Kling AI has announced the full availability of the Kling 4.0 model. Video generation now lasts up to 30 seconds in a single pass, up from the previous 15 seconds. The model generates stereo audio instead of mono and supports Portuguese, German, French, and Hindi. Kling 4.0 can be invoked not only on the kling.ai website but also from AI agents via MCP and from the terminal via CLI. Export to 10-bit HDR and an official API for developers and enterprises have been announced but have not yet been released.

image
image

What happened

Kling AI has opened Kling 4.0 to all users. Generation duration has doubled from 15 to 30 seconds per pass. Users can set up to 10 keyframes to control the timeline, and the set of combined references has been expanded to 15 slots: up to 10 images, up to 5 videos with a total duration of up to 30 seconds, and up to 7 characters. Instead of mono, the model generates stereo audio, and the language list has been expanded to include Portuguese, German, French, and Hindi. The model is separately available via MCP for connecting to AI agents, such as Claude, and via CLI for terminal automation: both tools support task submission, execution tracking, and result retrieval, and they consume Kling account credits. Announced but unreleased features include 10-bit HDR export at 1080p/4K resolutions and an API for developers and enterprises.

Context

From an evidence perspective, this is a distribution event rather than a scientific release: the announcement is entirely vendor-driven, with sources including the kling.ai blog, the Kling AI account on X, and a secondary reprint by runtimewire. The materials do not include a tech report, benchmarks, or evaluation methodology. Most claims appear to be incremental increases in control parameters—duration, number of keyframes, and references—without indications of a new architecture or retraining. A comparison with the previous version was published by the vendor itself in a blog post titled "Kling 4.0 vs. 3.0: Comprehensive Comparison & Upgrade Guide," which is a marketing document rather than an independent assessment. However, the verifiable part of the announcement is the actual availability via MCP, the protocol through which agents like Claude connect to external tools, and via CLI: this can be reproduced manually and tested on one's own tasks without taking the demos at face value.

Why this matters for the industry

The main industry signal is not the 30 seconds or the stereo audio, but the fact that Kling is transforming from a web application into a callable backend for agents and script pipelines: video generation can now be invoked from scripts, agents, and automated pipelines without waiting for an official API. This changes how video models are integrated into the workflows of studios and developers and gives startups an early window on a new distribution surface. A significant limitation remains: without an official API, pricing, and latency data, MCP and CLI are account automation with credits, not a reliable backend, so it is premature to build critical pipelines on them. If the promised API is released with limits and HDR export, scenarios debugged on MCP and CLI will migrate to the official contract, and Kling will establish itself as a full-fledged video backend vendor—this is currently an interpretation, not a guaranteed outcome.

Why this matters for users

Kling 4.0 can be tried right now on kling.ai: 30-second clips, 10 keyframes, and stereo audio are available, and all of this should be tested on one's own tasks rather than vendor demos. Developers can connect the model via MCP to their agents, including Claude, or run generations via CLI from the terminal and scripts; all operations consume Kling account credits. A prototype of an agentic video pipeline can realistically be assembled in one day: an agent receives a brief via MCP, manages the storyboard through keyframes, and the CLI fits into nightly batch tasks and CI. This is sufficient for experiments and failure-tolerant scenarios; for serious production, it is advisable to maintain one's own error monitoring, retries, and quality metrics from day one, as there are no guarantees with the account credit wrapper.

What is still unknown / limitations

There is no direct evidence of quality for the release: no vendor metrics or independent benchmarks of Kling 4.0 against 3.0 have appeared, and it is unknown if they will be available by the time the API is released. The claim that character consistency has become practical for the first time with 4.0 is a hypothesis, not an established fact: the conclusion precedes the evidence. Promises of stereo audio and improved speech synchronization with multilingual support cannot be measured based on the announcement data. The release dates for the API and 10-bit HDR, their pricing, limits, and latency are unknown. MCP and CLI work on personal account credits without SLA or guaranteed limits, so they are tools for prototypes, not for production.

Sources

Author

Look at AI, editorial team