Kling AI introduces Kling 4.0 model with 30-second clip generation and multi-control capability
Read more
Pandaily
pandaily.com

Kling AI introduces Kling 4.0 model with 30-second clip generation and multi-control capability

Kuaishou introduced its new video model, Kling 4.0, on September 28th, and on the same day opened limited access to the lighter version, Kling 4.0 Flash. Access to Flash is initially provided to Ultra Yearly subscribers, while the full model is planned for release in October, although an exact date has not been announced.

The main improvement lies in video duration: Kling 4.0 can generate up to 30 seconds of video in a single pass, which is double the previous limit of 15 seconds in Kling 3.0. This feature supports long shots and continuous storytelling within one clip. Additionally, a video extension feature will soon be available, allowing users to continue a clip multiple times to reach a duration of about two minutes.

Control over Generation and Editing

Another focus is placed on process management capabilities. One task can now combine up to 15 multimodal references, which can be drawn from 10 images, five video clips, and seven objects. This allows characters, movement, camera work, and sound to adhere more precisely to the provided material. Users can also set up to 10 keyframes to fix character states, scene changes, and plot points.

Text prompt input has been increased to 8000 tokens, giving the model the ability to expand and refine short prompts. Editing tools allow users to add, modify, or delete objects and backgrounds in an existing video, as well as adjust style, weather, color, material, frame size, or angle. The remake mode analyzes the linguistic features and pacing of a reference video to create a new advertisement based on that structure, targeting social media, product advertising, fashion, beauty, and consumer electronics.

In terms of output quality, Kling 4.0 supports 4K and 1080p formats with 10-bit HDR, which is also expected soon. An ultra-wide 21:9 format, two-channel stereo sound, and improved lip synchronization are also included. The company claims that motion, tracking shots, and rotating camera movements have become more stable. Voice support covers Chinese, English, Japanese, Korean, Spanish, Portuguese, German, French, and Hindi, including American, British, and Indian English accents, as well as several Chinese dialects such as Cantonese. The model can also display multilingual text, emojis, and logos on screen.

The Kling 4.0 Flash version is designed for high-volume content creation due to faster generation and lower cost. Kuaishou has also updated its creation page by consolidating links to images, videos, and audio into a single input field, adding a timeline for previewing and editing, and a canvas workspace with an assistant agent. API prices for the new model have not yet been published.

Similar stories

Inspur unveiled MetaBrain SD200 Ultra supernode with 128 domestic chips for running Kimi K3 model
Read more
pandaily.com

Inspur unveiled MetaBrain SD200 Ultra supernode with 128 domestic chips for running Kimi K3 model

Inspur Information has introduced the MetaBrain SD200 Ultra artificial intelligence supernode at the AICC2026 Artificial Intelligence Computing conference. This node is positioned as a capability-class system for processing tasks involving trillion-parameter models and agents. The official English name of the system is Inspur / MetaBrain.

According to company data and information agencies, the node integrates 128 domestic AI chips, provides access to 8 TB of unified accelerator memory, and 64 TB of host memory. It is capable of running the Kimi K3 model from Moonshot AI, which has 2.8 trillion parameters, on a single machine with a token generation latency of less than 5.85 milliseconds, equating to approximately 170 tokens per second for a single user, according to Inspur's estimates.

The architectural features of the node include the 3D Hyper Mesh factory, which ensures native, semantics-sensitive memory communication with a claimed latency of 0.69 microseconds. Short-reach copper interconnects and symmetric memory are also utilized, allowing accelerators to directly access the memory of remote nodes. Inspur claims that the AllReduce operation execution time has been reduced by approximately 3.5 times compared to previous solutions.

Furthermore, integrated 'superoperators' for the KDA, gated MLA, and MoE stages of the Kimi K3 model reduce the number of operators by about ten times while increasing inference performance by more than threefold. These figures are vendor claims. Materials also indicate the potential to support models up to 10 trillion parameters on a single node.

Another product, MetaBrain HC2000, is a rack with multiple accelerators designed for capacity-class inference, boasting a claimed tenfold tokens-per-investment throughput under agreed SLA constraints. Although various publications describe the accelerators only as domestic or local AI chips, none of the primary sources analyzed for this review name the silicon manufacturer or the SKU within the SD200 Ultra. Therefore, this article does not attribute the node to Ascend, Cambricon, or any other specific supplier.

Popular