Kling 4.0 is the video generation model that Kuaishou's Kling AI fully launched on the evening of October 15, 2026. Kling AI said the release focuses on three directions — realistic, controllable and professional: large-motion scenes are more stable, a single native generation can run up to 30 seconds, and the Omni Reference system can combine up to 15 reference items in one task. After a preview that began on September 28 with the Flash edition first, the full version is now open to all users, with API access for enterprises and developers to follow soon.
Long shots and big motion come first
AI video clips have mostly lasted 5 to 10 seconds, and characters often break down once they start running or turning. Kling 4.0 puts high-speed motion and complex camera work first, supporting pushes, pulls, pans, tracking and orbiting shots, so a 30-second generation can carry continuous storytelling instead of stitched fragments. Text input now accepts up to 8,000 tokens, and the model fills in camera planning and narrative detail from the prompt; a coming continuation feature will extend clips repeatedly, up to 2 minutes.
Omni Reference: state what must not change, once
Omni Reference combines up to 15 reference items per task — up to 10 images, 5 videos and 7 subjects — with characters, motion, camera moves and voice timbre specified separately. Up to 10 keyframe inputs pin down character states, scene changes and story beats. For advertising and short-drama teams, this control matters more than raw resolution: the same character keeps the same face across shots and product colors stop drifting, so storyboards can actually be executed.
Read the spec sheet carefully: 4K HDR is not here yet
The easiest detail to misread is image quality. Several reports mention 10-bit HDR output, but creators who tested the model note that the 10-bit 4K HDR spec is not officially live yet, and 1080P remains the practical ceiling for now; teams planning broadcast-grade delivery should wait for the follow-up update. On pricing, one tester's reference figures put a 30-second 1080P clip at about 450 credits, and drafting in 720P with Kling 4.0 Flash before rendering the final version is the safer workflow; official pricing on Kling's own pages prevails.
Access changes too: MCP and CLI
Kling 4.0 also connects to Kling's MCP and CLI, so creators can call Omni Reference and keyframe-to-video through AI agents and embed generation into automated pipelines. Kling discloses that the platform now covers 224 countries and regions, with more than 100 million users and over 50,000 enterprise customers. With 30-second takes, multi-item references and agent access arriving together, video generation competition is shifting from how good a single clip looks to whether a whole sequence can be delivered — exactly where short drama, advertising and film previsualization teams benefit first.
