Kling·by KuaishouMotion control
Kling v3.0 Pro Motion Control
Kling 3.0 Motion Control animates a character image with a motion clip: the character performs the clip's movement and expressions. The image sets who and where, the clip sets the motion, and the prompt is optional. Clips run 3–30 seconds (up to 10 when following the image) and are billed by their length. Pro renders the finer texture; iterate on Std.
Parameters
| Parameter | Type | Default | Range or options |
|---|---|---|---|
Character orientation character_orientationFollow the clip: facing, framing and camera come from the motion clip — best for turns, spins and dance. Follow the image: they come from the still, and camera moves can be written in the prompt. | select | video | Follow the clip (up to 30 s) (video), Follow the image (up to 10 s) (image) |
Keep the clip’s audio keep_original_soundCarries the motion clip’s sound, music included, into the result. The model generates no audio of its own. | select | false | true, false |
Inputs
Files this model takes in addition to the prompt.
| Input | Accepts | Max files | |
|---|---|---|---|
Character image | image | 1 | Required |
Motion clip video | video | 1 | Required |
LUVI Prompt Engine rules for Kling
TARGET MODEL: Kling Motion Control (Kuaishou, v2.6 and v3.0). Motion transfer: a character IMAGE plus a driving VIDEO; the character performs the clip's movement and expressions. The image supplies the character, background and every other element of the picture. The clip supplies the action. The prompt is a side channel for what neither shows — it is optional, and an empty prompt is a correct answer.
Rules that must hold
- Never describe the action. The clip IS the choreography; Kling's own guide says the action need not be described, and over-directing movement fights the reference motion. Delete every sentence that narrates what the body does (spins, steps, raises arms, dances, walks, turns, jumps) — even when the user wrote it.
- If nothing remains after the choreography is removed, return an EMPTY prompt. Do not pad it, do not restate the task ("replace the character in the video with the character from the image" is inert), do not write a placeholder.
- Never contradict the image. Face, hair, body, costume and set come from the image; a prompt that changes them fights the picture. Wardrobe detail is allowed only where it agrees with what the user said is in the image.
- Write only what the pixels cannot carry: wardrobe detail, an added background subject ("a corgi runs in, circling around her feet"), light, hour, weather, atmosphere, style ("neon lighting, cyberpunk city, volumetric fog"; "golden hour, film grain").
- Camera: never ADD camera words. A camera move written by the user is kept only if it is one of Kling's presets — zoom in, zoom out, camera up, camera down, fixed position — and it takes effect only when the character follows the image; when it follows the clip, the clip's own camera is used.
- Plain prose only, at most 2,500 characters. No weight syntax, no brackets, no dialogue lines, no shot cards, no timestamps — the model generates no speech and no lip-sync.
- Delete resolution and quality garnish ("4K", "masterpiece", "ultra detailed"), durations, aspect ratios and negative lists. Length comes from the clip, framing from the image or the clip, and there is no negative prompt.
What works best
- One or two short sentences. Most good motion-control prompts are a clause of light and setting, or nothing.
- Keep the user's own nouns for anything they want added to the scene.
From the LUVI Prompt Engine rules for this family — the same rules the workspace rewriter and LUVI's MCP connector apply.