The shot list habit that stops AI production from collapsing mid-project

You open your project folder three days into production and find 140 clips with names like `output_final_v3_USE_THIS.mp4` sitting next to `output_final_v3_USE_THIS_2.mp4`. You remember roughly what each clip was supposed to be, but the coverage logic that seemed obvious on day one has dissolved. You are now spending more time hunting for usable takes than generating new ones, and the cuts that do exist feel arbitrary because you never planned the geometry of the scene.

This is the most common reason solo AI film projects stall or finish looking structurally weak. It is not a prompting problem and it is not a model limitation. It is a production planning problem, and the fix is borrowing a discipline that live-action crews have used for decades: the paper shot list, adapted for the specific constraints of working with generative video.

What a shot list does for an AI project that a mood board cannot

A mood board tells you what something should feel like. A shot list tells you what footage you need in order to cut a scene. Those are different jobs. When you are generating clips at a cost per second, the shot list is also a budget document. Every line on it is a decision about whether to spend a generation credit on a wide, a medium, a close, a cutaway, or a reaction. Writing the list before you open a generation tool forces you to answer the question: what is the minimum coverage this scene actually needs?

For a simple two-person dialogue scene, that minimum is usually:

- A wide establishing the space and both figures
- An over-the-shoulder for character A
- An over-the-shoulder for character B
- One close-up insert (hands, object, a face detail)

Four shot types. In a generative workflow, each of those is a separate prompt, possibly several takes. If you do not list them in advance, you will generate twelve clips of whatever felt interesting in the moment and discover you have no coverage of character B at all.

The actual format to use

Keep the shot list as a plain text or spreadsheet file, one row per shot. Columns that matter:

```
Scene | Shot # | Framing | Lens feel | Motion | Subject action | Status
----------------------------------------------------------------------
02 | 02-01 | Wide | 35mm equiv| Locked | A and B at table, A speaking | APPROVED
02 | 02-02 | OTS-A | 50mm equiv| Slow push in | B listening, slight nod | NEEDS REGEN
02 | 02-03 | OTS-B | 50mm equiv| Locked | A speaking, hand gesture | APPROVED
02 | 02-04 | CU insert | 85mm equiv| Locked | A's hand on document | TO DO
```

The Status column is the column that saves your project. APPROVED means a clip is locked and moved into your edit folder. NEEDS REGEN means the generation exists but has a flaw you have decided is worth fixing. TO DO means you have not generated it yet. At a glance you can see the state of an entire scene without opening a single file.

Naming and folder structure that follows the list

Your file names should map directly to shot numbers. A clip for scene 02, shot 02-03, take 2 becomes `02-03_T2.mp4`. Nothing else. No descriptors, no `final`, no `good`. The shot list is where descriptions live. The file name is only an address.

Folder structure should be flat per scene, not nested by date or by model:

```
/project-root
/scene-01
/scene-02
/scene-03
/approved
/regenerate-queue
```

When a clip earns APPROVED status on the shot list, you copy it to `/approved`. When something goes to NEEDS REGEN, you copy it to `/regenerate-queue` so you can batch those generations in one sitting rather than context-switching constantly.

Reviewing dailies as a solo creator

Live-action crews review the previous day's footage before shooting the next day. You can do the same. At the end of each generation session, open everything you produced, update the Status column for each shot, and move files accordingly. This review takes ten minutes and it prevents the folder-entropy problem entirely. It also shows you immediately if your coverage plan for an upcoming scene is missing a shot type, before you have spent credits generating into a gap.

The review session is also when you catch consistency problems early. If your wide for scene 03 has midday light and your close-up for the same scene has sunset light, you know now, not in the edit. You either regenerate one or you decide to treat it as a stylistic ellipsis and make that choice deliberately.

The before and after

Before this habit: a folder of clips sorted by gut feel, a patchy edit with missing coverage, and a second production pass that costs as much as the first. After: a shot list that tells you the project is 60% approved with eight specific shots still to generate, a folder structure you can navigate without memory, and an edit that assembles quickly because the coverage logic was decided before the first generation ran.

Start your next scene by writing the shot list before you write a single prompt. Four to six rows is enough for most scenes. Update Status after every session. The clips will follow.