News

Releases & Models

CONFIRMED: Kuaishou launches Kling 4.0, AI video model with Omni Reference for up to 15 elements

Kuaishou confirms the official launch of Kling 4.0 with Omni Reference, able to combine up to 15 references to keep character and scene consistent across videos up to 30 seconds long.

Ilustração editorial abstrata representando geração de vídeo por IA, com trilhas de movimento neon e quadros de filme se transformando em uma cena urbana futurista.

Kuaishou launches Kling 4.0, pushing AI video toward longer, more consistent scenes

Kuaishou officially confirmed the launch of Kling 4.0, the next generation of its AI video generation model, which went live on October 15. The Chinese company, competing for space with Google, OpenAI and Runway in this market, frames the model as a leap in realism, control and professional-grade quality, exactly the three points where video generators still stumble.

What actually changes

The technical highlight is called Omni Reference: the system now combines up to 15 references in a single generation, adding up to 10 images, 5 videos and 7 different subjects. In practice, that lets creators take a character built in one scene and move it into a completely new environment without losing its look, something testers have already started showing publicly. Kling 4.0 also natively generates up to 30 seconds in a single take, accepts up to 10 keyframes to guide story progression, and expanded the prompt limit to 8,000 tokens, giving more room to describe scene, camera movement and dialogue with precision.

Why it matters for content creators

Until now, keeping a recognizable character across different scenes was one of the biggest bottlenecks in AI video generation: changing environments usually meant reinventing the look from scratch or accepting visible inconsistencies. With references of up to 15 elements and support for video as input, Kling 4.0 tries to solve this more directly, which matters for anyone producing ads, brand content or short fiction without a studio budget. Integration with MCP, the protocol that connects models to agents, and with CLI, the command line interface, also opens the door to automated workflows where an AI agent triggers video generation as part of a larger production pipeline.

The limit that still exists

Access to Kling 4.0 remains restricted: the only API route mapped so far is invite-only, through the fal platform, billed per second of generated video. In practice, most of the audience will test the model first through the Kling AI app and website, not through direct API integration. Kuaishou also did not publish comparative benchmarks against direct competitors, so the conversation about Kling 4.0's real quality against models like Veo or Sora still depends on independent testing, not on official company material.

MaxAssistant's take

The launch confirms that the AI video race has moved into a new phase: it is no longer about generating one pretty isolated clip, it is about maintaining character and scene consistency across a longer narrative, with fine control over each step. That is what separates a toy from a real production tool. Worth watching whether invite-only access opens up quickly and whether independent tests confirm the promise of control without losing visual quality.

Sources

Kuaishou Technology: Kling AI Officially Launches Next-Generation Kling 4.0; TestingCatalog on X: hands-on Omni Reference test.