Kling 4.0 can generate 30-second videos in a single pass

Chinese company Kuaishou has launched the video model Kling 4.0 globally on October 15, 2026, according to its own press release. The model creates video from text, images and other inputs.

According to Kuaishou, the model can generate up to 30 seconds of video in a single pass. The company highlights realism, control and professional quality, and says motion should be more stable in fast action and complex camera work. That is Kuaishou's own assessment.

Control with references and keyframes

Kling 4.0 supports what Kuaishou calls Omni Reference. This means you can give the model examples to work from. You can combine up to 15 references and up to 10 keyframe inputs. Keyframes are images that control how the video looks at specific points.

According to the API platform fal.ai, which is a third party, the 15 references cover up to 10 images, up to 5 videos totaling 30 seconds, as well as voices and elements.

According to fal.ai, the full model can also edit existing video, for example facial expressions, body movement, camera position and style. This is done with one main video and up to 4 supporting clips totaling 30 seconds.

Flash is the lightweight version

According to fal.ai, Kling 4.0 Flash has been in limited early access since September 28, 2026. It generates 3-20 second videos, is limited to 720p and has no keyframes. The full model generates 3-30 seconds and supports 720p, 1080p and 4K.

The full model has two-channel stereo audio. 10-bit HDR for 1080p and 4K has been announced as coming. The Flash specifications come from fal.ai and not from Kuaishou, and the sources disagree on how widely Flash has been rolled out.

Pricing and availability

Kuaishou's press release mentions no price for Kling 4.0. It is therefore unclear what a video costs, and whether Flash is cheaper than the full model.

Fal.ai says Kling 4.0 will come to the platform through the same API as earlier Kling models. That is a plan and not something that has happened yet.