AI filmmaking
What to Make First With AI Video (Not a Short Film)
Most people open an AI video tool for the first time and try to make a trailer. Ninety minutes later they have fourteen clips that do not belong together and a strong feeling that the tool is overrated.
Make ten seconds that work
Your first finished thing should be three shots, about ten seconds total, of somebody doing something ordinary. A person walks to a window and looks out. Someone sets a cup down and picks up a letter.
That is not a warm-up exercise you skip when you get good. It is the smallest unit that contains the entire craft: you have to establish a space, move the camera within it, keep a person looking like themselves across three shots, and cut them together so the sequence reads as one moment.
Why the trailer fails
A trailer is twenty disconnected shots, each in a different location, with different lighting, often with different characters. Every one of those differences is a place for the tool to drift — and because nothing connects to anything, you never find out whether you can make two shots agree.
It also hides failure. Twenty unrelated clips cut fast with music can look fine while teaching you nothing, which is why so many impressive-looking AI reels come from people who cannot yet make a scene.
The order that works
| Stage | What to make | What it teaches |
|---|---|---|
| 1 | One shot you are happy with | Prompt structure, and how long a generation takes |
| 2 | The same shot five ways | Which words actually change the output |
| 3 | Three shots that cut together | Shot size, camera position, continuity |
| 4 | Thirty seconds with a beginning and end | Structure, pacing, when to cut |
| 5 | A short film | Everything above, sustained |
Most people attempt stage five on day one. The stages are not difficulty levels — they are each a different skill, and stage three is where the actual filmmaking lives.
Stage two is the one people skip
Generating the same shot five times with deliberate single-word changes is the fastest way to learn a tool. Change only the shot size. Then only the light. Then only the lens. Watch what moves.
An hour of this is worth more than a week of trying things at random, because it replaces superstition with knowledge. Most prompt advice circulating online is superstition — someone changed six things at once, liked the result, and credited the wrong one.
Keep every generation, including the bad ones, with the prompt that made it. The failures are the dataset you learn from, and you will want to compare attempt one against attempt six.
Pick a subject that cannot embarrass you
Not your business's hero video. Not a client job. Not anything with a deadline. The first few things you make should have no stakes at all, because you will throw most of them away and you need to be able to do that without flinching.
An empty room in morning light. Someone's hands making tea. A street in rain. Dull subjects are ideal: there is nothing to hide behind, so you can see plainly whether the craft is working.
What finishing teaches that generating does not
The gap between people who make things and people who collect clips is editing. Cutting ten seconds together forces decisions that generation never does — how long to hold a shot, where exactly to cut, what to lose.
You can do that in free software. The tool matters far less than the habit of actually finishing, which is the only thing that converts practice into ability.