Text-to-video prompts: camera moves, pacing and motion

An image prompt describes a moment. A video prompt describes a shot: a moment plus what changes over a few seconds. The most common mistake is writing a video prompt like a story - three scenes, a character arc and a twist - and getting back a clip that tries to do all of it and does none of it well.

Think in single shots

A generated clip is one continuous shot. Write one location, one main subject and one action. If you need a sequence, render several clips and cut them together.

The shape of a video prompt

1. 01Shot and camera move - "slow aerial pull-back", "handheld tracking shot", "static wide shot".
2. 02Scene - where we are and what it looks like, as in an image prompt.
3. 03Motion - what moves and how: "waves rolling in", "leaves spiralling down", "steam curling up from the cup".
4. 04Pace and mood - "slow and calm", "energetic", "dreamlike slow motion".

Camera moves the models understand

• Static - the camera holds still; all motion comes from the scene. Safest for a clean result.
• Pan - the camera turns left or right on the spot.
• Tilt - the camera turns up or down, revealing height.
• Dolly / push-in / pull-back - the camera moves toward or away from the subject.
• Tracking - the camera travels alongside a moving subject.
• Orbit - the camera circles the subject.
• Aerial / drone - a high, moving view over a landscape.

Choose one. Combining "orbit while pushing in and tilting up" in a few seconds usually produces a clip that wobbles.

Give the scene something to do

Clips come alive when something in them moves naturally: water, weather, fire, smoke, fabric, traffic, animals, light. A still object in a still room with a static camera will look like a photograph with a slight shimmer.

Length and resolution

Models here make clips up to 30 seconds. Longer is not automatically better: a model has to keep the scene consistent for every extra second, and a tight 5-second shot often looks more convincing. Draft ideas short and at a low resolution, then render the keeper at full quality - the composer shows the price for each combination before you commit.

Start from a still for more control

If the look matters more than surprise, generate the first frame as an image, get it right, then animate it. The model keeps your composition and only has to invent the motion. The image-to-video guide walks through it, and the model comparison helps you pick where to render.