Kling Motion Control 3.0 video model

Kling 3.0 rolled out in the beginning of March 2026 – it’s a motion guided image-to-video tool inside the Kling VIDEO 3.0 line. The system takes a still reference image plus motion input and turns it into a short moving clip. Think character animation where the body movement and expressions follow guidance you give.

The developer is Kuaishou the Beijing tech company behind a large short-video platform. The firm says it launched China’s first short-video community in 2011 and now runs a global social content platform.

What makes Motion Control 3.0 interesting is the attempt to fix a common AI video issue. Characters often move but their faces hands or identity drift or break. Kling’s guide says the tool aims for tight control of body motion and facial expressions using a reference image. The release notes also say users can bind facial parts using multiple images or video sources to keep the character consistent.

The research report says the system handles body face and hand motion as separate parts inside one DiT-based model. That setup aims to keep full body animation more steady while the character moves.

So the bigger goal seems clear. Kling wants to push motion control from a simple animation trick toward something closer to directed character acting. That does not mean movie level results every time. But the product messaging and the new research paper both suggest the company is putting serious work into this area.

The paper also claims more than 10x faster inference through distillation.