Image to Video Prompts: What to Write When the Image Sets the Look

Image to video is the most controllable way to make AI video right now. You design the perfect frame first, as an image, and then ask the model to bring it to life. But most people write their video prompt as if the image didn't exist. They describe the woman, her red coat, the neon street, the rain, the film look, all over again.

That's the mistake. In image to video, the picture already sets the look. Your prompt should describe what the picture can't: what moves, how, and when.

Why re-describing the image hurts

When you describe the image again, the model tries to match your words and the picture at the same time. Any small difference (you wrote “red coat”, the image has a burgundy one) becomes a tug of war. The result is drift: faces slowly changing, colours shifting, details morphing as the clip plays.

Describe the subject in simple, general words instead: the woman, the car, the dog. The image already knows the details.

The five things to describe

1. The camera move

One move, with speed. This is the most important line.

Slow push-in toward her face.

2. The subject's action

What the subject does, in one clear verb. Keep it small and physical. Big actions (running, fighting, dancing) are harder to hold together.

She turns toward the camera and smiles.

3. The environment's motion

What moves around the subject. This makes the world feel alive even when the subject barely moves.

Rain keeps falling, neon signs flicker, steam drifts from a vent.

4. Light over time

If the light should change, say how. Otherwise leave it out, and the lighting stays as it is in the image.

A car's headlights sweep across her face.

5. Sound, if your platform supports it

Dialogue, sound effects, ambience, music. Our audio guide covers the layers.

Perfume bottle on a marble plinth with floating water droplets, a studio product render still — The image sets the look, the prompt adds the motion. Studio Product Render, an image look in Frame Juice Image, and Product Orbit 360, a camera move in Frame Juice Motion.

Robotic camera arm orbiting a perfume bottle on a plinth in a dark studio — The image sets the look, the prompt adds the motion. Studio Product Render, an image look in Frame Juice Image, and Product Orbit 360, a camera move in Frame Juice Motion.

A worked example

The start frame: a woman in a red wool coat on a rainy neon street at night.

Wrong:

A beautiful woman in a red wool coat stands on a rainy neon street at night, cinematic, film grain, moody, 4k, she turns and smiles.

Right:

Slow push-in toward her face. The woman looks up and turns toward the camera with a small smile. Rain keeps falling, reflections ripple in the puddles, a neon sign flickers behind her.

The second prompt is shorter and has almost no appearance words. Everything in it is motion.

Your image prompt, turned into a motion prompt

Frame Juice Motion is built around this. Paste your image prompt into the start frame, press READ, and it takes only the subject and ratio from it, so your video prompt stays about motion. The free version is here.

See Frame Juice Motion

Keep the action small

Image to video works best when the start frame and the end of the clip are close cousins. A person turning their head, a car pulling away, a flower opening, steam rising. If the action would change the scene completely (the person walks out of a building and into a forest), the model has to invent too much, and quality drops.

For bigger changes, use an end frame as well. Our start and end frame guide shows how.

Choosing the start frame

The prompt can only move what the image makes possible.

Leave room for the move. If you want a pull-back, the frame shouldn't be tightly cropped with important things at the edges.
Pose for the action. A person mid-turn or looking off-camera gives the model a natural next step.
Keep hands simple. Hands in complex poses often deform as they move.
Match the ratio. Generate the image at the same aspect ratio as the video. Cropping later throws away the composition.

Camera on a dolly track pushing in toward a woman seated at a kitchen table, with open space around her — Slow Push-In Drama, a camera move in Frame Juice Motion. The subject sits with space around her, so the push-in has somewhere to go.

Settings that belong in the tool, not the prompt

Clip length, aspect ratio, audio on or off, and which image is the start frame are usually settings in the platform's interface. Writing “8 seconds, 16:9” in the prompt rarely does anything. Set them in the tool, and keep the prompt for motion.

Quick template

[Camera move + speed]. The [simple subject] [one small action]. [Environment motion]. [Optional light change]. [Optional audio].