From Script to Shot Flow

Question

A script provides intent; shot flow provides visual states ordered in time. One fragile point in an AI video workflow is sending a plot description directly to the model and letting it choose the shot order.

The resulting clip may look good but be difficult to edit. It does not clearly establish what the previous shot left behind, what the next shot inherits, or when the audience should receive information.

Observation

A narrative beat needs to pass through three layers before it becomes a prompt:

emotional turn: the character moves from defensiveness to hesitation
visual evidence: she sees the old letter on the table
shot change: wide room context -> medium reaction -> close-up letter

AI video models are good at filling in images, but less reliable at deciding narrative logic. Shot flow fixes that logic before generation.

Method

I write a beat table before the prompts:

Beat 1: She enters the room, establishing the spatial relationships.
Shot: locked-off wide shot, door frame left, table center.

Beat 2: She notices the letter and slows down.
Shot: medium shot, slow push-in, her hand stops above the table.

Beat 3: The letter becomes the center of information.
Shot: close-up, letter edge under warm light, hand partially in frame.

Each shot leaves the state needed by the next. The door, table, letter, and character position are editing connections as well as background details.

Boundary

More detailed shot flow is not always better. Overloading a shot asks the model to handle character, emotion, action, shot size, light, props, and complex transitions simultaneously. A better approach gives each shot one narrative task.

one shot, one turn
one shot, one camera behavior
one shot, one inherited state

Lab Note

When moving from script to shots, start with these questions before asking how to turn a passage into video:

What must the audience learn in this shot?
What must the model remember in the next shot?
The intersection of those questions is shot design.