All Missions
W-21
Film Making · AI Pipeline
Fourteen Shots, One Blocking
Greybox Previz to 4K AI Film
A minute of finished war film, directed as grey blocks first — fourteen shots, five speaking parts, and not one of them left to chance in the prompt.
Client
Confidential
Year
2026
Sector
Film Making
Outcome
Fourteen Shots, One Blocking
Mission Brief
The objective, and how we met it.
// The Challenge
Prompt-driven video cannot hold a scene together: the camera wanders, the room changes shape, and a character is a different person in every shot. The question was whether a sequence with five named speakers and fourteen cuts could be generated at 4K and still read as one continuous scene shot by one crew.
// What We Deployed
Generated video drifts when you describe a scene instead of staging it. So the scene was staged. A greybox set was built in Blender — room, map table, five officers in full kit — and the whole sequence was shot in previz: fourteen cameras, real lenses from 28mm to 120mm, a continuous move across the map. That playblast is what drives the generation, as a video reference rather than a paragraph of adjectives, which is why the camera lands where it was put and the cast stay where they were stood. The officers never move: each is pinned to a fixed bearing around the table, so every shot is a different camera onto the same arrangement and a face in shot three is the same face in shot eleven. Voices were cast and recorded per character before a frame was generated, so performance drove the cut rather than being fitted to it afterwards. The previz cost nothing to run — it is a blockout, not a generation — and it is where every expensive mistake got made and fixed.
// BUILD
Before the performance
The character is made first.
A generated face drifts the moment the head turns unless there is something for it to be consistent with. These two plates are that something.
01 · Previz, labelled for review
Shot slate, lens, the five fixed positions and the line being spoken. This is the version a director marks up — every question about blocking gets settled here, where a take costs nothing.
02 · Previz, clean plate
The same frame with every label stripped. Burned-in text gets reproduced as artifacts in a video-to-video pass, so the model is given this one and the blocking is carried by the fixed positions instead.
03 · Generated at 4K
Same camera, same lens, same five people standing where they were put. The greybox decided the shot; the generation only had to light and dress it.
// THE FILM
Watch it end to end
Sector 4 — fourteen shots, one blocking
1:00 · 1920×1080 · 24 fps
Play the case film
Dialogue is performed; the score is original. Fiction throughout — no real operation or person is depicted.
// RESULTS
What It Proves
4K
Master · 24 fps
14
Shots, One Blocking
5
Speaking Parts
0
Credits On Previz
// SURVEILLANCE
Visual Record
Captured from the field.
01 PLATES · DECLASSIFIED
// SCOPE
Deliverables
What we shipped.
Disciplines
- Previsualisation
- AI Film Pipeline
- Sound Design
Tags
- Film
- AI
- Previz
01
Fourteen shots generated against a Blender previz, not a text prompt
02
Five characters on fixed bearings — identity holds across every cut
03
Voices cast and recorded before generation, so the cut follows performance
04
Previz runs on a blockout: no generation, no credits, unlimited takes
// LOG
Engagement Log
How the mission ran.
Every PAR2 engagement runs the same four-phase doctrine — adapted here to the specifics of this operation.
01
Recon
Mapped the terrain for Greybox Previz to 4K AI Film — film making · ai pipeline constraints, users, and the real objective behind the brief.
02
Strategy
Set the operating plan and success criteria, aligning previsualisation & ai film pipeline & sound design into a single line of attack.
03
Build
Engineered and iterated in the field — fourteen shots generated against a blender previz, not a text prompt, validated continuously.
04
Deploy
Shipped, monitored, and tuned. Fourteen Shots, One Blocking stands as the headline outcome of the deployment.