Kling 4.0

Kling AI's next-generation native multimodal video model, built for professional creators and commercial content teams and announced on September 28, 2026. Text-to-video, image-to-video, reference-based generation, and video editing run in one workflow, with up to 30 seconds of video in a single pass.

Kling AI has announced that Kling 4.0 will launch in October 2026, without a confirmed date. Kling 4.0 Flash has been in limited early access on Kling's own platform since September 28. Runware will add both models to its API as soon as they're released.

Yes. On Runware, Kling 4.0 and Kling 4.0 Flash will run through the same API as every other model: one endpoint, one API key, and the same request shape. If you already call Kling VIDEO 3.0 on Runware, moving up will be a model ID swap.

Yes. The Kling VIDEO 3.0 family is live on Runware now, including Kling VIDEO 3.0 Omni Pro, Kling VIDEO 3.0 Omni 4K, and Kling VIDEO 3.0 Turbo, alongside Kling IMAGE models.

A lighter, faster model in the Kling 4.0 family, built on Kling 4.0 with a subset of its capabilities and designed for frequent creation, combining fast generation with excellent value. It generates 3 to 20 seconds per pass at 720p, and supports text-to-video, image-to-video from a first frame, and Omni Reference, but not end frames or multiple keyframes. Kling hasn't announced its audio support or reference limits yet.

Kling 4.0 is built for finished work: clips up to 30 seconds, 1080p or 4K output, keyframe direction, and video editing. Kling 4.0 Flash is built for frequent, fast generation at 720p, such as drafting ideas before rendering the take you keep on Kling 4.0.

Generation length doubles from 15 to 30 seconds per pass. Up to 10 keyframes join start and end frames, Omni Reference grows to 15 combined inputs including up to 5 videos, audio moves to dual-channel stereo with dialogue in more languages, and prompts can run to 8,000 tokens, which Kling 4.0 can expand and refine from even a short prompt.

From 3 to 30 seconds in a single generation. Kling has also announced extending a video multiple times, up to 2 minutes in total, as coming soon.

720p, 1080p, and 4K, in 21:9 ultrawide, 16:9, 1:1, 9:16, or Auto. 10-bit HDR at 1080p and 4K is coming soon.

Omni Reference takes up to 15 assets per request: up to 10 images, up to 5 videos with a combined length of 30 seconds, up to 7 subjects (3 of them from video), and voice references, the only kind of audio reference. Images can be subject shots, three-view references, image grids, storyboards, or wireframes. Reference videos can guide performance, motion, camera movement, and pacing, and can be depth maps or untextured 3D renders.

Yes. Make targeted changes to expressions, body movement, camera position and movement, style, or background while the rest of the scene stays stable, remove elements, or replace characters and scenes. An edit takes up to 5 input clips, including 1 primary video, each 3 to 30 seconds long and up to 30 seconds combined.

Yes. It generates dual-channel stereo audio with tighter lip sync, and dialogue in Chinese, English, Japanese, Korean, Spanish, Portuguese, German, French, Hindi, and more, with accents and dialects such as American, British, and Indian English, Cantonese, and Sichuanese. On-screen text, logos, and emoji in the same languages stay legible as the camera moves.

Runware will publish pricing for Kling 4.0 and Kling 4.0 Flash when they launch. As with every model on Runware, you'll pay only for what you generate, with no subscription or minimums.

Our Migration Skills and docs help map an existing integration over — same request shape as every other Runware model, so the change is usually just field names and auth. See the migration guide for details.