Why the Thigh Wound Was Replaced With an Arm Wound

The thigh wound had no reference plate.

That was enough to remove it from the fight.

An earlier version put marks on Sadao’s arm and ribs, and on Toshiro’s thigh. The revised version moved them to Sadao’s shoulder and ribs, and Toshiro’s arm. This was not a story preference. It followed what I had already generated.

I wrote down the change plainly:

La versión anterior ponía las heridas en el brazo de Sadao, el muslo de Toshiro y las
costillas de Sadao. **Se cambiaron a hombro, brazo y costillas**, y el motivo no es de gusto:

The available plates were uneven. Sadao had plates for his shoulder, chest, neck, collarbone, and ribs. Toshiro had a plate for his arm. There was no thigh plate.

Once I looked at that inventory, the decision stopped being abstract. If Toshiro received the mark on his thigh, I would have to create a new plate and hope it matched. If he received it on his arm, I could point the prompt to an existing reference.

That difference mattered more than the original placement.

The reference library became a production constraint

It is easy to treat a wound location as a sentence-level choice. A character gets hit here rather than there. Change one body part in the prompt and keep writing.

That was not how this shot worked.

The location had to survive beyond the line that described the contact. The same mark had to appear in this shot and the next. For that, I needed more than the word “thigh.” I needed an image that showed the mark and could be named directly in the prompt.

The project already had wound plates in clean and open states for the supported areas. It did not have either state for Toshiro’s thigh. What I had already generated therefore set the practical boundary of the scene.

This was my rule:

Poner una herida en el muslo obligaba a generar una lámina nueva y a rezar para que
cuadrara. Poniéndolas donde ya hay lámina, **el prompt puede señalar la referencia exacta**
(`@character-kiri-hombro-abierto`) y decirle al modelo cómo tiene que verse esa marca. Eso
es lo que hace que la herida sea la misma en este plano y en el siguiente.

The phrase “and hope it matched” is the whole problem. A newly generated thigh plate might have worked. Using it would have added a new generation requirement and a new matching problem.

The arm plate already existed. The thigh plate did not. I chose the supported mark.

The token carries the visual decision

The most useful part of the revised setup was not a longer description. It was the exact reference token:

@character-kiri-hombro-abierto

That token points to the open shoulder plate for Sadao. The prompt can use it to tell the model how the mark should look.

This is narrower than asking the model to interpret a general wound description. The location is already chosen. The visible state is already represented. The prompt has something specific to indicate when the mark needs to carry into another shot.

That is why I now separate two decisions that look like one.

The first decision is dramatic: where does the character receive the mark?

The second is material: do I already have the plate for that character, body area, and state?

If the second answer is no, the first decision is not ready. I either have to make the missing plate or move the mark to a supported area. In this case, I moved it.

What the original thigh wound required

Keeping the thigh placement would have created work outside the fight prompt. I would first need a new thigh plate. Then that plate would have to match the existing character plates closely enough to use.

I did not treat that route as impossible. The problem was simpler: there was no existing thigh plate, so the route depended on generating one and hoping it fit.

That is not the same production position as selecting Toshiro’s arm, where the plate was already available.

I do not need every possible wound location in advance. I need the locations the sequence actually uses. But once the shot depends on continuity across cuts, unsupported placement is no longer just descriptive freedom. It is an asset gap.

The earlier prompt treated the thigh as available because it could be named. The revision treated it as unavailable because it could not be referenced.

That was the better definition.

The replacement was specific to each character

I could not solve the problem by selecting any wound plate from the library. The plate had to belong to the correct character and the correct area.

Sadao’s supported areas were shoulder, chest, neck, collarbone, and ribs. Toshiro’s supported area was his arm. The revised shoulder, arm, and ribs arrangement follows that division.

This matters because “we have a shoulder plate” is incomplete. The useful statement is “we have Sadao’s shoulder plate.” The character and placement are one reference decision.

The same applies to Toshiro’s arm. That plate made the arm a usable choice for him. Sadao’s larger set of supported areas did not make those areas available to Toshiro.

The plate inventory was not interchangeable.

What done looked like

The wound decision was done when the prompt could point to an exact existing plate for the mark, and that plate could define how it should look across this shot and the next.

That is a more demanding finish line than having a body part written into the action. A line of prose can establish where the hit lands. It cannot replace a reference that does not exist.

For this fight, the clean decision was to use shoulder, arm, and ribs because those plates were already on disk. The thigh failed that test.

I did not keep it because it sounded better in the action. I removed it because the visual evidence required to carry it was missing.

Today, audit the wound plates you already have before revising your next fight prompt. Match every planned mark to the correct character, body area, and available state. If a mark cannot point to an existing plate, either make that plate first or move the mark.

---

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