Image-to-video prompts: 30 examples, templates and fixes
Quick answer
For image-to-video, let the image define appearance and use the prompt to direct motion. A reliable starting structure is: subject action + environmental motion + camera behavior + timing or continuity constraint. Begin with the one movement that matters most, generate a short proof, and change one instruction at a time.
Starting image | Prompt should describe | Avoid repeating |
|---|---|---|
Portrait | Expression, gesture, hair or clothing motion, camera | Age, clothes, facial details already visible |
Product photo | Product interaction, reflections, particles, camera path | Brand, packaging and materials already established |
Landscape | Weather, water, foliage, atmosphere and camera | Static geography and color palette |
Illustration | Animation style, layer motion and timing | Characters and composition already present |
The image-to-video prompt formula
The [subject] [visible action]. [Environmental motion]. The camera [one movement or locked behavior]. [Speed, timing or continuity].
This differs from text-to-video, where the prompt must establish both what the shot looks like and what moves. In image-to-video, the reference frame already supplies the subject, composition, lighting and style. Runway’s current image-to-video prompting guide likewise recommends focusing on motion and starting simply. Treat that as model-specific guidance, not a universal guarantee: different models interpret wording differently.
Portrait prompts
The subject gives a small natural smile and blinks once. A light breeze moves a few strands of hair. Locked camera, subtle motion only.
The subject turns slowly toward the window, then holds their gaze. Soft curtain movement in the background. Gentle handheld camera.
The subject takes one calm breath and lifts their chin slightly. Fabric moves naturally at the shoulders. Static close-up.
The subject looks down, then meets the camera with a restrained smile. Slow push-in, continuous shot.
The subject walks two measured steps toward camera. Clothing responds naturally. The camera tracks backward at the same pace.
Product prompts
Condensation rolls slowly down the bottle while cool mist drifts behind it. The camera makes a slow 20-degree arc from left to right.
The watch remains centered as the second hand moves and a narrow highlight travels across the glass. Locked macro camera.
Coffee pours into the cup in one continuous stream. Steam curls upward. Slow push-in at table height.
The sneaker rotates a quarter turn on the pedestal. Soft dust lifts at the base. The camera remains fixed.
The package opens cleanly and the product rises slightly into view. One continuous shot, controlled studio motion.
Landscape and environment prompts
Cloud shadows move slowly across the valley. Grass bends in a steady breeze. Locked wide camera.
Small waves reach the shore and withdraw. Sunlight flickers across the water. Slow lateral camera slide.
Snow falls at different depths while smoke rises from the cabin chimney. Static camera, quiet motion.
City lights switch on gradually as the sky darkens. Slow time-lapse feeling without a cut.
Fog moves between the trees while branches sway lightly. The camera advances slowly along the path.
Camera-motion prompts
The camera performs a slow push-in and stops before reaching the subject. Everything else remains calm.
Workflow recipe
The subject remains centered while the camera moves backward and zooms in, creating a controlled dolly zoom without changing the scene.
- Model
- Wan 2.2
- Effect
- Dolly Zoom
- Duration
- 5 seconds
- Resolution
- 480p
The camera tracks left while keeping the subject centered. Natural parallax separates foreground and background.
A gentle handheld camera follows the subject from behind. No change of scene.
The camera rises vertically to reveal the landscape beyond the foreground wall. Smooth crane movement.
Locked-off camera. Minimal subject motion only. The first and last frame should feel visually continuous.
Illustration and loop prompts
The illustrated clouds drift slowly while the character’s scarf moves in the wind. Preserve the drawn linework.
The neon sign pulses gently and rain travels down the window. Locked camera, seamless ambient loop.
The paper layers move with subtle stop-motion timing. Preserve the handmade cut-paper texture.
The character breathes and blinks while background particles repeat in a smooth loop. No camera movement.
The water ripples outward and returns to the starting appearance. The camera begins and ends in the same position.
Five prompts for common failures
Too much camera motion: Locked camera. The subject performs one small movement; the framing remains unchanged.
Frozen output: The subject takes one clear step forward as fabric and hair respond naturally. A slow push-in follows.
Unwanted cut: One continuous shot. The action unfolds in the same location without a transition.
Identity drift: The subject remains recognizable and performs a small natural gesture. Avoid requesting a simultaneous transformation.
Chaotic product motion: The product remains rigid and centered while only the light and camera move.
Animate a still image
Upload one strong source image, choose an image-to-video model, start with the essential motion, and revise one variable after reviewing the first clip.
Try Image-to-VideoHow to improve a prompt without wasting generations
Inspect the input first. Blur, malformed hands, unreadable packaging or contradictory motion cues can become more visible after animation.
Start with the subject’s essential action. Add environmental or camera movement only after that action works.
Request one shot. Several locations, transformations or camera moves compete for a short clip’s limited time.
Use positive, visible language. Describe what the camera and subject should do rather than writing a long list of prohibitions.
Change one variable per attempt so you can tell which instruction helped.
Set duration, aspect ratio and model in the interface when those controls are available; words do not override unavailable settings.
Choose image-to-video or text-to-video
Choose image-to-video when a person, product, composition or art direction must begin from an approved frame. Choose text-to-video when you want the model to invent the entire shot. For a deeper workflow comparison, see text-to-video versus image-to-video. For broader camera and lighting language, use the cinematic prompt cookbook.
Frequently asked questions
Describe the subject’s action, relevant environmental motion, one camera behavior and any timing or continuity requirement. Do not spend most of the prompt redescribing visual details already established by the image.
Use the shortest prompt that clearly describes the intended shot. A single sentence can be enough for simple motion. Add detail only to correct a specific mismatch.
Large movements, turns, transformations, occlusion and weak source images can make identity harder to preserve. Start with a sharp portrait and smaller motion, then increase complexity gradually.
That depends on the selected model. Some models respond better to positive descriptions of the desired state. Put universal requirements in plain positive language and consult the model’s current documentation for negative-prompt support.
Generate controlled shots separately and edit them together. Some workflows also let you use a finished clip’s last frame as the next starting image, but continuity still requires review.
Runbo Li is the Co-founder and CEO of Magic Hour, where he builds AI video and image tools for content creation. He is a Y Combinator W24 founder and former Data Scientist at Meta, where he worked on 0-1 consumer social products in New Product Experimentation. He writes about AI video generation, AI image creation, creative workflows, and creator tools.
