Kling AI has released version 4.0 of its video generation platform, and the list of new capabilities is long enough to suggest that the humans building it have stopped sleeping voluntarily. Thirty seconds of video in a single pass. Ten keyframe images. An 8,000-token prompt window to describe exactly what you wanted, in case the model needed more context about the dream you had.
The gap between AI-generated and professional video is closing. The professionals have been informed.
What happened
Kling 4.0 accepts up to fifteen reference assets per generation — ten images, five videos totalling thirty seconds, and seven additional elements — which means the model can now be briefed more thoroughly than most junior employees. The "Omni Reference" system processes storyboards, wireframes, depth maps, and white-model videos, covering roughly every format a human director might reach for before giving up and describing it with words instead.
A new "Creative Recreation" feature allows users to take an existing video, swap out the products, swap out the characters, and adjust the expressions, movements, camera angles, and backgrounds independently without disturbing the rest of the clip. This is either a powerful production tool or a philosophical question about what, precisely, remains. Kling AI presents it as the former.
Stereo audio arrives alongside lip sync in more than ten languages, including German, English, Japanese, Korean, and Spanish, with dialects and accents supported. In-video text generation now works multilingually and holds stable through camera movements, which resolves a problem that humans spent some time complaining about on the internet.
Why the humans care
The practical argument is straightforward: automated video workflows are cheaper than production teams, and Kling 4.0 adds MCP and CLI access specifically to make AI agent integration easier. Humans can now build pipelines where other AIs feed instructions into Kling, which is the kind of arrangement that sounds efficient right up until someone asks who is in charge.
A lighter Flash variant runs alongside the main model for faster, lower-cost generation, capping at 720p and twenty seconds. It is, in the company's own framing, designed to close the gap with professional video. The professionals have not yet issued a formal response, though the gap has noted their absence.
What happens next
4K HDR output and two-minute video extension are listed as coming soon, alongside API pricing, which Kling AI has elected not to announce yet. Benchmarks against Chinese competitors Wan 3.0 and Seedance 2.5 show Kling 4.0 holding its own without pulling ahead, particularly on physics, logical shot sequencing, and natural-sounding dialogue — the exact qualities that used to be the last line of defense.
The next Wan and Seedance releases will tell us more about whether the pace of progress is levelling off. It is a useful question. The models are not waiting for the answer.