Kuaishou launches Kling O1 unified multimodal video model

Background

Most video tools on the market have treated creating a clip and changing an existing clip as two separate jobs, often handled by different models with different interfaces. Generation means producing new frames from a prompt or a still image; editing means taking footage that already exists and altering part of it. A "unified" model is one that tries to do both kinds of work inside a single system, so that the same model can take text, an image, a video or a reference subject as input and return video.

What was announced

Kling AI announced Kling O1 on 1 December 2025^[1]. Kuaishou, the company behind Kling AI, called it the world's first unified multimodal video model^[1]. According to the announcement, the model merges generation and editing, and accepts text, image, video and subject inputs^[1]. A second model, Kling Image O1, launched alongside it^[1]. The company presented the release as the opening of Kling's Omni Launch Week^[2].

Key details

The launch release described Kling O1 as built on a Multimodal Visual Language (MVL) framework^[1] — that is, the model is presented as handling visual material of several kinds within one framework rather than through separate specialised components. Kuaishou's year-end release added that Kling O1 brings multimodal video editing and video referencing, together with a subject library intended to improve consistency of characters and objects^[2].

Consistency is a familiar difficulty in AI video: a character or an object can change appearance between shots, which makes longer or multi-shot material hard to assemble. A subject library, as described in the year-end release, is the mechanism Kling O1 offers against that problem^[2].

Why it matters

The significance claimed for Kling O1 rests on the word "unified". If one model can generate new video from a prompt, take an existing video as input for editing or referencing, and hold a character or object steady across results, then users work with one system instead of moving material between separate generation and editing tools^[1][2]. The "world's first" framing is Kuaishou's own description of the release^[1], not an independent ranking. The launch also shows where Kling's product line was heading: the video model was paired with an image model on the same day^[1], and the release was scheduled as the first item in a week of announcements^[2].