How to Direct Characters, Camera and Pacing in Kling 4.0

Kling 4.0, the video model from Kuaishou's Kling AI, gives creators three main levers for control: a single native generation of up to 30 seconds, Omni Reference with up to 15 combined reference assets, and up to 10 keyframes. Together they shift the work from hoping for one good clip toward specifying who appears, how the camera moves and when each story beat lands. The model launched on October 15, 2026, according to a Kuaishou press release dated October 16. DigIntoAI has not tested Kling 4.0, so this guide describes the controls as Kuaishou documents them and makes no judgment about output quality.

Characters: lock identity with Omni Reference

Kuaishou says Omni Reference accepts images, videos, subjects and voices in a single task, up to 15 references in total, including up to 10 images, 5 videos and 7 subjects. The stated goal is to keep characters, actions, shots and audio closer to the intended result.

For character consistency, the supported image types are the most useful part:

• Subject images for a specific person, product or object
• Three-view character sheets for a figure seen from several angles
• Grid images and storyboards to show several moments at once
• Wireframes for layout

Video references can carry a character's performance and action. Kuaishou also lists white-model videos and depth-map videos as valid references, which suits creators who block out motion in 3D first.

Camera: name the move, and borrow it from a reference

Kuaishou says Kling 4.0 has better understanding of push-ins, pull-outs, pans, trucking shots, tracking shots and orbit shots, with smoother changes in shot size and transitions. Two practical routes follow from that:

1. Write the move into the prompt. Prompt input now reaches 8,000 tokens, according to Kuaishou, leaving room to describe shot size, lens movement and staging per beat. Kuaishou also says the model expands prompts "with a director's eye."
2. Use a video reference for camera movement. Kuaishou lists camera movement among the things a reference video can supply.

Pacing: 30-second takes and 10 keyframes

A single pass can run up to 30 seconds, which Kuaishou pitches for long takes and continuous storytelling, so action, scene changes and emotional build-up can play out without stitching.

Keyframes set the beats inside that time. Kling 4.0 supports up to 10 keyframe inputs, and Kuaishou says they let creators specify character states, scene changes and story beats. A simple approach is to plan the clip as a storyboard first, then assign each panel to a keyframe.

A reference video can also supply narrative pacing. Kuaishou describes an upgraded creative recreation mode: keep the pacing, composition and style of a reference video while swapping in different products, characters and assets, aimed at ads and social content.

Image output

Kuaishou says Kling 4.0 supports output of up to 10-bit HDR, aimed at backlit, night and neon scenes and at color grading in post.

Coming later, according to Kuaishou

• Video extension: Kuaishou says it "will soon support" multiple extensions of up to two minutes.
• 4K, 1080p and 21:9: the release says Kling 4.0 "will also support" these formats.
• API: Kuaishou says Kling 4.0 will "soon" be offered to enterprise customers and developers through API access. No date was given.

Kling 4.0 is already connected to Kling MCP and CLI, according to Kuaishou, so AI agents can call Omni Reference and keyframe-to-video generation. For a different approach to generated video, see our coverage of Vast's open real-time world model EdenLive.

Price, free access and regions

Kuaishou has not announced prices, plan tiers, free-tier access or regional availability for Kling 4.0 in its release, and Kling's public pricing pages did not show Kling 4.0-specific pricing when checked on October 22, 2026. API pricing is also not announced, since the API itself is listed as coming soon.

Quick reference