Kling AI has released version 4.0 of its video generator. According to The Decoder, clips of up to 30 seconds are now possible. In addition, ten keyframe images can be used as input, and prompt understanding now extends to 8,000 tokens.
What the new features mean
Keyframes are key images. You give the model fixed images that should appear at specific points in the video, and the AI calculates the transitions in between. With more of these inputs, the course of a scene can be steered more precisely instead of being left to chance. This is exactly where AI video has often differed from traditional production, in which direction and continuity are central.
Tokens are the units into which a language model breaks down a text. A higher limit means a prompt can be more detailed, for example with information on camera movement, mood, characters and plot. Whether the model also implements long descriptions reliably, however, only becomes clear in practical use.
Clip length also plays a role. With very short sequences, users have to stitch several pieces together, and characters or lighting can vary slightly from clip to clip. Longer clips can reduce this effort.
How it compares
In comparison with Wan 3.0 and Seedance 2.5, Kling 4.0 keeps pace according to the report, but does not lead. The Decoder sees this as possibly a sign that the recently rapid development of AI video models is levelling off. This is a cautious interpretation and not an established finding.
In practice, this means that a ranking alone should not be the deciding factor when choosing a tool. Anyone creating videos with AI can use their own test scenes to check how well a model follows instructions. Important aspects include controllability via keyframes, the possible length and consistency across the entire clip. Depending on the project, a model that is not at the top can still be a better fit because of its control options.
