Kling V3 Omni

Kling Video 3.0 Omni is part of the Kling AI 3.0 model family that Kuaishou launched on February 5, 2026, alongside Video 3.0, Image 3.0 and Image 3.0 Omni. It accepts and produces multiple modalities in one workflow, spanning text, images, audio and video, and generates clips up to 15 seconds with native audio in multiple languages, dialects and accents.

Under the hood it is a diffusion transformer aligned with a vision language model, so visual and text tokens share one embedding space instead of being handled by separate pipeline stages. According to the Kling-Omni technical report, the system adds a prompt enhancer module that uses a multimodal LLM to interpret complex instructions, plus a multimodal super resolution stage for fine detail. Training runs in phases: large scale pretraining on text and video pairs, supervised fine tuning with curriculum learning on harder multimodal tasks, then preference optimization (DPO) targeting motion quality and visual integrity.

The Omni variant is built around reference and editing workflows rather than plain text to video. You can upload a reference video and the model extracts a character's visual traits and voice so they stay consistent across new scenes, and a multi shot storyboard mode lets you set duration, shot size, perspective and camera movement per shot. Its editing tools cover adding, removing, replacing and restyling elements in existing footage, and a June 17, 2026 upgrade extended editing to 3 to 15 second clips with 4K input and output.

Kuaishou is open about the edges: the technical report states that integrating video understanding and generation remains a significant challenge, and some capabilities described in the paper, such as reasoning and visual signal features, are not yet supported in the online version. In Kuaishou's own benchmark comparisons against Veo 3.1 and Runway Aleph the model led on image reference and editing tasks, but those numbers come from the company's internal evaluation rather than independent testing.