How to plan coverage so your AI scenes cut together instead of falling apart

Most AI filmmakers work the way a nervous student does on the first day of a film set: they generate what feels interesting in the moment, then try to assemble those moments into a scene later. The footage looks fine in isolation. Then the edit arrives and nothing joins. The eyelines are wrong. The screen direction flips. Two shots that should feel like the same conversation feel like they were filmed in different buildings on different days.

The problem is not your prompts. It is that you are thinking about shots instead of thinking about coverage. Coverage is not a list of angles. It is a plan for how a scene will be understood from every necessary point of view, with enough material to give the edit options without losing spatial coherence. Cinematographers plan it before the camera rolls. You need to plan it before you generate a single frame, because unlike a real set, you cannot call for one more pick-up at the end of the day.

Start with the geography of the scene, not the look of it

Before you write a single prompt, draw or write out a top-down description of the scene space. Where is each character standing? Where is the camera relative to them? Which direction is north, figuratively speaking, so that you can keep your 180-degree line consistent across every shot?

You do not need a formal diagram. A rough written note works perfectly well.

```
SCENE 04 GEOGRAPHY
- Two characters: A (left of frame), B (right of frame)
- Camera line runs roughly east-west between them
- All OTS shots: camera stays south of this line
- Window is behind B, providing practical light source
- Door is frame-left in wides, frame-right in close-ups would break the line: avoid
```

This note becomes the rule-set you consult before every generation. If a prompt would produce a shot that breaks the line, you rewrite it first.

Build your shot hierarchy for the scene before generating anything

A standard coverage plan has a hierarchy: wide establishing shot, medium two-shot or single, close-up, and then specific inserts or reaction shots. You do not need all of these for every scene, but you do need to decide which you need before you start, because each constrains the others.

Write out the hierarchy as a generation queue.

```
SCENE 04 GENERATION ORDER
1. Wide: both characters in frame, window behind B, door frame-left
2. OTS on A (camera behind B's left shoulder, A centre-right of frame)
3. OTS on B (camera behind A's right shoulder, B centre-left of frame)
4. CU on A: same lens, same light, tighter
5. CU on B: same lens, same light, tighter
6. Insert: A's hands on the table
```

Generating in this order matters. The wide locks in the geography. Every subsequent shot is written to match what the wide established, not to be independently beautiful.

Write each prompt as a spatial continuation, not a fresh idea

When you get to shot 2 in your queue, your prompt should contain a spatial anchor to shot 1. Something that ties it into the same room, the same light, the same moment.

```
Over-the-shoulder shot from behind a woman's left shoulder,
her dark hair slightly out of focus in the foreground,
a man standing two metres away centre-right of frame,
fluorescent office light from above, window with grey daylight behind him,
the same sparse meeting room as the previous wide: white walls, single table.
Static camera. Mid-morning.
```

Notice that the prompt describes the room as a continuation. It names the light source established in the wide. It keeps the man on the same side of frame. This is not about making the prompt longer. It is about keeping the spatial logic intact.

The reaction shot is where AI coverage most often fails

Close-up reaction shots are where spatial coherence collapses fastest. The eyeline drifts. The character looks slightly off camera when they should be looking directly into the OTS angle. The light shifts. The framing goes from loose to tight with no apparent reason.

For reactions, always specify the eyeline explicitly and link it to the geography note.

```
Close-up on a man's face, looking slightly frame-left at someone
just off camera to his left. Eye level camera. He listens, still.
Fluorescent office light from above. White wall behind him, slightly soft.
Same room, same scene, continuous moment.
```

"Looking slightly frame-left" is a spatial instruction. It is what keeps this shot from marooning itself.

The before and after

Without a coverage plan, you get a set of individually compelling images that fight each other in the edit. The eye does not know where to rest. The geography shifts. The scene feels restless even when the performances are good.

With a coverage plan, the edit has options. You can cut on action. You can hold on a reaction. You can go wide to give the moment weight. The shots are in conversation with each other because you designed them to be.

For your next scene, write the geography note and the generation queue before you open any generation tool. Treat the wide as the law, and treat every subsequent shot as a description of the same room seen from a different position within it.