Motion

Initiates the transformation of a static image into a dynamic video using a text prompt describing the desired motion effect.

Accepts a URL encoded string based prompt (motion=your%20motion%20description%20here).

Video generation typically takes 30-120+ seconds depending on the model and duration.

Prompt Examples

Motion prompts fall into two broad groups. Camera prompts move the viewpoint through a scene that otherwise stays still, while subject prompts animate what is actually in the frame — people, products, and graphics.

Camera motion

• "camera pans left to right, zooms out for aerial drone view"
• "slow zoom in on the subject"
• "camera orbits around the building"
• "gentle pan across the landscape"
• "dramatic zoom out revealing the full scene"

Subject and scene motion

• "the group starts running toward the camera"
• "the model turns slowly toward the camera and smiles"
• "the logo assembles from its individual shapes and settles into place"
• "wind moves through the trees and the water ripples"

Parameter Interactions

The motion parameter works in conjunction with other motion-related parameters:

• Use motion-duration to control the length of the generated video (default: 5 seconds)
• Use motion-seed to ensure reproducible results across multiple requests
• Use motion-model to select which AI model generates the video (default: “klingai”)

Combining with the Rendering API

Sizing parameters such as w are not applied to the source image of a motion request.

Generative parameters such as bg-replace can be combined with motion in a single request, but the two transformations run in sequence and the combined render time pushes the request onto the asynchronous path — it returns 423 while it works, and must be polled until it completes.

In practice we recommend against chaining them. A combined request is slow, hard to cache-warm, and inherits every limit of both transformations: if bg-replace fails, the default bg-replace-fallback behavior returns the original image, motion is dropped along with it, and the response is a 200 carrying a still rather than a video. Check the x-imgix-failure-reason response header when debugging — a 2xx alone is not proof you received a video.

Render the composite image first, store the result, then animate the stored image with motion.

Example Usage

Each scenario below starts from a single static image and adds nothing but a motion prompt.