Guide

Image to Video Prompt Examples

Image-to-video prompts improve when the weak version and the revised version are visible side by side. These examples show how to turn vague motion requests into one-frame prompt handoffs with clearer action, camera, continuity, and constraints.

Before-and-after prompt revisionsOne-frame motion structure made visibleKling, Veo, and Seedance branch examples from the same frame

Before and after

Good image-to-video prompts replace vague taste words with visible motion structure.

The better prompt does not have to be longer for its own sake. It has to make the starting frame, subject motion, camera behavior, continuity anchors, and failure constraints easier to inspect.

Example 1: Product still

Source frame: A compact silver skincare can sits upright on a dark subway seat beside a rain-speckled window, lit by passing city reflections.

Weak prompt: Make this product cinematic, premium, glossy, with cool city vibes and nice camera movement.

Revised prompt:

Start from a reference frame showing a compact silver skincare can sitting upright on a dark subway seat beside a rain-speckled window. The can tilts slightly toward the camera, catches a moving streak of reflected light, and settles back into center frame. Use a slow push in with controlled handheld drift. Keep the can label readable, preserve the exact silver finish, and keep the subway seat and rain-speckled window stable. Avoid warped packaging, changing label text, extra objects entering frame, or abrupt zoom jolts.

Why it works: The revised prompt names the source frame, one product motion, one camera move, and the details that must not drift.

Example 2: Character reference

Source frame: A young woman in a blue bomber jacket stands on a rooftop at blue hour, hair tied back, city lights softly glowing behind her.

Weak prompt: Animate her naturally with emotion and make the shot feel like a beautiful movie scene.

Revised prompt:

Start from a reference frame showing a young woman in a blue bomber jacket on a rooftop at blue hour. She turns toward camera, the jacket catches a light gust, and her expression resolves into a subtle half-smile. Arc the camera gently from three-quarter profile into a medium close-up. Keep her face shape, hairstyle, jacket color, and rooftop background stable from start to finish. Avoid identity drift, changing clothing details, extra fingers, or exaggerated facial morphing.

Why it works: The revised prompt gives the model a single expression beat while preserving the identity and wardrobe anchors from the still frame.

Example 3: First-frame continuation

Source frame: A dancer frozen mid-spin in a neon-lit warehouse, one arm extended, reflective puddles glowing on the floor beneath scattered haze.

Weak prompt: Continue the dance with lots of energy, dramatic neon lighting, and cool camera tracking.

Revised prompt:

Start from a reference frame showing a dancer frozen mid-spin in a neon-lit warehouse with reflective puddles and scattered haze. The dancer completes the spin, lands into a low step, and resolves cleanly toward frame left. Track just enough to keep the dancer centered while haze moves in slow ribbons and puddle reflections ripple underneath. Keep the outfit, warehouse layout, and reflective floor stable. Avoid duplicate limbs, chaotic background changes, broken anatomy, or over-smearing the motion.

Why it works: The revised prompt turns vague energy into a readable start, movement, and resolved end beat.

Revision passes

Use the same four passes before you branch into model-specific wording.

The model branch should refine the handoff, not rescue a weak base prompt. Start by making the shared one-frame prompt structurally clear.

Pass 1: Anchor the source frame

Start with the visible subject, setting, lighting, and two or three details that must survive once the image begins moving.

Pass 2: Make one motion beat do the work

Name one subject action that can happen inside a short clip. If the prompt needs several verbs, choose the one that matters most.

Pass 3: Add camera after the action is clear

Use camera movement to support the subject motion. If the camera move is more complicated than the action, the shot becomes harder to revise.

Pass 4: Turn drift risks into constraints

Preserve the few details that would break the shot if they changed, then add negative constraints for common failures like warped products or identity drift.

Same frame, different branches

The shared workflow stays primary, then model pages change the emphasis.

These examples use the same product still so the branch difference stays visible: Kling gets more direct shot wording, Veo gets more continuous sequencing, and Seedance gets more readable motion rhythm.

Shared Tool C version

Start from the silver skincare can on the subway seat, tilt the can slightly toward camera, catch one moving reflection, and settle back into center frame. Use a slow push in, keep the label readable, preserve the silver finish, and avoid warped packaging or changing text.

Kling branch

Start from the silver skincare can on the subway seat. The can should tilt toward camera, catch one moving reflection, and settle back into center frame. Use a slow push in. Keep the label readable and the silver finish stable. Avoid warped packaging, changing label text, or aggressive zoom jolts.

Veo branch

Start from the silver skincare can on the subway seat, then let the can tilt toward camera, catch a passing streak of reflected light, and settle back into center frame as the camera pushes in slowly. Keep the label readable, preserve the silver finish, and keep the opening composition recognizable.

Seedance branch

Start from the silver skincare can on the subway seat. Let the can tilt toward camera, catch a streak of reflected light, and settle back into center frame with a clean visual rhythm. Use a slow push in that follows the tilt, preserve the label and silver finish, and avoid jittery camera motion.

Common mistakes

Most failed examples are under-specified before they are model-specific.

If the first output drifts, inspect the source frame and motion beat before adding more style language or switching branches.

The before prompt only describes mood

Mood words can help after the motion is clear, but they do not tell the model what should move or what should stay stable.

The prompt compresses a storyboard into one clip

One-frame image-to-video prompting is strongest when the output has one readable motion job rather than several scene beats.

The constraints arrive too late

If product text, face shape, wardrobe, or environment layout matter, preserve them explicitly before the first generation attempt.

Related paths

Use the examples page as the proof layer for the shared Tool C workflow.

Once the structure is clear, the next useful path depends on whether you want to generate a prompt pack or adapt the same frame for a specific model branch.

Model branches

Compare Kling, Veo, and Seedance once the shared prompt structure is already stable.

Earlier scene layer

Use Script to Shot Prompts first if the job is still a multi-shot planning problem rather than one-frame motion.