Image-to-Video Prompt Guide

The core principle

First lock the character, scene, lighting, and color in a still image. Then add only visible motion: a body action, a camera move, and changes over time. This keeps the video model from trying to solve composition, style, and movement at the same time.

Three hard rules

Three hard rules for image-to-video:
1. The first frame already defines the style; describe motion and change only
2. Keep one dominant movement per prompt
3. Turn multiple actions into a timeline instead of one long sentence

For example, this timeline is more stable than a long sentence with conflicting actions:

0-2s: The girl remains in the snow, shoulders moving slightly, eyes looking into the distance
2-4s: She slowly turns, her gaze settles naturally on the camera, wind lifts her hair and coat
4-5s: The camera moves in slightly and stops at a medium close-up

Use a timeline for multiple actions

When a shot contains a subject action, a camera change, and a secondary effect such as wind or a gaze shift, write explicit time ranges. The timeline tells the model the order of states and prevents every action from happening at once.

• Small head turns, hand movements, or breathing changes usually fit in 3-5 seconds.
• Running, turning, stopping, and opening a door are often better as two shots.
• Describe visible changes instead of abstract emotion labels.

The anchor-action method

Choose one main action for the torso or center of mass. Let hair, clothing, hands, eyes, and facial expression follow as satellite changes. Without this hierarchy, each body part may move independently and the result feels disconnected.

Anchor: torso or center-of-mass action
Satellite: secondary changes in hands, head, clothing, and expression

Example:
Anchor: the subject steps forward and turns
Satellite: hair moves in the wind, fingers tighten slightly, gaze follows the turn

Single-shot template

## First-frame responsibility
- Lock the character, scene, lighting, and color

## Video prompt
- Dominant action:
- Camera movement:
- Timeline:

## Constraints
- Do not change the first-frame style
- Do not add extra people
- Do not rebuild the background structure

Common mistakes

• Repeating “8K”, “cinematic”, “hyper-realistic”, and “dreamy” in the video prompt.
• Combining handheld motion, a smooth push-in, and a 360-degree orbit.
• Asking one shot to run, jump, fall, look up, speak, and turn.
• Changing a natural-light first frame into a high-contrast cyberpunk scene.

Recommended workflow

1. Stabilize the first frame
2. Write one primary action before adding emotion
3. Convert two or more actions into a timeline
4. Use an anchor action for full-body movement
5. Generate only 3-5 seconds per shot
6. If unstable, repair the first frame instead of adding more prompt text

For the full prompt design system, continue with prompt-director. For spatial continuity, see the scene consistency guide.