Skip to content
Open the builder

How to Write Your First AI Video Prompt, Step by Step

ahmedNovember 12, 20263 min read

A good AI video prompt reads like a short note from a director to a camera crew. It says what is on screen, what happens, and how the camera records it. This tutorial builds one prompt from a rough idea, one layer at a time, so you can see what each layer adds.

Step 1: Start with one sentence about the subject

Write down who or what the shot is about and what they are doing. Keep it literal. A model cannot film a feeling, but it can film a person doing something.

Layer 1: subject and action
An old fisherman repairs a net on a wooden dock.

This already works, but the model has to guess everything else: the time of day, the framing, the camera, the style. Every guess is a place where the result can drift away from what you imagined.

Step 2: Add the setting and one or two details

Place the subject somewhere specific and add details the camera can see. Pick details that support the story of the shot, not a list of every object in the room.

Layer 2: setting and visible details
An old fisherman with a grey beard and a yellow raincoat repairs a torn fishing net on a wooden dock. Small boats rock gently in the harbor behind him. Light rain falls.

Notice the order: subject, action, background, weather. Describe things roughly in the order the viewer should notice them.

Step 3: Choose the shot size and angle

Now decide how close the camera is and where it stands. One shot size and one angle per shot. If you are not sure which to pick, the shot types and camera angles library shows all 26 options with the exact wording.

Layer 3: framing
Medium close-up, eye level. An old fisherman with a grey beard and a yellow raincoat repairs a torn fishing net on a wooden dock. Small boats rock gently in the harbor behind him. Light rain falls.

Step 4: Add camera movement

Video models need to know whether the camera moves. If you say nothing, many models add a slow drift on their own. Name one movement, or say the camera is static. Two movements at once, such as a push in and a pan, often produce neither.

Layer 4: camera motion
Medium close-up, eye level, slow push in. An old fisherman with a grey beard and a yellow raincoat repairs a torn fishing net on a wooden dock. Small boats rock gently in the harbor behind him. Light rain falls.

Step 5: Set the light and the look

Lighting changes the mood more than any other single choice. Name the light source and its quality, or use a look name that carries a full set of decisions. See cinematic looks and lighting for 27 options.

Layer 5: light and look
Medium close-up, eye level, slow push in. An old fisherman with a grey beard and a yellow raincoat repairs a torn fishing net on a wooden dock. Small boats rock gently in the harbor behind him. Light rain falls. Overcast soft daylight, muted blue-grey palette, shallow depth of field, 50mm lens, cinematic film grain.

Step 6: Read it back and remove conflicts

Before you generate, read the prompt once as if you were the camera operator. Ask three questions:

  • Can every word be filmed? Remove words like “beautiful”, “amazing” or “masterpiece”. They add nothing the model can act on.
  • Does anything contradict? “Bright sunny day” and “light rain” in the same prompt force the model to ignore one of them.
  • Is there more than one shot here? If the action has a beginning, a middle and a cut, split it into separate prompts.

The same prompt as JSON

Some models and pipelines accept structured prompts. JSON keeps each layer in its own field, which makes it easy to reuse the style and camera settings across several shots.

{
  "shot": "medium close-up",
  "angle": "eye level",
  "camera_motion": "slow push in",
  "scene": "An old fisherman with a grey beard and a yellow raincoat repairs a torn fishing net on a wooden dock. Small boats rock gently in the harbor behind him. Light rain falls.",
  "lighting": "overcast soft daylight",
  "style": "muted blue-grey palette, cinematic film grain",
  "lens": "50mm, shallow depth of field"
}

You do not need to type this by hand. The Prompt Builder assembles both versions from the same choices.

Summary

Subject and action first, then setting, framing, motion, light and lens. Keep every word visible on screen, remove contradictions, and change one layer at a time. That order works for most text-to-video models and is a solid starting point for image models too.

Share

Leave a Reply

Your email address will not be published. Required fields are marked *