How do you use Kling motion control with an AI character?

Kling motion control copies the movement from a 3 to 30 second reference clip onto a still image of your character. Our method: rebuild the clip’s first frame with your character in an image-editing model, then run motion control with one prompt, “replace the girl in the video. Use the background from the image and keep it static.” It’s the rented, fastest route in our guide to making AI influencer videos.

What is Kling motion control?

A Kling feature that takes two inputs, a still image of a character and a video of someone moving, and makes a new video of the character doing that movement. Kling is a paid online service from Kuaishou, billed in credits. You don’t need a GPU of your own.

Kling’s motion control guide lists two versions (checked 2 Oct 2026): VIDEO 3.0 Motion Control and VIDEO 2.6 Motion Control, each in Standard and Professional mode. Kling has said 4.0 is coming; its motion control guide still lists 3.0 and 2.6. Version details are in our Kling vs Wan vs LTX comparison. The course plans to refilm its Kling lessons on 3.0.

What clip can you use?

A clip you filmed of yourself doing the move, or licensed stock footage. Never a real creator’s TikTok or Reel, or anyone else’s clip: it’s their work and their likeness, and the course only builds original characters. The clip also has to meet Kling’s own rules, or the motion comes out wrong.

These are the requirements from Kling’s motion control guide, checked 2 Oct 2026:

Filming your own is easy: phone on a tripod, plain background, just you, the full body in frame, 10 to 15 seconds of the move. That covers the motion rules in the table.

Why rebuild the first frame instead of using a portrait?

Because Kling needs the image and the clip to match. Its guide says the character’s proportions must match the motion video, full-body with full-body, half-body with half-body. A frame rebuilt from the clip itself matches by definition: same framing, same pose, same distance from the camera. A random portrait of her almost never does.

A portrait breaks Kling’s matching rule as soon as its framing differs from the clip’s. If you do try one, check the result for a body that doesn’t fit the motion.

It also gives you control of the scene. You can change her outfit or the room in the rebuilt frame, and the prompt tells Kling to keep that background.

How do you run it, step by step?

Get a clip, take its first frame, rebuild that frame with your character, then give Kling the rebuilt frame, the clip and the prompt. Check the result for hands, face and background before you post. The steps below are the course’s flow, with the settings named the way Kling’s guide names them.

1. Get a clip that meets the rules above. Trim it so the move starts in the first second.
2. Take the first frame. Pause on frame one and export it, or with ffmpeg: ffmpeg -i clip.mp4 -frames:v 1 frame.png.
3. Rebuild the frame with her. Use an image-editing model such as Nano Banana Pro or Qwen-Image-Edit. Give it the frame and your own AI images of her, and ask it to replace the person with her while keeping the pose, framing and position. The course’s trick: upload the images to ChatGPT, number them (“image one, image two”), describe the change in plain words, and let ChatGPT write the prompt for the edit model.
4. Check the rebuilt frame. Whole body and head visible, same proportions as the clip, one person, her face clearly hers.
5. Open Motion Control in Kling and upload the rebuilt frame as the image and the clip as the motion reference.
6. Character orientation: leave it on “Character Orientation Matches Video”, the default.
7. Prompt: replace the girl in the video. Use the background from the image and keep it static.
8. Pick the version and mode, generate, and check the result.

Her face in the video is only as good as her face in the rebuilt frame. That frame should come from her character LoRA, not from a one-off edit. Our consistent AI character guide covers locking the face.

Why does that prompt work?

Our reading of it: each part does one job. “Replace the girl in the video” tells Kling the task: the movement comes from the clip, the person comes from the image. “Use the background from the image” keeps your rebuilt scene, not the source clip’s. “Keep it static” stops the background from drifting while she moves.

Keep it that short. A long prompt describing her face or body competes with the image, and the image is what holds her look.

What does Kling motion control cost?

Credits per second of output. On 2 October 2026, Kling’s guide listed VIDEO 3.0 motion control at 9 credits a second in Standard mode and 12 in Professional, and VIDEO 2.6 at 5 and 8. A 10-second clip on VIDEO 3.0 Standard is 90 credits. What a credit costs depends on your Kling plan.

Every run costs credits, including the ones you throw away, so the cheapest habit is a clean clip and a checked frame before you press generate. If you’d rather pay for GPU hours than credits, the open-weight options are compared in Kling vs Wan vs LTX.

What can go wrong?

• Body proportions look wrong. The image and the clip don’t match. Rebuild the clip’s first frame instead of using a portrait.
• The video is shorter than the clip. Kling’s guide says fast or complex movement can shorten the output. Use steadier moves.
• The motion glitches mid-clip. The source has a cut or the camera moves. Use one continuous shot from a fixed phone.
• The wrong person moves. There are two people in the clip. Kling follows whoever fills most of the frame. Use clips with one person.
• Her face drifts. The rebuilt frame wasn’t really her. Remake it from her LoRA images. On VIDEO 3.0, Kling’s guide offers Element Binding: a face Element built from clear close-ups of her at several angles (it uses facial information only, not hair or clothing), and it only works with “Matches Video”.
• You used someone else’s video. Delete it. Film your own, or license stock. For dance trends, see how to make an AI dance video.
• The post is unlabelled. TikTok requires a label on realistic AI content. Add it every time.