Cinematic prompts · Part 2 of 5

The prompt operating systems AI filmmakers actually use

Six stacks and one psychology pre-brief. Pick the paper the job needs.

A directed shot is one sentence. An operating system is how you repeat that sentence without inventing a new movie every take. Pick a stack for the job, write the same scene inside it, and do not mix two stacks in one generation. The camera still has one job. The OS only changes how much you write around it.

What is a prompt OS?

A prompt OS is a fixed order of fields. You fill the fields. You do not free-write a paragraph and hope grammar appears.

Six generation stacks. One pre-brief you can drop into ChatGPT or Claude before any of them.

A. DP stack — size, angle, move, subject, lens, light, reveal
B. Clark script — short script the camera sees and feels
C. Progression lock — same people and room, new action and camera each beat
D. Seedance stack — scene, one move, tempo, psychological effect
E. JSON packet — shot, cinematography, and audio as separate fields
F. Short school — 30–60 words, one subject, one action, one named move

G is not a seventh generation format. It is a psychology lock you load first so every stack still encodes power, fear, or connection. Emily

The Godfather — restaurant, System A
Medium close-up, eye level. Locked-off static shot. Michael sits with his hands in his lap. 50mm, soft overhead. The stillness reveals the lie is over.

Goodfellas — Copa walk, System B
A warm, long-take medium wide follows Henry and Karen through a kitchen into a packed club. He steers her by the wrist. The camera tracks beside them, matching pace, never leaving their shoulders. 35mm. The track reveals the back door is how he belongs here.

What to start with

The Director’s sentence from the last lesson
The job of the clip (hero product, emotional beat, four-shot scene, fight)
Whether face and room must match a prior frame
The model you will paste into

The six stacks

A. DP stack
Use for a single hero shot, an ad, an emotional close-up.

Runway formula

Paste this skeleton

[Shot size] + [Angle] + [Move + direction + speed] + [Subject and action] + [Lens] + [Light] + [What it reveals]

B. Clark script
Use for a 10–15 second scene that has to feel directed, especially Veo.

Order: hook and shot type, who/what/where with exact verbs, light and texture, subject motion on its own line, camera motion on its own line, one composition term. Aim 120–180 words. Restate wardrobe and room every time. The model has no memory. Clark thread

Paste this skeleton

[Hook + shot type].
[Who, what, where, with exact verbs].
[Light and texture].
Subject motion: [what the body does].
Camera motion: [one named move, its speed, where it stops].
[One composition term].
Wardrobe and room, restated: [coat, props, place].

C. Progression lock
Use when one pair of people has to carry four beats.

Lock Character and Setting. Rotate Action, Camera, and Emotion. Breakfast, walk, wait, night is four cameras, not four new movies. Magiclight

Paste this skeleton

Same [character], same [wardrobe], same [room].
Action: [what changes this beat].
Camera: [shot size, angle, one named move, lens].
Emotion: [the inner state this beat lands on].

D. Seedance stack
Use when the move itself is the feeling.

Example they publish: a soldier collapses, the camera retreats, the retreat slows to stillness, an ultra-wide leaves him isolated. Seedance 80 · Kling camera verbs

Paste this skeleton

[Scene] + [one camera move] + [speed or rhythm] + [psychological effect]

E. JSON packet
Use for short ads, horror inserts, anything that breaks when prose gets novelistic.

Separate fields for shot, cinematography, and audio. Describe the shot the way a DP would, not a novelist. Genflick

Paste this skeleton

shot:
cinematography:
audio:
reveals:

F. Short school
Use when a long prompt invents a second movie, or the clip is already drunk.

30–60 words. One subject. One action. One named move. Sound only if it is in the room. imagine

Paste this skeleton

[Subject] [one action] in [place]. [Shot size]. [One named move]. [Lens]. [What it reveals].
30-60 words. Sound only if it is in the room. No second location. No second move.

G. Psychology lock (pre-brief)
Drop this into chat before you write A–F.

Low angle is power or threat. High angle is smallness. Dolly zoom is dread. Handheld is chaos. OTS is connection. ECU is feeling. 24–35mm is space or paranoia. 50mm is plain sight. 85mm and up is isolation. The camera must encode an inner state. Emily · Cinematic Prompting Guide

Pick a stack by job

8-second product hero: A or F
15-second emotional scene: B or D
Four-beat short: C (then lock face and room in the continuity lesson)
Fight or one-shot: B, with one move per beat
You are stuck: run G in chat, then trim the result to F

Worked swap (same rain, six operating systems)

Engine: a detective lights a cigarette in the rain. We should learn he is not as calm as the smoke.

A
Medium close-up, low angle. Slow dolly push-in over 4 seconds. A detective lights a cigarette in the rain. 35mm, shallow DOF, wet neon on the pavement. The push-in reveals the tension in his face.

B
A tense noir medium close-up finds a detective in alley rain at night. Navy coat, wet shoulders, unlit cigarette. He strikes the match and cups it against the wind. Neon breaks up on black puddles. He does not look up. The camera starts at his chest and dollies in for four seconds, no zoom, and stops on his eyes. 35mm, shallow DOF. The move reveals the calm is a performance.

C (beat 2 of 4, coat and alley already locked)
Same detective, same navy coat, same alley. He lights the cigarette. Medium close-up, low angle, slow dolly in, 35mm. Emotion: the lie is slipping.

D
A detective lights a cigarette in alley rain. Slow dolly push-in. Four seconds, even rhythm, then hold. The move tightens into his face until the calm reads as fear.

shot: medium close-up, low angle, detective cupping a match in the rain
cinematography: slow dolly in 4s, 35mm, shallow DOF, wet neon, no second move
audio: rain on coat, match flare, distant traffic
reveals: the calm is fake

F
Detective lights a cigarette in the rain. Medium close-up. Slow dolly in. 35mm. We see the fear in his face. Rain audio only.

Point at what changed: the paper. The engine did not. If System B adds a second location or a partner in the car, you left the OS and started a new pile.

Anti-patterns

Camera clause before the subject. Name who does what first. vflow
Two unrelated moves in one prompt. Kling in particular wants one verb. Slow dolly in only. VIDEOAI.ME · Kapwing
No camera movement. Models fight that negative. Write Locked-off static shot. Camera fixed on a tripod. Runway Gen-4

Kill test

Which stack did I pick, and why that job?
Can I point at one named move?
Did I restate coat and room if this is not shot one?
Did System B wander past 180 words into plot?
If I strip this back to F, does the shot still have a reveal?

If the last answer is no, you wrote a paragraph. You did not direct a shot.

Using AI as a writing partner

Load G first if you want the psychology map in play. Then force one stack.

Paste into your AI chat

Act as my DP, not a novelist. Do not add characters, locations, or plot.

Inputs: [Director’s sentence or sloppy prompt] · [job: hero / emotion / 4-beat / drunk-clip] · [model]

1. Pick one stack only: A DP, B Clark, C progression, D Seedance, E JSON, or F short.
2. Rewrite the shot in that stack. One named move. Subject and action before camera. What the move reveals must be explicit.
3. If I said this is beat 2 or later, lock the same coat and room. Change only action, camera, and emotion.
4. Then write the same shot in F (30–60 words) so I can compare.
5. Attack both. Where did a second move sneak in? Where did prose invent story? Do not pick a winner for me.

Next lesson is the move library. Each move gets a story job. This lesson is which paper you write that job on.