Motion prompting
Write image-to-video prompts that move with intention
The image already defines the scene. Your motion prompt should direct what changes, how the camera witnesses it and what must remain stable.
August 21, 20266 min read
Image-to-video works best when the prompt behaves like a compact shot direction, not a second description of the still. Start from motion, make time visible and protect the visual facts that matter.
Separate subject motion from camera motion
Describe the subject first: fabric catches a light breeze, the character turns toward the doorway, ripples spread from one falling leaf. Then give the camera one readable action such as a slow push in or a gentle orbit.
When both are packed into one vague phrase, the model must guess which movement belongs where. Two short clauses usually communicate more than a cinematic adjective.
Give the shot a beginning and an end
A useful prompt shows progression: stillness, a trigger, then a settled result. Try “the lantern is still, a gust lifts it, then it swings gently as the camera moves closer.”
Avoid demanding several unrelated events in a short clip. One primary action and one supporting detail create a stronger sense of intention.
Protect identity and composition
State the few things that should remain stable, especially face, outfit, product geometry or framing. Phrase constraints plainly and keep them shorter than the motion direction.
Generate a restrained first pass. If the identity holds, add energy in the next iteration by increasing camera speed, environmental response or the distance the subject travels.
Put the shot in motion
Bring a still frame into Image to Video and continue refining inside the full creator.