Kling 4.0 is Kuaishou's fourth-generation AI video model, and it arrives with two directed workflows that change how you build a shot: keyframe storyboards and omni reference. Both produce video with synchronized stereo audio, from 3 to 30 seconds at 720p or 1080p.
Keyframe mode is storyboard control. Upload a first frame, up to 8 middle keyframes and a last frame — Kling 4.0 animates the journey between them. The total frame count is capped by duration: 6 frames at 5 seconds, 10 frames at 9 seconds or longer. That cap keeps motion natural instead of teleporting between beats.
Reference mode is the omni pipeline: up to 10 reference images, 5 reference videos, 3 voice references and 7 subject elements — 15 items in total — attached to a single request. The model keeps characters, products and styles consistent while you direct the action in the prompt.
The signature interaction is @ mentions. Type @ in the prompt and mention your attachments directly: @first_frame, @keyframe_1, @image_2, @video_1 or a voice's custom name. Every mention maps to a real asset, so instructions like "the jacket from @image_2 appears on the model from @image_1" just work.
Audio is native and stereophonic. Kling 4.0 generates two-channel synchronized sound by default, and when you attach reference videos you can choose to keep their original audio instead.
Direct your first shot in the Kling 4.0 studio — keyframes, references and @ mentions all work in the browser.