Kling 4.0 adds shot control and 30-second clips to AI video

For anyone making an advertisement, an animated sequence or a short film with AI, the hard part has rarely been getting one striking shot. It has been getting the tenth shot to show the same character as the first, and making the cuts between them read as one story. Kling AI's new model, Kling 4.0, is built around that problem rather than around the picture alone.

The mechanism rests on two tools. The first is what the company calls multimodal reference assets: the model accepts up to 15 inputs, which can be images, videos or audio, to learn how a video should look and feel. The second is keyframes, with up to 10 control points that plan a character's state, a change of action or a scene transition. The distinction is simple: references tell the model what the film should look like, and keyframes tell it how the story should move from one moment to the next. Until now, most AI video tools worked mainly from a text prompt, which makes it hard to describe a character's appearance, a camera move or the rhythm of an action precisely.

Duration is part of the update too. Kling 4.0 generates up to 30 seconds of video natively in a single generation, and existing videos can be extended. But the company's spokesperson is clear that length does not solve the core issue. As clips get longer, the model has to keep character identity, spatial relationships, action continuity, camera direction, lighting and audio-visual synchronisation consistent, and a few seconds of convincing footage can be obtained through repeated attempts and selection in a way a longer work cannot.

For professional creators, the gain would be fewer wasted attempts, if the features perform as described. The company says the company's next priorities are more precise creative control and a better understanding of scripts and storyboards. No timeline for wider availability was given in the material available for this article, and the effectiveness of these tools will depend on how complicated the scenes are. The figures and claims come from Kling AI itself, and no independent test is reported here.

The test that matters, then, is not whether an AI model can produce a convincing shot. It is whether it can deliver the right shots, in the right order, consistently enough to serve a larger story. Human judgment remains essential for story development, shot selection and editing. What the update suggests is that the field is moving from making individual clips to making the production process itself more predictable.