Kling 3.0 Pro Motion Control

Kling 3.0 Pro Motion Control — Video-to-Video Motion Transfer

What is Kling 3.0 Pro Motion Control?

Kling 3.0 Pro Motion Control is a video-to-video motion transfer model from Kuaishou's Kling team. You give it two inputs — a single still character image and a 3-to-30-second reference (driving) video — and it returns a new clip in which your character performs the reference's exact movement while keeping its own face, expression and outfit. It separates appearance from action: the image decides who appears, the video decides how they move. The pro tier renders at 1080p for final delivery.

Because you show the model the motion frame by frame instead of describing it in a prompt, results are far more repeatable than prompt-only image-to-video — one reference clip can drive a whole cast of characters.

Key Features

• •Reference-driven motion transfer from a 3–30 s driving clip; output length matches the reference
• •1080p output on the pro tier (720p on the standard tier)
• •Facial identity, expression and outfit preserved through head turns, multi-angle motion and partial occlusions (hands, props)
• •Physics-accurate motion — grounded feet, natural balance and inertia, no "sliding on ice"
• •character_orientation to follow the reference (up to 30 s) or keep the still's facing (up to 10 s)
• •Optional prompt for scene, background or camera; keep_original_sound to retain the reference audio

Best Use Cases

Apply a trending dance or an expressive gesture to a custom character or avatar; drive brand mascots and concept art with a real performance; reuse one motion template across a series for consistent characters. It suits music videos, game trailers, IP campaigns and short-form social, replacing a motion-capture-and-rigging stage with a single reference clip. Dance, martial arts, walking, running and hand-gesture work all transfer cleanly — our test drove a wushu character through a full kung-fu form (high kick, lunge punch, horse-stance guard) with identity and footing intact.

Prompt Tips and Output Quality

Motion always comes from the video, so the prompt only adds scene, background or camera direction — keep it short. Use a clean, single-subject reference with continuous, moderate-speed motion and no hard cuts, and a character image with a clearly visible, reasonably large face. In a multi-person reference, the subject filling the largest area of the frame drives the motion. Expect sharp 1080p results that hold likeness across the whole clip.

How to Use Kling 3.0 Pro Motion Control

Kling 3.0 Pro Motion Control animates one still character with the motion of a reference clip. Supply an image (who appears) and a driving video (how they move), and the model returns a 1080p clip of your character performing that motion — face, expression and outfit kept intact.

Prepare your inputs

• •Character image: one subject, full-body or clean half-body, face clearly visible and reasonably large in the frame. JPG or PNG, aspect ratio between 1:2.5 and 2.5:1. A small or soft face is the main cause of likeness drift.
• •Reference video: 3–30 seconds, a single person, continuous and moderate-speed motion, stable framing, no cuts or scene changes. The output runs the length of this clip, so trim it to the beats you want.

Choose the right settings

• •character_orientation: video follows the reference's facing and allows output up to 30 seconds; image keeps the still's facing and caps output at 10 seconds.
• •prompt: optional, and scene-only — describe background, setting or camera. The motion always comes from the video, never the text.
• •keep_original_sound: keep the reference's audio track, or turn it off for a silent clip you will score yourself.

Recommended Settings

From our test run — a red-sashed wushu character driven through a kung-fu form — character_orientation=video, keep_original_sound=true, a roughly 5-second single-subject reference, and a short scene prompt ("…on a grey flagstone temple courtyard, steady wide shot") reproduced the high kick, lunge punch and horse-stance guard at 1080p with the character's face, costume and footing held throughout. For drafts and high-volume iteration, run the 720p standard tier first, then switch to pro for the 1080p final delivery.

Tips

• •Match the prompt's scene to the character image so the background stays coherent.
• •Only one character is driven — the largest subject in a multi-person reference.
• •Favour crisp, legible reference motion; very fast or blurry clips transfer less cleanly.

Examples to try

Each example below ran end to end through Kling 3.0 Pro Motion Control at the settings shown: a still character image plus a short single-person reference video, with character_orientation=video and the reference's own length (about five seconds) carried onto the output. The face, costume and footing come from the image; every bit of movement comes from the video.

• •Motion transfer from a reference clip (hero). Prompt: "The martial artist performs the form on a grey flagstone temple courtyard, steady wide shot capturing the full figure." The red-sashed wushu character reproduces the driving clip's kung-fu form — guard, high kick, lunge — with the face and costume intact and feet planted. Tip: a clear full-body character image with a large, visible face is the biggest driver of a clean result.
• •A different motion class — dance. Prompt: "The flamenco dancer performs on a dark wooden tablao stage under a single overhead spotlight, steady wide shot capturing the full figure." The dancer sweeps her arms, turns and steps while the ruffled red dress flares and her face and costume hold. Tip: favour continuous, moderate-speed reference motion; very fast footwork can blur.
• •Identity and ornate costume held through a turn. Prompt: "The armored warrior-mage performs in a vaulted stone hall lit by blue braziers, steady wide shot capturing the full figure." The knight-mage raises both arms and turns from profile to face the camera while every engraved plate, glowing rune and the cloak stay recognizable. Tip: stylised, game-style characters animate cleanly when the image is a clear full-body shot.

Also worth trying: "a sprinter driven through a start-line burst", "a chef guided through a plating flourish", or "a conductor following an orchestral cue" — each needs only a single-subject reference clip of the motion.

Does it keep my character's face?

Yes — identity, expression and outfit stay stable through turns and when hands or props briefly cover the face.

What resolution does the pro tier output?

1080p; the standard tier outputs 720p for drafts and high-volume iteration.

How long can the output be?

It matches the reference clip — up to 30 seconds in video orientation, or up to 10 seconds in image orientation.

Can the prompt change the motion?

No. This is motion transfer, not text-to-motion: movement comes only from the reference video.

What makes a good reference video?

One subject, continuous and moderate-speed motion, stable framing, and no cuts or scene changes.

How is 3.0 different from 2.6?

It preserves character identity more reliably, grounds the physics, and holds outfit and element detail better through turns.