Image to video prompts: 24 motion patterns that bring a still to life

An image-to-video prompt should describe motion, not content — the still already locked in the subject, composition, and lighting, so every word you spend re-describing them is wasted, and every word you spend on movement is not. The reliable formula is one camera instruction plus one subject or atmosphere instruction, stated plainly: "slow push-in toward the bottle, steam rising from the cup beside it." Below are 24 patterns organized by what you are trying to achieve, followed by before-and-after rewrites of prompts that fail. You can test any of them on our free image-to-video tool.

Why image-to-video prompts are written differently

With text-to-video, the prompt does two jobs: it invents the scene and it directs the motion. With image-to-video, the first job is already done. The model reads the still for subject, framing, palette, and light; the prompt's only remaining job is to answer one question — what moves, and how? (If you are still deciding between the two starting points, our text-to-video vs image-to-video comparison covers when each wins.)

That changes the writing rules:

• Describe motion only. "A woman in a red coat stands on a foggy pier" tells the model nothing it cannot see. "Her coat ripples in the wind as fog drifts across the pier" is a direction.
• One camera move per clip. Generations run 4 to 10 seconds. A push-in and an orbit and a tilt in one clip produces mush; a single committed move produces a shot.
• Pick a speed. "Slow," "gentle," "gradual" versus "fast," "sudden," "whips." Unspecified speed drifts toward the frantic.
• Respect what the frame can support. A prompt cannot invent what the pixels do not contain. Asking a close-up portrait to "walk out the door" forces the model to hallucinate legs, a door, and a room — and the shot falls apart.

One honest limitation before the patterns: image-to-video on PrismPoster rejects raw photos of identifiable real people. Your selfie or a photo of a friend will be refused at upload — a real person can only appear through consented likeness. Generated stills, product shots, artwork, landscapes, and characters all animate without issue, and everything below assumes one of those as the source.

Camera-move prompts: when the subject should hold still

The camera move is the workhorse of image-to-video. Most product shots, landscapes, and posters look best when nothing in the scene moves except the viewpoint.

1. Slow push-in — slow, steady push-in toward the subject, everything else still. The default. Adds intent and focus to almost any still.
2. Pull-back reveal — camera slowly pulls back, revealing more of the scene around the subject. Works best when the still has visual interest at its edges for the model to extend.
3. Lateral tracking — camera tracks slowly from left to right across the scene, parallax between foreground and background. The parallax cue matters; it separates depth planes instead of sliding a flat postcard.
4. Orbit — camera arcs slowly around the subject, keeping it centered. Strongest on single objects with clear silhouettes — products, statues, buildings. Keep it to a partial arc; a full 360 asks the model to invent the object's hidden side.
5. Crane up — camera rises slowly, tilting down to keep the subject in frame. Turns a street-level still into an establishing shot.
6. Tilt down from sky — camera tilts down from the sky to settle on the subject. A natural opening beat for the first clip of a sequence.
7. Handheld drift — subtle handheld sway, as if held by a standing camera operator. Adds documentary texture without a directional move. Keep the modifier "subtle" or you get seasickness.
8. Focus pull — focus shifts slowly from the foreground object to the background. Needs a still with genuine depth — two distinct planes — to have anything to pull between.

Subject-motion prompts: when something in the frame should move

Subject motion is riskier than camera motion, because the model has to redraw the subject frame by frame. Small, physically simple movements succeed; large repositioning fails. Favor motion that is secondary — the subject stays put while part of it moves.

1. Fabric in wind — her dress and hair move gently in the breeze, body still. The single most reliable way to make a character still feel alive.
2. Steam and smoke — steam rises slowly from the cup, curling in the light. Near-foolproof; steam has no anatomy to get wrong.
3. Water in motion — waves roll gently toward the shore in a continuous loop-like rhythm. Water reads well even when imperfect.
4. Slow product rotation — the product rotates slowly on its axis, lighting fixed. Better on simple shapes; label text can swim on a full rotation, so keep the turn partial. (More product-specific technique in our AI product photography guide.)
5. Breathing and blinking — the character breathes slowly and blinks, otherwise still. For generated characters and portraits — the "living painting" effect.
6. A turn of the head — the character slowly turns their head toward the camera. The upper limit of safe character motion in one clip. Anything beyond it — walking, gesturing, interacting — belongs in a fresh generation, not a prompt.
7. Flame and candlelight — the candle flame flickers, shadows shifting softly on the wall. The shadow clause does half the work.
8. Falling elements — autumn leaves drift down through the frame / snow falls steadily in the foreground. Adds motion in front of the subject without touching the subject at all — the safest pattern on this list.

Atmosphere prompts: when the mood should move

Atmosphere prompts animate the environment and light rather than any object. They pair well with a slow camera move, and they are the patterns most worth combining.

1. Rolling fog — low fog drifts slowly across the ground from left to right. Give fog a direction or it pulses in place.
2. Shifting light — sunlight slowly intensifies, shadows lengthening across the scene. A time-passing cue that transforms landscapes.
3. Rain beginning — light rain starts to fall, droplets streaking past the lens. The lens clause puts rain in the foreground plane where it reads clearly.
4. Dust in a light beam — dust motes drift through the shaft of light. Only works when the still already contains a visible beam.
5. Moving clouds — clouds drift slowly across the sky, their shadows crossing the landscape. The shadow clause again earns its place.
6. Flickering signage — the neon sign flickers irregularly, its glow pulsing on the wet street. A night-scene staple.
7. Heat shimmer — heat haze shimmers above the road in the distance. Subtle, and best kept in the background plane.
8. Curtains and interiors — sheer curtains billow gently at the open window. The interior counterpart of fabric-in-wind, and a quiet way to animate an otherwise static room.

Before and after: fixing prompts that fail

The fastest way to internalize the rules is to watch a weak prompt become a working one.

Re-describing the image. Before: A vintage red car parked on a rainy street at night, neon reflections, cinematic lighting. After: Slow push-in toward the car; rain falls steadily, neon reflections shimmering on the wet street. The before-version is an image prompt pasted into a video field — all nouns, no verbs. The rewrite spends every word on motion.

Asking for too much. Before: The woman walks along the beach, turns around, waves at the camera, then the camera flies up into the sky. After: Her hair and dress move in the sea breeze as she stands looking at the waves; slow push-in. Four actions in one clip guarantees a broken middle. If the walk and the wave matter, they are separate generations, cut together afterward — which is how finished videos of any length actually get made.

No speed, no direction. Before: Make it move. Camera movement, wind, dramatic. After: Gentle handheld drift; the tall grass sways slowly in the wind, clouds drifting right to left. "Dramatic" is a mood, not a direction. Named speeds and directions give the model something to obey.

A workflow note: stills built for animation animate better

The prompts above work on any accepted still, but the sources that animate best are the ones with depth cues (foreground/background separation), soft directional light, and room around the subject for the camera to move into. If you are generating your source stills, it is worth composing for motion from the start — our image-to-video guide covers source-image strategy in depth, and the prompt-writing guide covers the still side of the equation. Inside PrismPoster, the Image Studio and Video Studio share a library, so a still you generate is one click away from being animated at 360p up to 4K, in 16:9 or 9:16.

Frequently Asked Questions

What should an image-to-video prompt include?

One camera instruction, one subject or atmosphere instruction, and a speed. For example: "slow push-in; steam rises from the cup." The image already defines subject, composition, and lighting, so the prompt should describe only what moves and how fast.

Why does my animated image look warped or distorted?

Usually the prompt asked for more motion than the pixels can support — large subject repositioning, a full rotation, or several actions in one clip. Cut back to one small move, add a speed word like "slow" or "subtle," and keep character motion to secondary movement such as hair, fabric, or breathing.

Can I animate a photo of myself or another real person?

Not from a raw photo — PrismPoster's image-to-video rejects photos of identifiable real people at upload. A real person can appear only through Creator Cast, a consent-based likeness the person sets up and controls. Generated characters, products, artwork, and landscapes all animate normally.

How long can the animated clip be?

A single generation runs 4 to 10 seconds, in 16:9 or 9:16, at 360p up to 4K. Longer finished videos are made by generating several clips and cutting them together on a timeline — the same way every longer AI video is made.

Do motion prompts work the same on every image-to-video tool?

The principles transfer — describe motion not content, one move per clip, name a speed — because they follow from how the generation works, not from any one product. Exact phrasing tolerance varies by tool, so treat the patterns here as starting points and adjust to what your results tell you.

Try the patterns on a real still

Reading motion prompts teaches less than running three of them against the same image and comparing the results. The free image-to-video tool lets you do exactly that in the browser, and a free account comes with a one-time grant of 200 credits — no card — which covers several 360p clips at 20 credits each: enough to test a push-in, a fabric-in-wind pass, and an atmosphere combination on your own still today. Credits never expire, so whatever you do not spend on experiments stays for the shot you actually keep.