Prompting · 16 June 2026 · 2 min read

Prompting AI video: seven techniques that work

Subject, action, camera, light: how to describe a shot so the model has less to guess, with a weaker and a stronger version of each.

By The Crewnema team

Prompting a video model is closer to directing a very literal crew than to searching the web. It does what you describe and fills every gap with a guess. These seven techniques are about leaving fewer gaps. You will mostly be editing the script and shot descriptions that Crewnema writes for you, so they apply there too.

1. Name the subject precisely

“A woman” could be anyone. “A woman in her thirties in a green apron, tying her hair back” is a person the model can keep hold of. Give age range, clothing and one distinctive detail, and keep the wording the same from shot to shot.

2. Describe the action, not just the look

Instead of: “A barista and a cup of cold brew.”

Try: “A barista slides an iced cold brew across the counter and smiles at the camera.”

Video needs something to do. A look is a still; an action is a shot.

3. Say where the camera is

Framing decides a lot: a close-up of hands, a medium shot at the counter, a wide shot of the whole café. Pick one per shot and say it. “Close-up” and “over the shoulder” are words the models understand; vague terms like “cinematic” are not instructions.

4. Give it light and a mood in a few words

“Soft morning light from a window on the left” does more than “beautiful lighting”. Two or three concrete words about light and colour are enough. Long lists of adjectives mostly cancel each other out.

5. Keep shots short and lines shorter

A shot of four to six seconds is easier to get right than one of fifteen. If a shot has spoken words, they have to fit: Crewnema counts how long a line takes to say and tightens lines that will not fit the shot. Write the way people talk, in short sentences.

6. Hold the cast and the product steady

Use the same description of each person every time they appear, and attach a photo of the real product. Crewnema also measures whether a face holds across shots and repairs the ones that drift. The longer guide to this is keeping the same face and product in every shot.

7. Say what you want, not what you do not

Models often act on a word even when it follows “no”. “No crowd” can produce a crowd. Describe the empty street instead: “an empty street at dawn”. If something keeps appearing, remove it from the description rather than forbidding it.

Make your first video

Crewnema is in private beta and lets businesses in a few at a time. Request access and we will write when yours is open.