Storyboard before you generate

Most money wasted on AI video is spent discovering what you wanted. You generate, look at it, realise the shot is wrong, change the prompt, generate again. Four iterations later you have one usable clip and a bill for five.

The fix is not a better prompt. It is doing the thinking in a medium that costs nothing.

Planning is free; pixels are not

Writing a shot list costs nothing. Rewriting it costs nothing. Reordering it, changing the location, swapping which product appears in which shot — all free.

Generating five seconds of 1080p video with audio on a premium model is the expensive operation, and it should be the last thing you do, not the thing you use to explore.

This is not an AI insight. It is how film has always worked: you storyboard because reshooting is expensive. The economics of AI video are different in scale but identical in shape.

What a good plan pins down

Before anything renders, each shot should have an answer to five questions:

1. What is in frame? The subject, and what else shares the shot with it.
2. What happens? The action, however small. “Turns toward camera” is an action; “looks premium” is not.
3. Where is the camera and what is it doing? Static, push-in, tracking, macro slider. Be specific.
4. What is the light doing? Direction, quality, colour. This carries more of the mood than anything else.
5. How long is it? Four to six seconds is a normal shot. Ten is a long hold, and long holds expose model weaknesses.

If a shot cannot answer all five, generating it is a coin flip.

Draft passes are worth it

When you do start generating, the first pass should be cheap on purpose.

A 480p draft on a fast model costs a fraction of the final and answers the only question that matters at this stage: is the composition right? Framing, motion and whether the subject reads correctly all survive the resolution drop.

Only once the shot works structurally should you re-run it at final quality. Two generations at draft and final together often cost less than three at final while you iterate.

Regenerate the shot, not the film

This one is structural, and it is worth insisting on from whatever tool you use.

If your four-shot film is one generation, fixing shot three costs a full re-render. If each shot is independent, fixing shot three costs one shot. Over a project with normal amounts of iteration, that difference is most of the budget.

It also changes how you work. When a fix is cheap you actually make it, instead of shipping the version that was almost right.

Keep every take

Related, and frequently missed: do not overwrite.

Version three is not automatically better than version two. Motion might be improved while the product got worse. Unless you can put them side by side and pick, you are choosing blind — and if the old take is gone, you cannot go back.

What this looks like in practice

A twenty-second, four-shot advertisement done well:

1. Upload references and wait for analysis. Free.
2. Brief the Director; get a four-scene storyboard. Free.
3. Rewrite two of the four scenes. Free.
4. Generate all four as 480p drafts. Cheap.
5. Fix the one shot that is wrong; regenerate that shot only. Cheap.
6. Re-run all four at final quality. The real spend, once.
7. Cut, add audio, export.

Most of the work happens before step four. That is the point.