Kuaishou releases Kling 1.5

Background

Text-to-video models take a written description and generate a short clip from it. That is a harder problem than generating a still image, because the model has to keep objects consistent from frame to frame, produce believable motion, and respect what the prompt actually asked for. Kuaishou's Kling family is one of the models built for this task, and in September 2024 the company released a new version, Kling 1.5^[1].

What the update changed

The core of the release was a jump in output resolution: Kling 1.5 generates video at 1080p, up from what the previous version offered, and Kuaishou also reported better dynamics and better semantic responsiveness^[1]. Resolution is the easy part to describe and the easy part to notice — 1080p refers to the vertical resolution of the frame, enough detail to hold up on an ordinary screen. The other two improvements are about behaviour rather than pixel count. "Dynamics" refers to how motion is rendered: whether moving subjects travel smoothly, whether fast action stays coherent, whether the clip looks like footage rather than a sequence of slightly different stills. "Semantic responsiveness" refers to how closely the result follows the prompt — whether the requested action, the requested subject and the requested relationships between them actually appear.

Kuaishou stated that in professional mode Kling 1.5 outputs 1080p HD video^[2]. The distinction matters in practice, since a professional or higher-quality mode is typically the setting in which a model runs its full pipeline, while a faster mode trades detail for speed.

The motion brush

The release also introduced a motion brush feature^[1]. A motion brush is a way of directing a generated clip by hand rather than only through text: the user marks a region of the frame and indicates how that region should move, and the model generates motion that follows the direction while keeping the rest of the scene consistent. Control of this kind addresses a common complaint about text-to-video systems, which is that a prompt is a blunt instrument — describing a camera move or the path of a single object in words is awkward, and the model may satisfy the description while moving the wrong thing. A brush lets the user point at the part of the image that should move.

Performance claims and pricing

Alongside the feature list, Kuaishou said internal testing showed a 95% performance increase over the previous version^[2]. That figure is a vendor-reported result from the company's own evaluation, not an independent benchmark, and the source does not break it down by metric; it is best read as the company's summary of how much better the new model is on the tests it runs itself. On the commercial side, Kuaishou kept pricing unchanged for the new version^[2] — the upgrade was not accompanied by a price increase.

Why it matters

Three things tend to decide whether a video generator is usable: how good a single frame looks, how well motion holds together, and how much control the user has over the result. The Kling 1.5 release touched all three — resolution for fidelity, dynamics and semantic responsiveness for coherence and accuracy, and the motion brush for control — which is why it is a useful marker of where the field was heading. The unchanged pricing is the part that concerns users rather than researchers: a version bump that improves output without raising cost lowers the price of a given quality of generated video. For Kuaishou, which runs Kling as a product rather than only a research demo, that combination — better output, same price, plus a new hands-on control — was the substance of the release.