After an initial early access phase, Kling 4.0 is now open to everyone

Kling 4.0 is available with videos up to 30 seconds, ten keyframes, and fifteen multimodal references to control the sequences.

Thirty seconds of video generated in a single pass, ten keyframes to organize the progression of a scene, and up to fifteen references to guide characters, sets, or camera movements. Kling 4.0, developed by Kuaishou, expands to broader availability following an initial early access phase launched in late September with Kling 4.0 Flash. The model adds several features designed to create longer sequences and control visual and audio elements, while consolidating generation and editing into a redesigned creative environment. Thirty Seconds of Video in a Single Generation

Duration is one of the main evolutions of Kling 4.0. While Kling 3.0 allowed for the production of sequences up to fifteen seconds, the new version doubles this limit to reach 30 seconds in a single generation, with a minimum of three seconds.

This extension makes it possible to organize multiple actions within the same video: character entry, camera movement, framing change, dialogue, or product appearance.

For creators, it notably reduces the need to stitch together multiple short generations to build a sequence. Continuity of characters, movements, and scenery can be refined within a single prompt.

Kling also announced a video extension feature allowing sequences to be extended forward or backward, across multiple operations, up to a total of two minutes. This functionality remains among the newly announced features slated for later availability. Ten Keyframes to Organize Shots

Kling 4.0 introduces a system of ten flexible keyframes that can be positioned at different points along the timeline.

Unlike first-and-last-frame control, which was already available in video generation, this feature allows users to define several intermediate steps.

The creator can specify when a character should appear, at what point a product should be revealed, or how the camera should evolve between multiple compositions.

These references also serve to organize multi-shot sequences featuring framing changes, tracking shots, pans, or circular camera movements.

However, the keyframes guide the model without acting as absolute editing constraints. Their presence does not guarantee that every transition or action will be executed at the exact intended moment. Up to Fifteen References in a Single Video

The Omni Reference system is also evolving to accept several types of assets simultaneously.

A single generation can combine up to fifteen reference elements, subject to limits specific to each category: ten images, five videos, and seven subjects, while adhering to the overall cap of fifteen elements.

The reference videos can represent a cumulative duration of thirty seconds. Voice references can also be associated with characters.

Each asset serves a specific purpose. One photograph might define a character's face, another might specify the features of a product, while a video serves to guide a movement or camera behavior.

Kling also accepts depth maps or simplified volumetric representations to guide movement.

This approach makes it possible, for instance, to use the structure of an existing video as a reference and then replace certain visual elements with other characters, objects, or products.

The fidelity of these transformations nevertheless depends on the quality of the assets and their interpretation by the model. Complex Movements and Stereo Audio

Kuaishou announces improvements regarding the stability of fast movements, particularly in combat, dance, sports, and action scenes.

The model aims to preserve spatial relationships between characters, pose consistency, and camera movements when multiple actions occur simultaneously.

The audio component evolves with support for two-channel stereo audio, compared to a single channel in the previous generation.

Kling also announces improved lip-syncing and expanded language support, notably including French, English, German, Portuguese, Japanese, and Korean, as well as several accents and dialects.

On-screen text generation has also been reworked. The model can produce lettering, logos, and emojis while seeking to maintain their legibility during camera movements.

Prompts can now reach up to 8,000 tokens, allowing for greater detail regarding composition, dialogue, movements, and the progression of a sequence. From 4K to 21:9 Cinema Aspect Ratio

Kling 4.0 offers outputs in 720p, 1080p, and 4K, with several aspect ratios including 16:9, 9:16, square, and 21:9 ultrawide.

The 4K resolution is not new to the Kling family: it was already available in the previous generation.

The announced evolution notably involves support for 10-bit HDR in 1080p and 4K, designed to retain more detail in highlights, shadows, and color transitions.

However, this feature is still listed as coming soon in the official documentation. The expanded availability of Kling 4.0 does not mean that all showcased features are immediately accessible. Generate and Edit from the Same Environment

The creation interface of Kling AI has been reorganized to bring together several operations previously scattered across different tools.

The new workspace accepts text prompts and multimodal references within the same interface. It also includes a timeline to review generations, add assets, and continue editing.

The Canvas and Agent features serve to organize project elements and assist with certain steps using conversational prompts.

Kling 4.0 also incorporates capabilities for editing existing videos: changing an expression, replacing an element, modifying the background, or tweaking a camera movement.

The documentation allows for up to five input video clips, including one primary video, for a maximum cumulative duration of thirty seconds.

These operations allow creators to work from existing footage rather than systematically re-generating from scratch. Kling 4.0 Flash Retains a Lighter Offering

The Kling 4.0 Flash version, available in limited early access since September 28 for Ultra Yearly subscribers, remains positioned for faster and less demanding generations.

It