How to Write Image-to-Video Prompts: Motion, Camera and Pacing

A good image-to-video prompt describes what changes, not what is already in the picture. Your still has already settled the face, the outfit, the room and the light, so spend your words on how the subject moves, what moves around them, what the camera does and how fast it all happens. Then add one line saying what must stay the same.

Start with the right still

Most bad clips are decided before you type a word. On SpicyGen, animating an image starts from a picture you generated, a saved character or one of the preset looks. It does not take uploads of real photos, so the starting frame is always something you made here. Run it past the starting-image checklist in the SpicyGen user guide first. Two questions matter most: is every object the action touches already in frame, and are the face, hands, limbs and torso visible rather than cropped or covered?

That second point sounds fussy until you ask a cropped portrait to "run her hand through her hair". The hand has to come from somewhere, so the model invents one, and invented hands are where warping starts. Match the action to the framing. A tight close-up suits eye movement or a slow smile. Head and shoulders leave room for a head turn and visible breathing. If she is going to stand up, walk or sit on the bed, you need a full-body shot with the setting in place.

What to describe

The subject's main action

Pick one action and write it as a body part, a direction and an end point. "She moves sensually" gives the model nothing to animate. "She turns her head from the window toward the camera and holds eye contact" gives it a start, a path and a place to stop. Verbs do the work here. Adjectives like sultry describe a mood the image already has; to show the mood, put it in an expression: "a slow smile forms" or "her eyes narrow slightly".

Don't re-describe the character; "the woman" or "he" is anchor enough. Repeating the hair color, outfit and body type invites the model to redraw them, and it rarely redraws them identically.

Secondary motion

This is what separates a living clip from a puppet show: hair settling after a turn, a silk robe shifting at the hem, a candle flame, a curtain moving in a draft, the chest rising with a slow breath. Pick one or two details that would plausibly move in that room. Any more and the frame starts to shimmer as everything competes for motion.

The camera

Give the camera one job, or tell it to stay still. A slow push-in, a gentle pan, a slight handheld drift and a locked-off shot all sit well under a moving subject. An orbit while the subject also turns is asking for trouble. Camera angle and movement aren't separate settings on SpicyGen, so write them plainly in the prompt, in their own sentence. For the full vocabulary of moves and shot sizes, see our guide to AI video camera movements and shot types.

Pacing

There is no speed slider, so pacing comes from your words. "Slowly", "gradually", "unhurried" and "pauses for a moment" all give the motion room to read. Sequence words like "then", "after a beat" and "finally" tell the model the order when there is more than one beat.

Match the number of beats to the length. Clips on SpicyGen can run up to 30 seconds. A few seconds holds one action comfortably. A longer clip can carry two or three beats, but each one needs its own sentence, and the last one should be a pose the character can hold.

What stays the same

Close with a line of constraints: "Her face, hair and lingerie stay the same. The room and lighting don't change." It looks redundant, but it gives the model permission to leave things alone. Artifacts you never want, like a morphing face or extra fingers, belong in the negative prompt field. If the same person appears across several clips, use a saved character, which keeps the same face from one generation to the next.

A structure you can reuse

[Who, in two or three words] [main action: body part, direction, end point], [pace].
[One or two secondary motions].
[Camera: one move, or static], [pace].
[What stays the same].

Action goes first because it is the thing you most need to land. The camera gets its own sentence so it doesn't get tangled up with the subject's movement. Four or five plain sentences is plenty. Longer prompts don't buy more control; details just get dropped or blended.

Before and after examples

A boudoir portrait

The still: a woman in her thirties sitting on the edge of a bed in black lace lingerie, looking toward a window, warm lamplight.

Before:

Beautiful woman in her thirties, long dark hair, black lace lingerie, luxurious bedroom, warm lighting, sexy, cinematic, masterpiece, moving

This is an image prompt with "moving" tacked on the end. Everything in it is already in the picture, and "moving" doesn't say what moves. Expect a near-still clip with a faint zoom, or a face that slowly morphs as the model reinterprets "beautiful".

After:

The woman slowly turns her head from the window toward the camera and holds eye contact as a faint smile forms. Her hair slides off one shoulder as she turns, and the sheer curtain behind her sways slightly. The camera stays still at eye level. Unhurried pace. Her face, hair and lingerie stay the same.

A man in a doorway

The still: a man in his forties leaning against a doorframe in an open white shirt, low amber light, full body in frame.

Before:

Handsome man in his forties looks at camera, smiles, winks, walks toward camera, runs hand through hair, leans on wall, camera orbits and zooms in, dramatic

Six actions and two camera moves for one clip. The model will rush some, skip others and merge the rest, and an orbit around a walking figure is a reliable way to bend his limbs.

After:

The man pushes off the doorframe and takes two slow steps toward the camera, then stops and tilts his head with a half smile. His open shirt moves with each step. The camera slowly pushes in to a medium close-up as he approaches. The doorway and amber lighting stay unchanged.

One action with a clear stop and one camera move that follows it. The wink and the hand through the hair can have their own clips.

A longer clip with beats

For a longer clip of an anime-style character, a woman in her late twenties at a vanity mirror, spell out the order:

The woman sits at the vanity and looks at her reflection. After a moment she lifts a brush and slowly draws it through her hair. Then she sets the brush down, turns on the stool to face the camera and rests her chin on her hand. Lamplight flickers softly on the mirror. The camera holds a static medium shot. Her face and outfit stay the same throughout.

Three beats, each in its own sentence, ending on a pose she can hold.

Common failures and how to fix them

Warping faces and bodies

Usually this is too much motion for the frame: a fast camera move on top of a fast body movement, or an action that needs a hand or leg the image doesn't show. Slow everything down and let the subject or the camera make the big movement, never both. Pick a still where the relevant limbs are visible and add the "stays the same" line.

A frozen subject

You get a gentle zoom on what is basically a still image, or a blink and nothing else. Almost always, the prompt described a mood instead of an action. Swap the adjectives for a verb attached to a body part, say where the motion ends, and add one secondary motion. If it still stalls, make the action start straight away: "she immediately begins to turn toward the camera".

Too many actions

The clip looks rushed, gestures blend into odd in-between movements, or the second half of your prompt never happens. Cut to one action for a short clip. A longer clip can take two or three beats with sequence words, each in its own sentence. Anything beyond that is a second clip.

Actions that fight the image

Asking a standing figure to lie back on a bed that is out of frame, or bringing in a second person who isn't there, forces the model to invent the missing pieces, and it does that badly. Choose a still that is a logical starting point for the action, or generate a new one first. Keep to one person where you can; scenes with several people moving at once fall apart quickly.

Change one line at a time

When a clip comes back wrong, resist rewriting the whole prompt. Change one sentence (the verb, the camera line or the pace word) and run it again from the same still. Change everything at once and you'll never know which line fixed it. Keep the versions that work in a note sorted by framing, and you'll soon have starting prompts you trust for every kind of shot.