Kling AI has released version 4.0 of its video generator, adding tools for shaping scenes, reusing video templates and producing speech in more languages. The update also introduces Kling 4.0 Flash, a faster, lower-cost variant with more limited output. The expanded controls offer creators more ways to guide a generation, while a comparison with other models suggests that added capabilities do not guarantee a clear lead in video quality.
More room to guide a video
Kling 4.0 can generate up to 30 seconds of video in one pass. Users can provide up to ten keyframe images to help define transitions, character states and story beats. The system’s prompt comprehension has also expanded to 8,000 tokens, giving users more room to describe what they want.
Its upgraded Omni Reference system accepts up to 15 reference assets in one generation. These can include up to 10 images, 5 videos with a combined maximum of 30 seconds, and 7 elements. Kling AI says users can also draw on storyboards, wireframes, image grids, white-model videos and depth-map videos as references.
These inputs give creators several ways to communicate visual direction. Keyframes can mark important moments, while reference material can help establish what a scene or character should look like. The system accepts a mix of materials, so a generation can be guided by more than a written prompt alone.
Editing templates and individual details
A feature called Creative Recreation lets users start with an existing video as a template, then replace products or characters. Kling AI says users can adjust details such as facial expressions, body movements, camera angles and backgrounds without changing the rest of the clip.
That approach makes the feature useful for modifying a specific part of a video while keeping its broader structure. Instead of describing every element from scratch, a user can begin with an existing sequence and choose which parts to alter. The degree to which changes preserve the rest of the clip is the company’s stated capability.
Audio, languages and a mixed model comparison
Kling 4.0 adds two-channel stereo audio and improved lip sync across more than ten languages, according to the company. Supported examples include German, English, Japanese, Korean and Spanish, along with various dialects and accents. In-video text generation also works in multiple languages and is designed to stay stable as the camera moves.
A comparison described in the source puts Kling 4.0 alongside Wan 3.0 and Seedance 2.5, both identified as Chinese competitors. Kling holds its own, but does not pull ahead, particularly in logical shot sequencing, physics and natural-sounding dialogue. That result may point to a slower pace of improvement in AI video models, though the next releases of Wan and Seedance will provide more context.
Some announced capabilities are not available yet. Kling AI lists 10-bit HDR output at 4K and 1080p, as well as a video extension function that would stretch clips to up to two minutes, as coming soon. The source describes them as upcoming rather than current features.
Flash trades output limits for speed and cost
Kling 4.0 Flash is designed to generate videos faster and at lower cost than the main model. Its maximum output is 720p resolution and 20 seconds of video, compared with the main model’s ability to generate up to 30 seconds in one pass.
The creation interface has also been redesigned. Kling 4.0 can be accessed through MCP and CLI, which makes it possible to build automated workflows with AI agents. Pricing for the Kling 4.0 API has not been announced, and API support is listed as coming soon.
For creators, the update combines longer clips and more reference controls with new audio and language features. Flash presents a quicker, lower-cost route with tighter output limits. The comparison with Wan 3.0 and Seedance 2.5, meanwhile, leaves open how much the expanded feature set translates into an overall improvement in the finished video.