Cinematic prompts · Part 5 of 5
How the best people block ambitious shots
Chain the beats: cause, reaction, new geography. One camera sentence per beat, a motivated change of view, and a written end frame.
An ambitious shot is still one job per beat. The difference is the beats are chained. Write the scene as cause, reaction, new geography. Give each beat one camera sentence. Motivate every change of view with a body, a whip, or an object crossing the lens. End-frame the last image so the next generate has a place to land.
What is a block?
A block is the shot planned in space before anyone hits Generate. Who stands where. What the camera is allowed to do. What earns a change of view. AI flipped the old order into prompt-then-pray. The people whose clips hold together put the order back.
Traditional: block the space, place the bodies, plan the camera, shoot.
Working AI version: lock face and room, write cause and reaction, assign one move per beat, generate, land on a described end frame.
Goodfellas — the Copa walk
Cause: he wants her inside his life.
Reaction: the staff parts.
New geography: kitchen becomes club.
Camera job: stay with them. One track. No extra move.
Touch of Evil — the opening crane
Cause: a bomb goes into a car.
Reaction: the street keeps living.
New geography: the crossing.
Camera job: one long rise and travel that makes time feel continuous.
Both shots are ambitious because the camera has a path with a reason. They are not ambitious because they stack five verbs.
What to start with
- The four locks from the last lesson (face, coat, room, key)
- The scene as three beats: cause, reaction, new geography
- One named move per beat
- How the view will change (a turn, a whip, a body across the lens)
- An end frame in words
If you cannot say the end frame, you cannot start the next generate.
Five patterns
1. The camera never leaves the person
Umesh’s museum one-shot: the world morphs from painting to flood to ship to desert to war while the lens stays on the guard. Shared shapes sell the transitions. The reusable rule is continuity of attention, not continuity of location. Umesh
Use when the job is “this is still her story” even as the room dies.
2. Motivated POV handoffs
Open third person. The camera sees the threat first. She spins. Whip pan into her eyes. A body crosses the lens to get back out. The change is physics, not “now cut to POV.” Clark
Use when we must know something she does not, then join her.
3. Scale tells the feeling
Clark’s Titan sequence: low canopy wide so he is a god, push to the face so he hears, pan with his gaze, bird’s-eye so he is an insect, dive and track the run, pull back as two forces meet. Screenwriting and camera technique are the same list. Clark
Use when power is the point. Low makes him large. High makes him small. Do not skip the middle.
4. Plan the path in space first
If the fight or the room is complex, block it before the model invents geography. Clark’s hybrid work puts a real rack focus and a real space under the generate. The camera already knows the distance. Clark hybrid
Kōda’s still-to-prompt pass is the cheap version of the same idea. Steal a composition that already exists in space. Then generate. Kōda
5. One move per world, even in a one-take
A door that opens onto five worlds only holds if each world gets its own camera sentence. Dolly back in world one. Low-to-orbit in world two. Do not write five moves in one paragraph and hope they sequence. Umesh is the clean proof: attention stays on the person, the world changes around that attention. Umesh
PJ Ace’s Odyssey pipeline is the production version. Short prompts. Named moves. Edit while you are still generating. A bad first frame is not rescued by a longer paragraph. Imagine thread · PJ Ace
The directing template
- Write the scene as cause → reaction → new geography.
- Assign each beat one camera sentence.
- Motivate every perspective change with a body, a whip, or an object crossing the lens.
- End-frame the last image.
- Keep the four locks identical across beats.
That is the whole advanced lesson. Longer prompts do not make it more advanced.
Worked swap (same detective, three beats)
Engine: alley rain, navy coat, cigarette. Now something arrives.
Beat 1 — cause (we see it first)He lights the cigarette. Medium wide, eye level. Locked-off static. 35mm. At the far end of the alley a car door opens. The stillness reveals he has not noticed.
Beat 2 — reaction (handoff)He turns toward the sound. Whip pan with his turn. Land in a close eyeline. Handheld, slight operator instability. The whip reveals we now know only what he knows.
Beat 3 — new geography (scale)He walks toward the car. Low angle, medium shot. Slow dolly out as he approaches. Camera physically retreats. 35mm. The pull-back reveals the alley is longer than the job.
Shared lock on all three: same face still, same navy coat, same alley reference, key from camera left. Point at what changed: the job of the camera. If beat 3 grows a new street or a dry coat, the block failed.
Kill test
- Can I mark cause, reaction, and new geography on the page?
- Does each beat have one named move?
- What physical thing earns the POV change?
- Is the end frame written in nouns I can reuse?
- Did I ask the model to change worlds and camera grammar in the same sentence?
If the last answer is yes, split the generate.
Using AI as a writing partner
Use chat to block. Use the video model to execute one beat at a time.
Paste into your AI chat
Paste beat 1 only into Veo, Kling, Seedance, Runway, or Sora. Then beat 2 with the end frame of beat 1 as the landing pad. Ambitious means chained. It does not mean crowded.
Next lesson is a one-page cheat sheet for how those same sentences behave in each model.