Kuaishou launches Kling 2.0

Background

Kling is the AI video generator built by Kuaishou, the Chinese short-video company. It was already a commercial product well before version 2.0: in March 2025 Kuaishou reported that Kling's annualized revenue run rate had passed US$100 million ^[2]. An annualized run rate takes revenue from a given period and scales it to a full year, so it is not the same thing as a full year of booked revenue — but it does show that customers were paying for the service at a substantial pace, which is unusual for a young generative-video product.

The launch

In mid-April 2025 Kuaishou unveiled Kling 2.0 and billed it as the world's most powerful AI video generator ^[1]. The same month, the company rolled the model out to global users ^[2]. According to Kuaishou, the new version led on three fronts: motion quality, semantic responsiveness and visual aesthetics ^[2]. In plain terms, those cover how convincing the movement in a clip looks, how faithfully the result follows what the user actually asked for, and how good the individual frames look. Those three have been recurring weak points in AI-generated video: figures that drift or deform, prompts that are only partly obeyed, and images that look flat or artificial once they move.

Multi-modal control

Kling 2.0 introduced what Kuaishou calls a Multi-modal Visual Language (MVL), together with multimodal video editing ^[3]. Multi-modal simply means handling more than one kind of input at once — text, images, video — instead of relying on a text prompt alone. A "visual language" in this setting is a way of representing what the user wants in a form the model can act on, rather than describing everything in sentences. Multimodal editing applies the same idea to changing existing footage: instead of generating a new clip from scratch, a user can take a video and alter it, with the instruction carried by more than one type of input. That shift matters because much real production work is revision — adjusting an existing shot — rather than creation from nothing.

Reception and the business picture

Kuaishou's CEO said that Kling AI 2.0 was launched in April 2025 and met with a strong reception around the world ^[3]. The leadership claims attached to the release — strongest generator, best motion, best prompt-following, best aesthetics — are the company's own characterisations of its product. Alongside them sits the more checkable number: an annualized revenue run rate above US$100 million as of March 2025, before the 2.0 release ^[2].

Why it matters

AI video generation began as a demonstration technology: short clips that were impressive precisely because they were hard to make. Kling 2.0 marks a stage where the emphasis moves to the things a paying user cares about — control over the output, the ability to edit rather than only generate, and distribution to users in many countries at once ^[2][3]. The revenue figure puts the product in a small group of generative-media services with real commercial traction ^[2].

The release also shows how the competition in this field is framed. A company no longer claims only that its model can produce video; it claims specific advantages in movement, instruction-following and look, and it pairs those claims with a global rollout and a revenue disclosure ^[1][2]. Whether Kling 2.0 was in fact the strongest generator at the time is a question that independent comparisons would have to settle, but the way it was launched — technical vocabulary, worldwide availability and commercial numbers presented together — became the template for how video models are brought to market.