Spatial Anchors: Terrain, Light, and Time

Question

Spatial stability in AI video requires more than writing "the same room" in every prompt. The model does not know which physical facts "same" should preserve. It may keep the mood while changing the walls, preserve a sense of luxury while losing the table, or retain a character while reversing the light.

The key is to break an abstract setting into recognizable, repeatable anchors.

Observation

"Luxury auction room" names a setting type. These details establish a specific space:

surface: walnut wall panels, cream marble floor
layout: long dark table centered, display case behind auctioneer
light: warm amber key light from frame right
direction: entrance on frame left, windows behind the table
object: black velvet jewelry case stays on the table

Shots can repeatedly confirm these anchors. Seeing the same wall, table, and beam of light helps viewers recognize the same place. It also gives the model a structure within which to generate changes.

Method

I create an anchor card for each important space:

ROOM NAME: private jewelry auction room
WALL: walnut panels
FLOOR: cream marble
KEY PROP: black velvet jewelry display case
LIGHT: warm amber from frame right
SCREEN DIRECTION: door left, auction table center, window back
DO NOT CHANGE: table material, prop position, light direction

Each shot uses only the anchors it needs. A wide shot needs layout; a close-up needs the key prop and light; a reaction shot needs screen direction. Trim spatial anchors to the shot's task instead of inserting the entire card.

Boundary

Too many anchors can compete. The model may preserve decoration while losing the direction and props that matter to the edit. The most important anchors usually fall into three groups:

direction: relative positions of people and objects in the frame
light: where the light comes from
object: which prop identifies the space

Material and mood can receive less emphasis. Direction and key objects must remain coherent.

Lab Note

Judge a spatial prompt by whether it can answer these questions:

How will the next shot prove this is the same place?
Where does the character enter, and where do they move?
Which object must stay in place?
Would a change in light direction break continuity?

Without answers, the space is only a background, not a scene that can be edited.