Start a project
WorkBlogResourcesAboutContact

Blog/Anatomy of a shot

What an 808-Word Duel Prompt Keeps After a 63 Percent Cut

How I cut a duel prompt from 2,164 words to 808 while keeping its three impacts, hard dawn light, sword technique, and unfinished exit.

By Jeff Zambrano··6 min read

What an 808-Word Duel Prompt Keeps After a 63 Percent Cut

I cut the duel prompt from 2,164 words to 808, a 63 percent reduction. The swords still had to meet three times. The shot still had to start in motion and end with another attack coming.

Those were the terms of the rewrite. If shortening the prompt changed the fight, I had shortened the wrong things.

The longer version was continuous prose, with repeated rules, uppercase instructions, and stars. Its swordsmanship direction appeared as four long, repeated laws. The shorter version used named blocks and concentrated that technique into a dense passage.

The work was deciding which sentences carried a visible requirement and which sentences merely asked for that requirement again.

I started with what the shot could not lose. Japanese swordsmanship, including grips and footwork. Hard dawn sunlight. Three clashes. A camera that catches fragments without revealing the entire duel. Motion already underway when the shot begins, and motion still underway when it ends.

That is a fairly demanding shot before anybody adds a style adjective.

The rewritten opening states the scene directly:

SCENE CONTEXT
Dawn duel on raked sand, already under way. Two swordsmen close the last four metres and their blades meet three times in fourteen seconds. The camera lives inside the fight and catches it only in fragments.

This paragraph contains the ground, time of day, starting state, distance, impact count, duration, and viewing restriction.

The last sentence carries a large part of the film direction. I need the camera inside the encounter. A clear view of the whole exchange would change the shot, even if the robes, blades, and sunlight were right.

That is why I would keep the fragment instruction before keeping another description of how intense the duel feels. The instruction tells me something I can check in the image.

The same test applies to the beginning. “Already under way” has a separate job from the detailed first frame. The context establishes the state of the scene. The blocking then makes that state visible: both men at a full run, mid-stride, weight forward, blades traveling, sand already airborne.

This is repetition with a purpose. Each block answers a different question.

Removing every repeated idea would be a poor editing rule. I need to remove duplicate explanations while preserving the relationship between the scene and its execution.

The largest structural change was giving the action a clock:

ACTION
0 to 4 seconds: both men cover the last four metres at running speed and the blades meet the first time, steel on steel, and slide apart.
4 to 8 seconds: the second exchange, closer, with the camera whipping across to the far man as his hips turn into the cut.
8 to 11 seconds: the third clash, the tightest of the three, the two guards locked for a fraction of a second at chest height.
11 to 14 seconds: they break and the next attack is already loading, and the shot ends with that movement still travelling.

The original prompt had no timings. The rewrite gives each exchange a place and reserves the final stretch for separation and the next attack.

That final stretch matters as much as the steel contact. Ending after the third clash would give the shot a different shape. I wanted the cut to interrupt movement that was continuing.

The shorter prompt protects that ending by describing what is physically happening at the boundary. The men break apart. Another attack loads. Movement travels into the last frame.

The impacts also remain distinct. The first contact slides apart. The second comes closer and connects the camera movement to a hip turn. The third brings the guards together at chest height. The count survives, but so does the difference between the contacts.

I did not compress all of that into “three fast sword clashes.” That would preserve a number while discarding the action.

The camera block carries another set of decisions: chest height, a distance of 1.5 to 4 metres, lateral whips between the men, and focus riding the near shoulder. It asks the camera to arrive a fraction late to each clash.

Those details explain how the fight becomes fragments. The scene context names the visual intention. The camera block gives it movement, position, and timing.

The format block then keeps those whips inside one continuous take. That distinction earns its space. A lateral whip is part of the shot’s camera movement.

Physics stayed concrete too:

PHYSICS
Every footfall throws sand that falls back under its own weight. Blades carry mass: they swing through and past the body, and the arms follow the steel rather than driving it. Robes lag behind the torso by a beat and settle. Real time at 24 frames per second, a shade faster than life.

Sand falls. Steel continues through its swing. Cloth follows the torso and settles.

Those are visible relationships. Each gives the movement something to answer to. Cutting them would leave the duel with its events intact but less direction about the material behavior connecting those events.

Sword technique received the same protection. I identified swordsmanship as the place where the model fails. That made it a poor place to chase further savings. I compressed the repeated laws into concentrated direction instead of treating every block as equally expendable.

Light became more specific as the prompt became shorter. I changed it from a broad hard, raking sunlight instruction to source, angle, temperature in kelvin, and airborne dust density.

I still kept the hard dawn sun and described its physical conditions.

The references stayed attached to the people, sword, and place that appear in the shot. Short descriptions identify the black robe, brown robe, worn sword grip, and raked ground. The reference instructions carry the matching requirement.

I also removed negative phrasing. The rewrite describes the required movement and continuity in positive terms. It gives the take a behavior to maintain.

My 63 percent reduction measures the text. It does not establish how faithfully a generation followed it. What I can inspect here is whether the shorter version still contains the shot.

Today, mark the instructions in your prompt that determine something visible in the first frame, each impact, and the last frame. Protect those while you cut the repeated explanations around them.


You just watched one shot come apart, decision by decision. None of them landed there by luck.

ShareXLinkedInWhatsApp
Jeff Zambrano

Jeff Zambrano is a Madrid-based AI film director. His debut, Borrowed Wounds, reached the Top 50 of the Higgsfield Global Film Festival out of 6,700 films. He writes these notes from his own productions.

WhatsApp