Kuaishou releases the Kling 2.1 series

Background

AI video generation models take a written description, and in some cases a reference image as well, and produce a short clip. Viewers and developers usually judge the output on a few things: whether the objects in the scene move naturally, whether they behave the way physical objects should, and whether the result actually follows what the prompt asked for. Kling is the video generation line from Kuaishou, and the 2.1 series is its update from the middle of 2025^[1].

What was released

Kuaishou released the Kling 2.1 series at the end of May 2025^[1]. Rather than a single model, the release came as a set of three tiers: a 720p Standard mode, a 1080p High Quality mode and a 2.1 Master mode^[1]. The tiers differ mainly in resolution and in how much generation quality they are meant to deliver, so users can pick between faster, cheaper output and more detailed output.

Alongside the new tiers, Kuaishou said the series improved motion, physics simulation and prompt responsiveness^[2]. Each of those points to a different part of the problem. Motion is about how believable movement looks across the frames of a clip, for instance whether a walking figure keeps a steady gait rather than drifting or stuttering. Physics simulation is about whether objects obey everyday physical expectations: things fall down rather than up, solid objects do not pass through one another, and cloth or hair reacts plausibly to movement. Prompt responsiveness is about instruction-following, that is, how much of what the user described actually shows up in the clip and how much the model substitutes its own ideas.

Pricing and access at launch

According to reporting by QbitAI, both 2.1 versions kept the same credit prices as their predecessors^[3]. Credits are the units users spend to generate video, so holding the price steady means the new versions arrived at the same cost per generation as the models they replaced.

QbitAI also reported that, at launch, the Standard (720p) mode supported image-to-video only^[3]. Image-to-video means the model animates a still picture supplied by the user rather than starting from a text description alone, which makes the input image the main control over what appears in the clip.

Where it sits

The launch was not only a product announcement. In its financial results for the second quarter of 2025, Kuaishou listed the Kling 2.1 series as having launched in May 2025^[2], which places the release inside the company's own account of the quarter's progress. The series is therefore both a technical update to the video model and a milestone the company reported to investors.