Start a video prompt with three decisions: what is on screen, what it does, and how the camera moves. Then add the setting, lighting, and details that must remain consistent. If two instructions conflict, resolve that conflict before generating; adding more descriptive words will not tell the model which instruction matters most.
This guide uses a simple structure that works for text-to-video and can be adapted for image-based workflows in Veo3o1.
The six-part prompt structure
Write the prompt in this order:
- Subject — the person, object, animal, or scene that matters most.
- Action — one main movement or event.
- Environment — location, time, weather, and useful background details.
- Camera — shot size, angle, and one camera movement.
- Light and mood — the visual atmosphere.
- Constraints — details that must stay fixed or elements to avoid.
Example:
A handmade ceramic cup sits on a dark wood table as steam curls upward and morning rain moves across the window behind it. Close product shot at table height, slow camera push-in, soft cool daylight with a warm reflection on the glaze, stable cup shape, no added text.
The prompt is specific without trying to describe ten shots at once.
Subject: name what must remain readable
Start with one clear subject. “A busy city” gives the model many possible priorities. “A bicycle courier in a yellow rain jacket” creates a visual anchor.
For image-based generation, connect the prompt to the input:
The person shown in the reference images...
Then repeat only the identity details that matter, such as clothing color, hairstyle, packaging shape, or material. The character consistency guide explains how to pair those details with reference images.
Action: use one main verb
Short clips benefit from one readable action. “Walks,” “turns,” “opens,” “rises,” or “rotates” gives the shot a spine. Several sequential actions may be compressed, skipped, or blended together.
Weak:
A chef cooks, serves dinner, talks to guests, and the restaurant becomes a rooftop party.
Stronger:
A chef places the finished dish on the counter as steam rises into warm kitchen light.
If the story needs multiple beats, generate separate shots and edit them together.
Environment: include details that affect the shot
Describe background details that change lighting, motion, or meaning. “At sunset beside a windy coast” is more useful than a long inventory of objects that never interact with the subject.
Choose compatible signals. “Bright noon,” “dark candlelight,” and “neon night” in the same short prompt create competing lighting directions unless a transformation is the point of the shot.
Camera: specify shot size before movement
Camera language becomes clearer when it follows a hierarchy:
- shot size: wide, medium, close-up, macro;
- angle: eye level, low angle, overhead;
- movement: push-in, pull-back, pan, tilt, orbit, tracking;
- pace: slow, restrained, rapid, handheld.
Example:
Medium shot at eye level, slow tracking movement beside the runner.
One deliberate camera move is usually easier to control than “drone shot, zoom, orbit, whip pan, handheld.”
Timing: match motion to duration
Veo3o1 exposes 4, 6, and 8 seconds where supported. Reference Images mode currently uses 8 seconds. A four-second shot should reach its visual point quickly; an eight-second shot can support a slower reveal or a more gradual camera move.
Do not fill extra time with unrelated events. Extend one meaningful motion instead.
Constraints: protect the important details
Use constraints sparingly. A concise ending such as “same red coat throughout, stable product proportions, no text” can prevent common drift. A long negative list can overwhelm the positive description of the shot.
When using First & Last Frames, describe how the scene travels from the first image to the second. When using Reference Images, describe how the referenced subject behaves in a new shot. The mode comparison explains the difference.
Three reusable prompt templates
Product shot
[Product] on [surface/environment]. [Single product or environmental motion]. [Shot size and angle], [one camera movement]. [Lighting and mood]. Keep [critical shape/color/material] unchanged, no added text.
Character shot
[Character identity] [single action] in [environment]. [Shot size and angle], [camera movement]. [Lighting and mood]. Keep [clothing/face/hair details] consistent.
Establishing shot
A [location] during [time/weather], with [one meaningful environmental motion]. Wide [angle] shot, [camera movement], [lighting and atmosphere], realistic scale and restrained motion.
Iterate like an editor
Save the prompt and selected settings for every useful result. Change one variable per attempt: first the action, then the camera, then the lighting. An optional seed can help you compare repeated settings, but it does not replace clear instructions or guarantee identical output.
Before generating, run this checklist:
- Is there one primary subject?
- Is the main action possible within the selected duration?
- Does the camera have one clear job?
- Do the lighting instructions agree?
- Are references and prompt describing the same identity?
- Is the aspect ratio appropriate for the destination?
- Is the displayed credit cost acceptable for this stage?