AI Video Prompt Guide
Image-to-Video Prompts: How to Write Motion, Camera Movement & Timing
In image-to-video, the starting frame already tells the model what the subject, composition, lighting, and style look like. Your text prompt should usually focus on what changes next: camera motion, subject action, environmental motion, direction, speed, and timing.
Test a prompt
Upload a starting frame, choose a camera movement, and generate a short motion test.
Open Image-to-VideoChoose a camera move
Compare 24 camera movements with diagrams and copy-ready starting prompts.
View Camera Movement PromptsBuild the camera clause
Combine movement, subject action, environment, framing, speed, and target model.
Open Prompt BuilderDirect answer
Let the image describe the image; use the prompt to describe motion
Both current Runway and Google guidance follow this principle for image-to-video: the input image provides the visual starting point, while the prompt should emphasize what moves and how it moves. Repeating every visible detail can add ambiguity without adding useful control.
A beautiful cinematic woman in a red coat inside a bright glass train station with warm sunlight and realistic photography.
If the uploaded image already shows this, most of the text repeats information the model can see.
The camera slowly dollies in toward the subject as she turns toward the window. Background travelers move naturally behind her.
This prompt spends its words on camera motion, subject action, and environmental motion.
Prompt anatomy
Five motion components to control
You do not need all five in every generation. Start with the smallest combination that describes the motion you care about, then add another component only when the first result shows why it is needed.
Camera motion
Dolly in, pan left, orbit right, tracking shot, crane up, locked camera.
Subject action
Turns, walks, reaches, looks up, runs, remains still.
Environmental motion
Curtains move, dust trails, water ripples, traffic passes, reflections shift.
Direction & speed
Slowly forward, rapidly left, steady backward, matching the subject.
Timing
Begins still, then moves; one action follows another; movement ends on a specific frame.
A practical image-to-video prompt structure
This is not a mandatory syntax. It is an editing framework that makes each motion variable easy to identify. Runway explicitly recommends simple prompts and iteration rather than forcing every possible component into one generation.
Copy-ready examples
6 image-to-video prompt examples with camera movement
Each example changes the temporal behavior of the starting image rather than spending most of the prompt on static visual description.
Locked-off subject motion
Locked camera. The subject slowly turns toward the window while curtains move gently in the background.
Why it works: Use this when the frame should stay stable and the motion comes from the subject or environment.
Open Locked-Off Shot guideSlow dolly in
The camera slowly dollies in toward the subject, ending in a medium close-up as the subject raises their eyes toward camera.
Why it works: The physical move, speed, end framing, and subject action are separated clearly.
Open Dolly In guideProduct orbit
The camera slowly orbits left around the product, keeping it centered while reflections shift naturally across the surface.
Why it works: The orbit defines camera motion while the changing reflections describe environmental motion.
Open Orbit Left guideTracking a moving subject
A smooth tracking shot follows the cyclist from the side at matching speed while trees slide past in the background.
Why it works: The camera is tied to subject motion and the background movement reinforces direction.
Open Tracking Shot guideAerial reveal
An aerial camera slowly pulls backward and rises from the cabin, revealing the surrounding forest and distant mountains.
Why it works: The prompt describes what changes over time instead of re-describing the starting image.
Open Aerial Pull Back guidePan to reveal
The camera slowly pans left from the doorway to reveal the subject at the desk. The subject remains still.
Why it works: A simple rotational move works well when the intent is to reveal adjacent space without camera travel.
Open Pan Left guideStart with a strong source image
The source image becomes the first frame, so blurry faces, distorted hands, unclear geometry, or other artifacts can be carried into motion and may become more noticeable.
Prefer a frame whose subject, composition, depth cues, and important background relationships already support the motion you want to generate.
When should you describe visuals again?
Motion-first does not mean visual description is forbidden. Add visual information when the video must introduce something the source image does not already establish.
- • A new object or element enters the scene.
- • The subject transforms or changes appearance.
- • Two existing elements must interact in a specific way.
- • The scene undergoes a dramatic visual transformation.
A better iteration workflow
Test the base motion
Start with the one camera or subject movement that matters most.
Inspect the failure
Decide whether the problem is direction, speed, framing, subject action, or scene motion.
Change one variable
Add or revise one instruction instead of rewriting the full prompt.
Save the working pattern
Reuse successful camera language as a baseline for similar shots.
Model differences matter
Standard cinematography terms travel well across AI video systems, but prompt interpretation, supported durations, negative-prompt behavior, and image controls can differ. Keep the motion description model-neutral first, then adapt it to the model's current controls.
Kling 3.0 adds precise shot control, reference-driven consistency, and a multi-shot storyboard in Kling 3.0 Omni. See the Kling camera movement guide.
Google's Veo 3.1 supports image-to-video and first/last-frame generation. See the Veo camera movement guide.
See the Runway camera movement guide for Runway's current motion vocabulary and iteration guidance.
Seedance 2.5 emphasizes longer one-take control and reference-driven camera language. See the Seedance camera movement guide.
Example: Runway's current Gen-4 guidance favors positive phrasing such as “Locked camera. The camera remains still.” rather than “No camera movement.” Google Veo documentation exposes its own prompt and model controls, so model-specific rules should be checked before copying a workflow unchanged.
Image-to-video prompt FAQ
What should an image-to-video prompt include?
Start with the motion that matters most. A practical prompt can include camera motion, subject action, environmental motion, direction, speed, and timing. You do not need every element in every prompt.
Should I describe the image again in the prompt?
Usually no. The input image already provides the visual starting point, including subject appearance, composition, lighting, and style. Re-describe visual details only when you need a transformation, a new element, or a specific interaction that is not already clear from the image.
How long should an image-to-video prompt be?
There is no universal ideal length. Start with the smallest prompt that clearly communicates the motion, generate a test, then add one useful detail at a time. Clear physical actions are more useful than long lists of vague cinematic adjectives.
How do I add camera movement to image-to-video?
Name the physical camera action directly: dolly in, dolly out, pan left, tilt up, tracking shot, orbit, crane, aerial pull-back, handheld tracking, or locked camera. Add direction, pace, and end framing only when they matter.
Should I use negative prompts?
This is model-specific. Runway's current Gen-4 guidance favors positive phrasing such as “Locked camera. The camera remains still.” instead of negative instructions. Other models may expose separate negative-prompt controls, so follow the target model's current documentation.
Put the prompt into a real workflow
Choose a movement from the visual guide, build a clean camera clause, then upload a starting frame and test the motion in the Image-to-Video tool.
Sources & verification
This guide separates general image-to-video principles from model-specific behavior. The motion-first guidance is cross-checked against current Runway and Google documentation, and the examples are written as model-neutral starting points rather than guaranteed commands.