Kling 4.0 doubles its AI video clip length to 30 seconds

Kuaishou's AI video unit has announced Kling 4.0, a model that doubles the longest clip it generates in one pass from 15 seconds to 30, Bloomberg reported on September 28, 2026.

Most of the rest of what the company said is about control rather than length. Kling 4.0 takes more than a dozen reference inputs at once, in text, image and video formats, and a user can set up to 10 keyframes, the still images that establish how a particular segment of the video should look.

Bloomberg reported that the model is still being tested with selected users, that a lite version is already open to annual Kling subscribers, and that the full release is coming in October. Kuaishou framed the launch as an attempt to close the gap on ByteDance, the Bloomberg report says.

Nothing else has been published. Kling's own site still leads with the 3.0 series, and there is no model card, API documentation or price for 4.0. Third-party pages that appeared before the announcement go further than the company has: OpenArt's Kling 4.0 page advertises output up to 4K HDR, 15 references and dialogue in nine languages. The pre-release settings published on September 24 by the API platform reAPI, drawn from Kling's own creator app, listed 720p and 1080p and no 4K, with a Flash tier capped at 20 seconds. Treat the extras as unconfirmed until Kuaishou publishes them.

That leaves 4.0 measured against the model it replaces. According to the live parameters listed by the API platform WaveSpeedAI, Kling 3.0 generates 3 to 15 seconds per request, in a tier set that includes native 4K, and supports up to six shots in a single request. The headline change in 4.0 is the 30-second ceiling.

The launch lands while the business around it is being reorganised. The Standard reported that Kling AI is raising its first pre-IPO financing at an US$18 billion pre-money valuation and aims to file for a Hong Kong listing in early 2027, after an earlier plan to spin the unit off at US$20 billion.

For a creator, the practical difference is an edit point. A 30-second generation covers a full spot or a short-drama beat that previously had to be stitched from two passes, and stitching is where character and lighting drift. Ten keyframes move shot planning into the generation request itself, closer to a storyboard than a prompt. None of that can be scheduled yet: the full model is an October release and no rate has been published.

Kling 4.0's lite version is live for Kling's annual subscribers, and the company says the full model follows in October.