Kling 3.0

What it is

Kling 3.0 is a series of AI models from Kuaishou for video and image generation. Kuaishou describes the series as an All-in-One framework that covers text, image, audio and video in one system.^[1][2] The series was launched globally on 5 February 2026 as four models: Kling Video 3.0, Kling Video 3.0 Omni, Kling Image 3.0 and Kling Image 3.0 Omni.^[6]

Background

Kuaishou launched the series in February 2026.^[1] Kling AI reported more than 60 million creators, over 600 million generated videos and over 30,000 enterprise users as of December 2025.^[6]

Main features

Kling 3.0 generates up to 15 seconds of video in one pass, with simultaneous audio-visual generation.^[2][3] Compared with Video 2.6, it adds multi-shot generation, which means one request can produce several connected shots. It also improves character consistency and includes native audio with character-level lip-sync, so each character's mouth movements can match that character's speech.^[3] Lip-sync means matching the movement of the lips to the spoken sound.

According to Kling's own description, the model also supports multilingual speech and accents, and it renders on-screen text that can be read.^[3] The series supports native 4K output.^[4] Native output means the model renders at that resolution directly, rather than enlarging a smaller image afterwards.

Position in development

In its first-quarter 2026 results, Kuaishou reported that Kling's annualised revenue run rate (ARR) reached about US$500 million in March 2026.^[2] Kuaishou's second-quarter 2026 results added that Kling revenue exceeded RMB 650 million in the first quarter of 2026, up more than 300 percent, and reached RMB 850 million in the second quarter.^[2][4] A Kling 3.0 Turbo model followed in the second quarter of 2026.^[4] The series was succeeded by Kling 4.0 in October 2026.^[5]

Key terms

A model series is a group of related models released together, usually with different variants for different jobs. The series includes video and image models, and the names of its members, such as Omni and Turbo, indicate variants within the series.

Native output means that the model renders at a given resolution directly, rather than producing a smaller picture and enlarging it afterwards. Enlarged pictures usually look softer than pictures rendered at full size, which is why the difference matters for professional use.

Multi-shot generation matters for storytelling. A scene is usually built from several shots at different framings, and generating those shots in one request makes it easier to keep characters and settings consistent across cuts. Character-level lip-sync means each character's mouth movements are matched to that character's own speech, which is more demanding when several people appear in one scene.