Take the performance, change the performer

Motion Control is the second task type inside the Video Enhancer.

Motion Control sits under Video Enhancer in the generation type dropdown, then behind a second task type dropdown that also holds Video Upscaling. The form defaults to upscaling, so this switch is a step people miss.

The two task types share a form and almost nothing else. Selecting Motion Control replaces the model, adds a required character image input and a prompt field, and changes what the source video is being used for.

Kling 3.0 Pro is the only motion control model, labelled in the picker with a 30 second maximum and a high fidelity marker. As with upscaling, there is no comparison to work through.

That 30 second figure is the ceiling in Video Orientation. The Image Orientation mode caps at 10 seconds instead, so the limit printed beside the model name is the best case rather than a fixed number.

One model, labelled with its 30 second ceiling.

The source clip supplies the motion, not the picture.

The Video Source block takes an upload, a direct video URL, or one of your previous completed tasks, accepting MP4 up to 1000 MB at a minimum of 300 by 300 pixels.

What this clip contributes is the movement rather than the appearance. Its subject will not appear in the output, so the thing to judge when choosing it is whether the motion is clean, well framed and unambiguous.

The character image is the subject that will appear in the output. It accepts an upload, a direct image URL or a previous completed task, taking JPEG, PNG, WEBP, GIF and AVIF up to 100 MB at a minimum of 300 by 300 pixels.

This block offers three input methods rather than the five available in image to video. The saved AI Actor and AI Character options are not present here, so a character you want to reuse has to come through as a file, a URL or a previous task.

The character image decides who appears. Three input methods, not five.

The prompt describes the motion you want, up to 2,500 characters.

The prompt field takes up to 2,500 characters and describes the movement or scene you want produced. It works alongside the reference video rather than replacing it.

Vague prompts produce unpredictable results here more than elsewhere, because the model is reconciling three inputs at once. Being explicit about how the character should move, or what the camera should do, is what keeps that reconciliation on track.

Character Orientation offers Video Orientation, labelled for complex motions, and Image Orientation, labelled for camera movements. This decides what the model transfers, and it also sets the maximum length.

Video Orientation is for a character performing actions such as walking, gesturing or dancing, and allows up to 30 seconds. Image Orientation applies camera moves like panning, zooming and tracking, and caps at 10 seconds. Front facing character images suit the first mode, and angled shots often work better in the second.

The single most consequential control in this form.

Optional extra elements, added one at a time.

An optional Elements section lets you add further items to the generation through an Add Element control. It is genuinely optional, and a first pass is usually better run without it.

Adding elements increases the number of things the model is reconciling at once. Establish that the character and the motion combine correctly before introducing anything else into the frame.

A single toggle, Keep Original Sound, preserves the audio from the source video in the output. Where the reference clip carries dialogue or ambience worth keeping, this saves reattaching it later.

Where the source is only there to supply movement and its audio is irrelevant, turning this off avoids carrying sound into an edit that will replace it anyway.

Keep Original Sound carries the audio across from the source clip.