Kling 3.0

Positioning Overview

kling-3.0/motion-control is a motion-driven video generation pipeline: you provide a reference image of a person and a reference video containing motion, and it outputs a new video choreographed by that motion. The character's appearance comes from the image, while motion and expressions come from the video; an optional prompt is used to supplement the scene, style, and camera. It is suited for developers who need to transfer live-action performance onto a specified character, such as short dramas, talking-head videos, and e-commerce showcases — any scenario that requires reusing the same set of motion footage.

Core Highlights

• Decoupled character and motion inputs — the character's appearance is determined by the reference image and the motion rhythm by the reference video, so the same performance can be applied to different character assets.
• Output duration adapts to the character orientation mode — when the character orientation follows the video, output can be up to 30 seconds; when it follows the image, up to 10 seconds, so you can choose based on the shot type.
• Supports retaining the reference video's original audio — the reference video's audio track is kept by default, so motion and sound are output in sync, eliminating the need for separate audio alignment.

Capability Description

Dual-asset-driven motion transfer: you must provide two inputs — a person image and a motion video. The image determines the character's appearance, and the video determines motion and expressions; the generated result inherits both.

Character orientation mode switching: character_orientation can be set to video or image. The former suits complex motion with a maximum duration of 30 seconds; the latter suits camera movement with a maximum duration of 10 seconds.

Prompt supplements scene and camera: the optional prompt can be up to 2500 characters and is used to supplement the scene, style, and camera direction without changing the motion and expressions determined by the reference video.

Selectable resolution and audio track retention: output can be 720p or 1080p, and you can choose whether to retain the reference video's original audio; higher resolution corresponds to higher billing.

Use Cases

• Short drama and talking-head creators reusing motion assets — upload a character portrait and a performance video to get a finished clip of that character performing the same motion, for serialized content production.
• E-commerce and brands creating character-based product demos — pair a model or virtual character image with a product demo motion video to produce a character showcase clip for detail pages and ad placements.
• Motion content that needs original audio preserved — with original audio retention enabled, directly output a video with a synchronized audio track for publishing as social media short videos.

Limitations and Selection Guidance

Both input assets have hard requirements: the person image must show the head, shoulders, and torso clearly, have an aspect ratio between 2:5 and 5:2, and be no larger than 10MB; the reference video must be 3 to 30 seconds long, a single continuous shot with the subject visible, and no larger than 100MB. Motion and expressions are determined by the reference video, and the prompt can only adjust the scene and style — it cannot rewrite the motion itself. If you need text-to-video or image-to-video, choose other pipelines in the same series; long narrative content still needs to be generated in segments and stitched together afterward.