, and it guesses generic. This is the cheat sheet I use to stop that.
It's way cheaper, and you see what's wrong before you pay full price.
Change five and you'll never know which one fixed it.
Every shot needs a lens, an angle and a movement.
What it feels likeWide, you feel the space
Use it forEstablishing the place, big environments
What it feels likeNatural, documentary
Use it forTwo people in one frame, scenes that feel real
What it feels likeNeutral, like the human eye
Use it forMedium shots, dialogue
What it feels likeIntimate, flattering, soft background
Use it forClose-ups of faces, emotion
What it feels likeTiny details, very shallow focus
Use it forHands, objects, textures
Tension, focus on a person
Release, reveal, calm
Energy, follows a conversation
Feels like a real camera, not a render
Never use a digital zoom: it flattens the image and looks fake.
Honest, neutral
Power, confidence
Connection between two people
Overview, layout
Texture, detail, tension
SHOT 1 | 0.0–2.0 s | 35 mm, f/2.8, eye level, slow dolly push-in IN FRAME: [who and where, left/right] ACTION: [what happens, with body language and facial expression] AUDIO: [dialogue between quotation marks, ambience]
Two men sit side by side in a 2000s airplane cabin, 4:3, shot on 35 mm film, no film grain. Camera: handheld with subtle micro-vibration, never static, every shot has a slow smooth movement, never a digital zoom. SHOT 1 | 0.0–1.6 s | 85 mm, f/2.0, close-up of the young man in three-quarter profile, slow dolly push-in. He looks at the older man, calm and patient, and says: "Third. Tell the camera what to do." SHOT 2 | 1.6–3.9 s | 35 mm, f/2.8, eye level, two-shot from the aisle side, slow lateral slide. The older man leans over his laptop, curious, frowning, then slowly nods. AVOID: static locked-off camera, digital zoom, film grain, plastic skin, subtitles.



The characters, set and voices in these examples are AI-generated.