Blueprint the fight before motion

Watch the finished fight once without trying to decode every effect. Track only three things: who owns the frame, where the threat comes from, and which direction the next shot must face. The spectacle stays readable because those answers remain stable even while the camera, crowd, and armor change.

Read the result backward

The source does not begin motion generation with a single heroic image. It first locks references for the exhausted hero, the four-meter monster with its skull face and crystal blade, the enemy army, and the snowy location. The hero also gets a scene-specific sheet that shows pain and fatigue, because that state has to survive across the sequence.

At this stage, the crowd is one reusable samurai-army reference. The source does not split it into heavy and light classes until the first generated batch reveals cloned silhouettes. That correction belongs to the next lesson; preflight only establishes the army that the shot must preserve.

Give every reference one continuity job

Give each approved image one job you can test later. At preflight, that job is continuity: hold what has already been designed without claiming to solve a failure the source has not seen yet.

Reference	What it must hold
Exhausted Roko	Face, costume, body state, and fatigue
Monster	Scale, skull-like head, and arm-blade construction
Enemy army	The established samurai appearance before crowd variation is diagnosed
Crimson-tree view	The front-facing landmark and threat direction
Empty reverse view	What the camera sees when it turns back across the field
Fog wall	A bounded depth field rather than an infinite background

The two location views are the decisive move. One faces the crimson tree; the other is deliberately empty. Together they define the unseen side of the set. When the camera pulls back, arcs, or looks over a shoulder, it has a reference for the reverse geography instead of inventing a new landscape.

The proof is not that the board contains many images. It is that every recurring person, the enemy army, and each camera direction has one stable visual source before motion begins.