Kling 4.0 AI video model adds 30-second clips and 10 keyframes

Kling AI, the video generation platform operated by Chinese tech company Kuaishou, has released Kling 4.0, a model update centred on motion performance, creative control and longer storytelling. Details of the release were published on October 16, 2026. The model can natively produce up to 30 seconds of video in a single generation.

Kling 4.0 accepts up to 15 multimodal reference assets, covering images, video and audio, and up to 10 keyframes that let creators plan character states, changes in action and transitions between scenes. Users can also extend videos they have already made, leaving more room for continuous action and stories told in several parts.

Beyond the text prompt

AI video tools have largely depended on written prompts, which struggle to pin down complex camera moves, character actions and scene changes. Even when single shots look convincing, stitching them together can produce characters whose appearance drifts or movements that break continuity. A Kling AI spokesperson said the new features target two problems in particular: creative control and narrative continuity.

The two tools do different jobs. References tell the model what a video should look and feel like, while keyframes define how the story should progress. In principle this could cut down on generating clips over and over and picking the best by hand, though how well it works will depend on how complex the scenes are.

Consistency becomes the battleground

According to the spokesperson, the main technical bottleneck for longer AI video remains stability, controllability and coherence across extended durations. Longer clips require a model to hold character identity, spatial relationships, motion, camera direction, lighting and audio-visual sync steady. The company argues that visual fidelity is turning into a baseline expectation, with the next priorities being finer control and a stronger grasp of scripts and storyboards.

Kling AI expects video tools to expand from single generation tasks into a workflow spanning ideation, reference input, shot generation, editing, extension and iterative refinement. For people making ads, animation or films, reproducing characters reliably and lowering the cost of revisions matter as much as realism.

More control does not mean AI can make a film on its own: story development, shot selection and final editing still rely on human judgment. The real test for tools like Kling 4.0 is whether they can deliver the right shots in the right order with enough consistency to make production predictable rather than a lottery of repeated attempts.