⌘ K

Image to video prompts

Image-to-video prompt guide and examples

Use these AI video prompt examples when you already have a keyframe and need controlled camera movement, subject motion, environmental motion, and timing without rebuilding the scene. Keep the still-image description separate from the motion instruction so you can identify whether a failure comes from composition or animation.

Example prompt

Keyframe: premium running shoe on wet black asphalt at night, low three-quarter camera angle, cyan and amber reflections, product silhouette and label sharp, realistic contact shadow, 16:9, no text. Motion: five-second slow push-in, subtle rain ripples and reflection shimmer, one soft light sweep from left to right, shoe remains completely still and unchanged, no orbit, no zoom jump, no product morphing, end frame holds for half a second.

Copy into ChatGPT Image / GPT Image; adjust product names and hex colors.

Steps

  1. Step 1

    Generate the still first

    Lock the subject, framing, material, light, and aspect ratio in a clean keyframe. Do not hide an unresolved composition problem inside a long motion prompt.

  2. Step 2

    Separate motion notes

    Write camera motion, subject motion, environmental motion, and duration as separate instructions. This makes it easier to remove the exact movement that causes distortion.

  3. Step 3

    Choose one primary movement

    Start with a push-in, pull-back, pan, tilt, or locked camera—not several at once. Add small secondary motion such as rain, steam, hair, or fabric only after the main movement holds together.

  4. Step 4

    Protect fixed details

    State what must not change: face identity, product shape, label, clothing, background layout, or camera axis. Repeat only the critical constraints, not the full image prompt.

  5. Step 5

    Test and reduce motion

    When shapes warp or the frame drifts, shorten the clip, reduce camera travel, remove one moving element, and test again from the same keyframe.

Multi-step pipeline: Workflow index · Prompt library

FAQ

What motion is safest for product videos?
Slow push-ins, light shimmer, steam, and fabric movement usually hold shape better than orbit shots.
What belongs in an AI video prompt?
Include the starting frame, one main camera action, limited subject or environmental movement, duration, ending behavior, and details that must remain fixed.
Why does an image-to-video result change the subject?
Large camera moves and too many moving elements give the model more opportunities to reinterpret the frame. Reduce motion and explicitly lock identity or product geometry.
Should I describe the whole image again in the motion prompt?
Usually no. Use the keyframe for appearance and keep the motion prompt focused on movement, timing, camera behavior, and preservation constraints.